Pests can cause changes in the physiological and structural characteristics of crops, causing the electromagnetic wave reflectance of crop leaves to change (Cao et al., 2022; Shi et al., 2018). Remote sensing technology has gained widespread recognition for its unique contribution in mapping the location, extent, and severity of crop pest outbreaks and assisting in analyzing potential factors that drive outbreak predictions (Xu et al., 2022). At present, remote sensing technology can acquire various crop information, including crop types, canopy temperature, physical characteristics, leaf nitrogen content, chemical composition, leaf inclination, structural traits, and crop density (Gu et al., 2024). Remote sensing has developed various systems for pest detection and monitoring, which utilize both active and passive radiation across a broad spectrum from gamma rays to microwaves. These include optical remote sensing, radar remote sensing, infrared remote sensing, as well as time-series and multi-temporal remote sensing. These technologies cover multiple bands, from visible light to infrared and microwaves, as well as different data formats and resolutions (Zheng et al., 2023). Luo et al. (2023) reported that after trees are attacked by wood-boring insects, the transport of water or nutrients may be interrupted, which affects the physiological state of the leaves and leads to changes in the spectral and structural information of the leaves and canopy. Figure 2A illustrates the remote sensing signal response to the biological and morphological changes caused by wood-boring pests. Meng et al. (2022) applied remote sensing technology to capture the regional scale of southern pine beetle infestation severity, enabling spatially explicit early warnings and detection of tree mortality caused by the beetle. Their results highlight the unique contribution of remote sensing in mapping the location, range, and intensity of beetle outbreaks, as well as in analyzing potential factors that drive outbreak predictions. Figure 2B illustrates the severity and spatial distribution of southern pine beetle damage after monitoring.

Researchers have explored remote sensing feature extraction, recognition, and classification algorithms at various scales. Notable methods include statistical discriminant analysis, machine learning, regression models, and spectral decomposition algorithms using single or multiple data collections. Xu et al. (2025) introduced a new dual-selective fusion transformer network for HSI classification to address the challenges of effective context scaling and ineffective self-attention features in context fusion when dealing with HSI scenes characterized by various land cover types and abundant spectral information. The study demonstrated that the proposed model significantly enhanced land cover classification accuracy, outperforming existing approaches. Figure 2C illustrates the data processing flow of the dual-selective fusion transformer network. Gu et al. (2024) employed motion structure combined with multi-view stereo algorithms to generate multispectral point clouds from drone-based multispectral images, which can also be used to estimate the 3D spatial distribution of crop photosynthetic parameters. Although the multispectral point clouds generated by structure from motion combined with multi-view stereo are sparse and incomplete, this method holds potential for application in the study of the 3D spatial distribution of crop parameters. Figure 2D shows a schematic of the fusion of lidar point clouds with multispectral images. However, when using remote sensing monitoring methods, certain limitations must be considered. In terms of scalability, remote sensing methods are well-suited for large-scale crop pest diagnosis and inspection, allowing for the identification of the area affected by pest damage. However, in practice, crop pest detection demands high resolution and precision from the equipment. Under complex climatic conditions, ordinary sensors cannot obtain accurate pest monitoring information in typical cloudy and foggy regions, and there are challenges in accurately identifying and distinguishing pest species in cases of overlapping spectra. For the future development of remote sensing technology, integrating multi-source remote sensing data for crop pest monitoring has already become a key direction for future research (Chen et al., 2025).

The crop pest image recognition method mainly utilizes high-definition cameras and visual sensors to create image data collection terminals, which are mounted on fixed or movable platforms for capturing images of crops in targeted areas. After image acquisition, the data are transmitted to local or cloud servers equipped with algorithms for pest feature extraction and matching, enabling comprehensive analyses of pest species, population density, and development trends. Figure 2E shows the structure diagram of an intelligent rice canopy pest and disease monitoring system reported by (Li et al., 2022). The typical process for crop image recognition and detection includes image acquisition, image preprocessing, image segmentation, feature extraction, and classification (Lv et al., 2022). Zhang et al. (2022) attempted to solve the problem of noticeable differences in the same pest species under different shapes, colors, sizes, and complex backgrounds, by proposing a crop pest recognition method based on an improved capsule network (MCapsNet). In MCapsNet, capsule networks are used to enhance traditional convolutional neural networks, with an attention module introduced to capture the most critical classification features and speed up network training. Figure 2F shows the architecture of MCapsNet.

The performance of pest image recognition algorithms is crucial for pest detection applications. Machine learning has been widely applied to detect field crop pest images, and research on the detection algorithm architectures in existing image recognition methods mainly centers on support vector machine (SVM), convolutional neural network (CNN), and Transformer architectures. Ebrahimi et al. (2017) developed an automatic pest detection method for thrips in strawberry greenhouses based on a support vector machine classification approach, mainly utilizing the ratio of large to small diameters as a regional indicator and hue and saturation as color indicators to design the SVM structure. Experimental results indicated that the average error in pest recognition detection was less than 2.25%, and the segmentation results for thrips detection are shown in Figure 2G. SVM performs well with linearly separable data, but it may not be as effective as CNN in handling non-linear data. CNN can effectively capture local features in images and gradually build more complex and abstract feature representations. This ability makes CNN particularly effective in pest detection and recognition tasks. By stacking deep convolutional layers, CNN is the most effective method for automatically learning and extracting useful information from pest images, thereby enabling high-performance pest detection and recognition (Cheng et al., 2017). The most used algorithm at present is the CNN-based improvement of You Only Look Once (YOLO). The YOLO framework treats object detection as a regression problem, solving it with a single forward pass through the convolutional neural network. YOLO performs both object localization and recognition simultaneously across the entire image (Teng et al., 2022). By adopting global feature representations, multi-scale predictions, and dividing the input image into fixed grids, the YOLO series incorporates several enhancements during its iterative process to improve performance (Popescu et al., 2023). Xiong et al. (2024) focused on the green stink bug and small green leafhopper on kiwi and applied an improved YOLOv5s image recognition algorithm at high, medium, and low-density conditions. The results achieved a map of 96.3% and recall of 94.0%. The transformer model can analyses the temporal changes in pest behavior, offering powerful support for dynamic monitoring and early warning of pest activities. Gao et al. (2024) developed an intelligent detection model using transformer deep learning for fast detection of cotton pests and diseases. Experimental results indicated an accuracy of 0.94, a mean average precision f 0.95, and a frame rate of 49.7. Liu et al. (2022) combined deep learning algorithms to design a fixed light-induced insect image collection device to recognize and predict field pest conditions. Based on the ResNet V2 model, they trained T_ResNet V2 using a migration learning dataset. The recognition accuracy on the test set was 84.6%, and after reasonably combining indoor and field datasets, the recognition accuracy of the SM_ResNet V2 model reached 85.7%. The structure of the device is shown in Figure 2H. Despite considerable progress in crop pest image recognition methods for pest detection, which demonstrate their efficiency in practical applications, challenges arise in real agricultural environments. Variable lighting conditions, crop diversity and changes in the size, color and shape of pests at different stages make it difficult to accurately and robustly recognize pests in images, significantly increasing the complexity of identification and detection. Notably, there is limited research on existing image recognition methods for underground and concealed pests. Moreover, when detecting small pests with similar modes in the same spatiotemporal context, the dense distribution and small size of these pests call for further research and development of detection devices and algorithms to improve image recognition accuracy. These are key bottlenecks and challenges that need to be addressed to enhance pest detection precision.

Gas chromatography gas sensors operate based on the differential distribution and adsorption of substances. Inert gases such as hydrogen, nitrogen, and helium are used as the mobile phase to separate and quantitatively measure different gases, essentially combining the high separation efficiency of the chromatographic column with high-sensitivity detection technology. While it provides accurate and objective results, its development and application have been restricted by challenges such as high detection costs, time consumption, complicated analytical procedures, and sample destruction (Weis et al., 2016). Ranger et al. (2010) used gas chromatography and gas sensors to verify the attraction of volatiles produced by nursery plants to ambrosia beetles and to demonstrate the ability of these beetles to attack selected trees when they are attracted by these volatiles. Gas chromatography gas sensors have evolved into several types, including GC-IMS, GC-O, GC × GC-MS, MS/GC × GC-MS/O, and GC-MS, with notable achievements in fields such as tea aroma detection. The miniaturization of instruments, portability, user-friendly human-machine interaction, and the ability to be mounted are crucial products urgently needed in pest detection. In the miniaturization of instruments, gas chromatographic sensors have developed microchip capillary electrophoresis gas chromatography (MCECGC). This technique uses the principle of capillary electrophoresis, where the electric field drives different sized charges to move and separate the gas components into the gas chromatographic part with a coated column, allowing for qualitative and quantitative analysis of the components. Compared to traditional capillary electrophoresis equipment, this is a sophisticated technique for analyzing complex mixtures, capable of achieving rapid separation and high resolution with minimal sample volumes. These miniature detection methods are still in their pregnancy, with only a few applications having been discussed (Nam et al., 2021; Wang et al., 2024; Wu et al., 2024).

Electrochemical gas sensors detect gases through electrochemical reactions occurring between the target gas and the working electrode and reference electrode. Electrochemical sensor detection methods are renowned for their sensitivity and versatility, enabling real-time and on-site analysis with minimal sample volume. A typical sensor consists of a permeable membrane, electrodes, an electrolyte, filters, and a housing. This technique involves detecting the presence of analytes in the sample through changes in voltage, current, or resistance (Bulemo et al., 2025). In gas detection applications, gases enter the sensor and undergo oxidation or reduction reactions with the electrolyte or electrode surface, producing signals such as current, which are proportional to the analyte concentration (Patra et al., 2022). By measuring the signal strength, the gas concentration can be derived, and it has broad applications in fields such as food safety, environmental protection, and aerospace. As the main electrochemical sensing techniques, voltammetry and amperometry have distinct features. Amperometry performs well in continuous monitoring of specific substances, while voltammetry is particularly effective in analyzing detailed sample information (Shimali et al., 2024). Menart et al. (2017) developed an electrochemical gas sensor for formaldehyde gas. The sensor mainly consists of a tri-electrode mesh modified with polyacrylic hydrazine and a poly-electrolyte capable of voltametric measurements. Experimental results showed that the system exhibited linear response over a formaldehyde concentration range of 4–16 ppm under a 120-min accumulation time, with a determination coefficient (R²) of 0.976. Mirelis et al. (2018) employed intermittent pulse amperometry technology to investigate the equilibrium and kinetic binding of target analytes on the surface of electrochemical aptamer sensors. By applying potential pulses using IPA and detecting changes in charge transfer rate within<100μs, they confirmed that changes in sensor current were quantitatively correlated with the target analyte concentration. The intermittent pulse amperometry method offers unprecedented sub-microsecond time response and serves as a general approach for evaluating fast sensor performance, ultimately achieving time resolution up to 2 Ms. Figure 3A shows a schematic representation of the response structure of the aptamer-based electrochemical sensor using rapid, two-millisecond interrogation with intermittent pulse amperometry. Although electrochemical detection technologies face challenges such as limited sensor lifespan, high sensitivity to temperature requiring internal compensation for stability, high maintenance demands, and the volatility and leakage of liquid electrolytes affecting the sensor’s lifespan, the replacement of liquid electrolytes with solid-state and organic gel electrolytes has become the future development direction for electrochemical sensors (Williams, 2020).

The electrochemical sensor-based flexible wearable gas sensor, developed for monitoring plant physiological states, can continuously detect the organic volatile gases emitted by plants in situ, reflecting crucial physiological and pathological parameters. Additionally, the flexible sensor exhibits outstanding mechanical flexibility and strong adhesion to the non-flat surfaces of biological materials, providing a unique potential for real-time, in-situ monitoring of crop growth. Ibrahim et al. (2022) developed an electrochemical, wearable plant sensor for the in-situ monitoring of methanol emissions from plant leaves. This sensor uses platinum nanoparticles modified with the conductive polymer poly(2-amino-1,3,4-thiadiazole) as an active catalyst for the electrochemical oxidation of methanol at specific potentials. The sensor is integrated into a miniature gas collection chamber which is mounted on the leaf surface and covered with a hydrophobic gas diffusion membrane. It offers a detection range of 0.5–500 ppm and a response time of under 20 seconds, Figure 3B shows a schematic representation of the structure of a wearable electrochemical VOCs sensor for monitoring methanol emissions from maize. Li et al. (2021) presented an electrochemical gas sensor array that can be attached to leaves. The sensor patches are applied to a stretchable substrate and incorporate a variety of graphene-based sensing materials and flexible silver nanowire electrodes. This sensor can detect VOCs gas fingerprints generated by stressed crops in real time, accurately detecting and classifying 13 types of plant volatile with over 97% accuracy. Figure 3C shows a flexible sensor mounted on the tomato leaf. Despite challenges remaining in terms of material costs, wireless power, networking transmission and the destructive impact of pests such as stemborers and leaf-eating insects, remarkable progress has already been made in plant growth monitoring using flexible gas sensors. Plant wearable devices thus hold great potential for simple, accurate, and continuous large-scale health monitoring (Yan et al., 2024).

Semiconductor gas sensors detect gases based on the chemical adsorption properties of metal oxide semiconductors. Semiconductor materials are typically divided into n-type semiconductors, including SnO₂, ZnO, TiO₂, WO₃, In₂O₃, Fe₂O₃, and CeO₂, and p-type semiconductors, such as CuO, NiO, Co₃O₄, Cr₂O₃, and Mn₃O₄. When the target gas reacts with the semiconductor material on the sensor surface, conductivity changes are used to detect the gas. The key lies in the interaction between the gas and the material’s surface, involving theories such as the chemical adsorption oxygen model, grain boundary barrier model, bulk resistance model, and space charge layer model (Li et al., 2023). The detection principle is based on n-type semiconductor materials, where electrons act as electron donors. The electrons from the conduction band move towards the electrode, generating holes that are captured by the electron donors, completing the electrode reaction, thus increasing conductivity and forming an anodic photocurrent. In contrast, for p-type semiconductors, holes act as electron acceptors. Electrons on the electrode neutralize the holes in the valence band, lowering the conductivity and forming a cathodic photocurrent. These changes reflect the gas concentration by altering physical quantities like conductivity.

Co₃O₄, In₂O₃/CoOOH composites, as well as perovskite nanocrystals and conductive polymer blends, have gained widespread attention as novel gas-sensitive materials, demonstrating excellent stability and performance in detecting analytes such as H₂, CO, NO, NH₃, and VOCs. Koga (2020) employed pulsed laser ablation to successfully prepare Co₃O₄-NP aggregates with uniformly dispersed Pd additives. Their research revealed that at 5% Pd loading, the material achieved optimal H₂ sensing performance with the highest density of single Pd atoms, which enhanced oxygen adsorption, free electron density, and the formation of high-concentration adsorbed ions. Furthermore, the excellent electron mobility and large surface area of these materials make them ideal for building gas sensors, solving problems such as high working temperatures, low sensitivity, and poor moisture resistance in single metal oxide materials. Liu et al. (2025) enhanced the gas sensing performance by constructing p-n heterostructures, which aid in the redistribution of electrons, improving the material’s response speed and sensitivity. Initially, sodium hydroxide (NaOH) solution was used to etch the synthesized metal-organic framework (MOF)-derived In₂O₃ nanotubes and Co-MOF precursors, leading to the preparation of MOF-derived In₂O₃/CoOOH composites. The structure and morphology of the composite Finally, in situ infrared spectroscopy was employed to study the NOx interaction in In₂O₃/CoOOH composites. At room temperature, the response value (Rg/Ra) for 10 ppm NOx was 84%, with a detection limit of 0.1 ppm, approximately 2.5 times that of pure In₂O₃. The response and recovery times were 141/78s, which is faster than the 179/86s for pure In₂O₃. material was then analyzed using techniques such as XRD, FT-IR, SEM, TEM, and XPS. Figure 3D shows the simulated sensing mechanism and energy bands of the In₂O₃/CoOOH composite material before and after exposure to nitrogen oxides. Jang et al. (2023) developed a novel organic-inorganic hybrid gas sensor by blending perovskite nanocrystals with conductive polymers to improve gas absorption capacity and overcome the challenges of low sensitivity and poor stability in organic semiconductors. CsPbBr₃ perovskite was embedded in the conductive polymer matrix to enhance gas sensing performance while maintaining sensing speed, and amphoteric ion polymer ligands were used to modify the surface of the perovskite nanocrystals. The lowest detection limits were compared in the PCA-w-NCs system, with NO₂ detection limit of 0.000318 ppb, SO₂ at 0.00112 ppb, and CO₂ at 0.00479 ppb. Figure 3E shows a schematic representation of the preparation method for the P₃HT/perovskite blended film. Yuan et al. (2022) developed Pd-modified Co₃O₄ hollow polyhedral (Pd/Co₃O₄-HP) by pyrolyzing a Pd-doped MOF precursor and designed an ethanol gas sensor. The experiments demonstrated that at a working temperature of 150°C, the Pd/Co₃O₄-HP gas sensor achieved 1.6 times higher sensitivity compared to Co₃O₄-HP, with rapid response (12 seconds) and recovery (25 seconds) to 100 ppm ethanol vapor. This study provides an effective approach for investigating the spillover effect in the sensing mechanisms of precious metal-modified oxide semiconductor sensors. The preparation and characterization of Pd/Co₃O₄-HP are depicted in Figure 3F.

The electronic nose, based on semiconductor sensor technology, is an instrument that consists of an array of electronic gas sensors with partial specificity and pattern recognition algorithms. This gas sensor array technology is also termed an electronic nose. Inspired by the human olfactory system, during detection, the target gas is collected into a chamber with an array of electronic chemical gas sensors with partial specificity via a sampling pump. Different gas-sensitive sensors react specifically to various organic compounds, causing changes in the sensor’s resistance. The sensor reaches a stable state within a few seconds. The data acquisition system records real-time changes in the response signals from each sensor. After processing with suitable pattern recognition algorithms, the type and concentration of the detected gas are determined, enabling the identification of both simple and complex gases. Xu et al. (2025) leveraging the enhanced adsorption capability of semiconductors and dual mesoporous characteristics, developed an innovative method for fabricating zinc tin oxide semiconductors with dual mesoporous structures and tunable oxygen vacancies using direct solution precursor plasma spraying technology. This method was employed for detecting 2-undecanone, a key biomarker for rice aging. Experimental results showed that the developed sensor effectively detected adulteration in different rice varieties. The sensor’s sensitivity to VOCs from japonica rice was as high as 62.9%, about 4.5 times that of indica rice. The results of testing the gas-sensing properties of all materials on the platform of a side-heating gas sensor are shown in Figure 3G. Zhou (2011) focused on common pests in rice cultivation, such as the striped stem borer and brown planthopper. They used the developed electronic nose to detect organic volatile gases produced by rice plants under pest stress, compared with healthy rice plants. The experiment ultimately verified the feasibility of using the electronic nose for pest detection in rice plants. Despite the potential of the electronic nose in various fields, its application for pest detection in complex field environments still faces challenges. First, environmental factors such as temperature and humidity may affect sensor sensitivity. Interference from multiple coexisting gases leads to inaccurate identification, necessitating the development of new materials and technologies adapted to varying temperature environments. Secondly, sensors have poor long-term stability, requiring regular calibration or replacement. The cost and limited lifespan of high-performance sensors also add to the maintenance burden, while complex data processing demands significant computational resources. This limitation restricts the widespread use of electronic noses in high-precision applications, necessitating urgent solutions.

The light source, as the core component of trace gas sensors in infrared absorption spectroscopy, should be selected based on the operational scenario, determining the spectral range of the target gas. The light source must have sufficiently high absorption intensity for the gas, be tunable within the wavelength range, avoid interference from other gas absorption lines, and match the selected spectral technique. Sources emitting infrared, visible, and ultraviolet spectral wavelengths can be selected. Furthermore, it’s important to note that different light sources have different radiation bandwidths. Lasers generate narrowband radiation, whereas thermally infrared coherent light sources generate broadband radiation. These distinct characteristics dictate the choice of spectroscopic technique: tunable laser absorption spectroscopy (TLAS) for narrowband laser sources, and non-dispersive infrared (NDIR) spectroscopy for broadband thermal sources. The choice of light source determines the type of optical multi-pass cell and detector used in the system, which indirectly affects the gas sensing system’s overall detection accuracy. The use of a single light source to detect multiple gases simultaneously in infrared absorption spectroscopy trace gas sensors has attracted significant interest from many researchers. In single light source systems, mid-infrared frequency comb spectroscopy technology has attracted attention for its outstanding ability to detect multiple gases. The frequency comb laser plays a critical role in this sensing technique by combining high spectral resolution, broad bandwidth coverage, and high-precision cavity interactions through individual comb teeth. Unlike conventional lasers, they can emit light pulses with thousands to millions of different “frequencies” that are coupled into the cavity. These pulses interact with molecules in the cavity as they make thousands of round trips in the gas pool, allowing for gas species identification (Liang et al., 2025). Liang et al. (2021) addressed the issue of resonance mode mismatches between mid-infrared light sources and optical cavities, which previously led to laser beam emission. They used the modulated ring-decreasing frequency comb interference method and employed mid-infrared cavity-enhanced frequency comb spectroscopy technology for human breath gas detection. Their research enabled the simultaneous detection and monitoring of four breath biomarkers—CH₃OH, CH₄, H₂O, and HDO—with detection accuracy reaching ultra-high sensitivity at the trillionth level. The schematic diagram of the mid-infrared frequency comb detection process is shown in Figure 4A. Ben et al. (2016) applied infrared frequency comb spectroscopy technology to detect complex molecules of various gases. The study demonstrated a significant increase in spectral resolution and specificity in identifying multiple gas molecules.

The optical multiplexing structure of the light source is important for the simultaneous detection of multiple gases. It passes infrared light from different lasers through an optical multi-pass cell in sequence, enabling multiple gases to be detected simultaneously. Our team has developed a wavelength modulation spectroscopy technique that combines second harmonic detection and a laser power fluctuation correction method. Based on a fiber optic switch integrated with time-division multiplexing technology, this near-infrared distributed feedback long-path absorption spectroscopy laser sensor enables different lasers to sequentially scan and measure methane (CH₄), carbon dioxide (CO₂), and oxygen (O₂) in the atmosphere. Experimental results indicate that, depending on concentration and noise levels, the detection limit (1σ) for CH₄ can reach 0.034 ppm, for CO₂ it can reach 11.921 ppm, and for O₂ it can reach 0.14% (Dang et al., 2024). The multi-channel multiplexed multi-gas sensor based on the fiber optic switch is shown in Figure 4B. In another study, Dong et al. (2017) employed a single broadband light source and a fast-switching detector device to detect multiple gases. The method primarily used a stepper motor to rotate and switch between thermoelectric detectors for carbon monoxide (CO), carbon dioxide (CO₂), and methane (CH₄). The study demonstrated that the stability of the multi-gas sensor was optimal when the switching motor operated at a speed of 0.4π rad/s, achieving a detection limit of 8.83, 8.69, and 10.29 ppm for the three gases in one second of dynamic operation. The multi-detector rotating switch device is shown in Figure 4C. Furthermore, regarding the optical aperture for gas cell light passage, Dong et al. (2019) proposed a concentric reflection model design method to address the spatial waste issue in classic Herriott cells and improve mirror utilization efficiency. They developed a dual-ring structure optical multi-pass cell that implements multi-gas detection, enabling a single Herriott optical multi-pass cell to support multiple light paths and facilitate the simultaneous detection of either single or multiple gases. The optical multi-pass cell structure with two incident apertures is shown in Figure 4D.

An optical multi-pass cell is designed to increase the effective optical path length for gas absorption. This is achieved by coating the mirror surface with various high-reflection materials, satisfying infrared absorption spectroscopy detection requirements across a range of wavelengths. Typical optical multi-pass absorption cell structures include White, Herriott, ring-type, and discrete lens long optical path cells. According to Lambert-Beer’s law, the strength of absorption is directly proportional to the effective interaction length between the laser and the gas. For gas molecules to absorb enough light, the light must travel several miles or even further through the gas sample in an optical cavity that is measured in feet. The length of the optical path dictates the detection accuracy of the gas sensor, so enhancing the effective absorption path length is a critical method for improving the detection sensitivity of the system. Precise cavity alignment is challenging, as mismatches between the laser and cavity resonant modes cause the beam to be rejected. Enhancing the efficiency of mirror utilization and reducing the volume of optical multi-pass cells are currently key areas of focus in research. Many researchers have developed various high-precision finite volume long optical path multi-pass cells tailored to different application scenarios, considering factors like light source type, light coupling methods, optical path structure, detection techniques, and target gases. New optical-mechanical structures are emerging, and optical multi-pass cells still have vast potential for improvement.

The design theory for classic optical multi-pass cells dates to 1942, when John U. White at Esso Research Laboratory in the United States designed the White optical multi-pass cell using three spherical mirrors with the same curvature radius (White, 1942). In 1962, Herriott at Bell Labs developed the Herriott optical multi-pass cell with a ring-shaped spot reflection structure using two identical concave mirrors. In recent years, researchers have optimized the design of optical multi-pass cells by improving the type and number of mirrors, effectively increasing the utilization of mirrors. This has promoted the development of long optical path multi-pass cells in finite volumes, further advancing the design theory for finite volume long optical path multi-pass cells (Herriott et al., 1964). The reflection path of a laser between two spherical mirrors is typically described using the ABCD matrix and vector reflection principles. However, in optical multi-pass cells with dense spot patterns, the laser reflected between the two spherical mirrors is often off-axis, violating the paraxial conditions, making the ABCD matrix no longer applicable. Although the modified ABCD matrix can eliminate errors arising from the paraxial approximation, the ray tracing process itself involves repeated complex matrix calculations and iterations, making the design process very complicated. The literature has reported the application of relevant algorithms to improve the traditional optical multi-pass cell design theory. Liu et al. (2021) first used the Monte Carlo algorithm to determine the distance between two mirrors. Then, they used the Nelder-Mead simplex algorithm to locally optimize the incident parameters, performed ray tracing, and simulated the size and shape of the spots. Finally, they applied a clustering method to automatically select independent circular patterns, resulting in optimal circular patterns with five, seven, and nine circles. The numerical simulation results are shown in Figure 5A. In another study, Chen et al. (2022) proposed an intelligent method involving a parallel multi-population genetic algorithm to optimize the optical path length (OPL) and size of the traditional Herriott optical multi-pass cell. Following several iterations, they designed an optimal multi-pass cell with an OPL of 6.3 m, with mirror diameters and distances of 25.4 mm and 30.5 mm, respectively (Ozharar and Sennaroglu, 2017). introduced a new method using rotational symmetry, spherical aberration and offset-related focal lengths to create a unique optical design. This innovative design uses astigmatic mirrors to form spot patterns beyond the standard Lissajous figure.

The practical development process of optical multi-pass cells typically includes theoretical design, parameter optimization, simulation verification, and practical validation, as shown in the design flowchart in Figure 6. In the practical design theory, optical multi-pass cells involve multiple variables, and the results vary under each parameter. The design typically follows the principle of optimal optical path design at the limit volume. The process begins with the determination of the volume and related dimensions based on engineering requirements. The ABCD matrix and its improved ray propagation theory are then used to calculate parameters such as the lens diameter and the spatial angular range of incident and outgoing light. The mirror distance is also dynamically adjusted. Simulations are then performed to calculate the optimal base length, with multiple iterations resulting in the best theoretical parameters and achieving the maximum optical path at the limit volume. Once the optimal theoretical parameters have been determined, multiple simulation verification iterations are carried out using well-established optical design software such as Zemax and Tracepro (Prabhat et al., 2023). If the simulation verification is successful, the optical-mechanical components are prepared and a test platform is built for optical experiments, with the aim of achieving the actual optical path. During gas detection experiments, researchers observed that the gas exchange rate is also a key indicator of the performance of optical multi-pass cells. This rate is primarily influenced by factors such as the structure of the gas chamber, the diffusion rate of the gas, and the speed of system signal processing. A higher gas exchange rate allows more substances to be fully interacted with by the light, resulting in more comprehensive detection of the sample gases. To avoid gas trapping and repeated laser detection caused by turbulence and eddy currents when gases pass through the optical multi-pass cell, researchers typically use fluid simulation software such as ANSYS and COMSOL. These programmers are used to simulate and validate the velocity of gas flow and the positions of pores (Liu, 2019; Yao et al., 2007).

In the field of optical path structure innovation, Wang et al. (2023) innovated the Herriott optical multi-pass cell with two 3mm-diameter holes at 90° in the second mirror. The outgoing light is reflected by M1 and M2 and returns to the cell to form a double-ring pattern with outer and inner circles. This is reflected 111 times, achieving an optical path of 73.926m. For multiple mirror structures and the White improved optical multi-pass cell. Cheng et al. (2022) proposed an improved ABCD matrix model. Using three non-coaxial mirrors, they developed a V-shaped multi-pass optical cell. This structure is very similar to a folded cavity in a laser cavity, meaning that the beam propagation matrix can be used for simulation and design. Ultimately, this achieves three effective OPLs of 49.6, 97.6 and 173.6 m and the spot patterns presented on the spherical mirrors M1, M2, and M3 of the three different multi-pass cells (MPCs) are shown in Figure 5B. Hudzikowski et al. (2021) used a genetic algorithm and four spherical mirrors to develop optical multi-pass cells with 16m and 23.8m optical path lengths. The MPC with an OPL of 23.8m was sealed and filled with 200ppm CO₂ in nitrogen at 533 hPa. CO₂ was detected at 2004nm using WMS, indicating a detection limit of 0.4ppmv in 10s. The spot pattern model for the best circular pattern presented by the four mirrors is shown in Figure 5C. The research and development of a new ring-type multi-pass cell was reported. Jouy et al. (2014) reported the development of a new ring-type multi-pass cell, using an 8cm diameter metal cylinder. To avoid optical fringes caused by stray light and main beam interference, they developed an absorption mask. The cylindrical chamber, with a volume of just 40 cm3, has an inner surface engraved with a ring mirror. Light enters the gas chamber through a small-angle incident hole directed toward the optical axis. After passing through multiple reflective holes, the light reflects up to 51 times, achieving a path length of 4m. The performance was evaluated by measuring the isotope ratio of CO₂ under environmental abundance, with experimental results indicating a measurement accuracy of 2ppm over an integration time of 600s. The micro-sensor consists of a QCL mounted in an HHL, an annular multi-pass cell, and a QCD system with a preamplifier, as shown in Figure 5D.

The development of ultra-compact, small-volume optical multi-pass cells has also been reported. Cui et al. (2020) reported the development of an ultra-compact optical multi-pass cell device based on tunable diode laser absorption spectroscopy technology, using a 1.65µm near-infrared distributed feedback laser and a quartz tuning fork (QTF) as a thermal detector. The palm-sized methane (CH₄) optical multi-pass cell device, with dimensions of just 78mm × 40mm × 40nm, achieved a minimum detection limit of 52 ppb within a 300ms integration time. The system structure is shown in Figure 5E. Feng et al. (2021) reported a compact ring-shaped optical multi-pass cell with an effective optical path length of 8.35m and 84 reflections, displaying a two-layer polygonal star-point pattern. Using a 1.653µm DFB laser, it detected methane in the atmosphere, achieving a detection limit of 22 ppb with a detection precision within a 57.8s integration time. The system is based on tunable diode laser absorption spectroscopy and a new dual-layer annular multi-pass cell, as shown in Figure 5F. Tuzson et al. (2020) developed a drone-mounted methane sensor. It uses a ring-shaped optical cell and a low-noise, intermittent continuous wave laser. Under stable laboratory conditions, the sensor achieved a detection limit of 0.1 ppb in 100 seconds. The CH₄ sensor installed on the six-axis aircraft is shown in Figure 5G. Based on the above research, it is evident that optical multi-pass cell designs have more possibilities and flexibility. By increasing the number of mirrors or changing the types of mirrors, more optical multi-pass cell structures with different reflection types can indeed be obtained. However, the difficulty of obtaining design parameters increases significantly due to the influence of multiple parameters on the beam’s reflection behavior. Additionally, in ray tracing, the vector reflection principle offers greater advantages, as it eliminates the need to adjust the beam propagation matrix based on the placement of mirrors. The introduction of intelligent algorithms further enhances the accuracy and convenience of optimizing optical multi-pass cell parameter design.

In trace gas infrared sensing systems, the function of the detector is to capture the weak light intensity signal after infrared light has passed through the optical multi-pass cell and fully interacted with the target gas. This light signal is then converted into an electrical signal through the photoelectric effect. Due to the weak nature of the original electrical signal, it usually needs to be filtered and amplified by a signal amplifier before being input into an analog-to-digital(A/D) converter for signal acquisition and digitization. In the whole sensing system, the light intensity change signal captured by the detector is usually related to the concentration of the target gas. As such, the detector is a critical component for high-sensitivity gas detection, and its detection accuracy and response sensitivity directly influence the system’s performance. The most used infrared detectors for gas detection today include thermopile detectors, pyroelectric detectors, and various improved detectors based on new materials and structures. Pyroelectric detectors respond to changes in the crystal’s temperature when infrared radiation hits it. This generates a measurable charge signal which is then converted into a current signal. Thermopile detectors consist of stacks of thermocouple units. Materials with different thermoelectric properties and polarities form the cold and hot ends. A temperature difference generates charge carrier movement, which in turn produces a voltage output. Advances in microelectromechanical systems technology are enabling the development of smaller, more integrated detectors.

In the field of infrared absorption spectroscopy trace gas sensors, some researchers have applied photoacoustic/photothermal technologies to enhance detectors and develop high-precision detectors. Figure 7A shows the structure of the opto-thermal-elastic coupling QTF sensor proposed by (Dai et al., 2023). This sensor utilizes a CH₃NH₃PbI₃ perovskite. Perovskite thin film coated on the surface of a QTF and forms a Schottky junction between the QTF surface and a silver electrode. This structure exploits the optoelectronic and thermoelastic effects coupling between CH₃NH₃PbI₃ and QTF, thereby enhancing the sensor’s detection sensitivity. Ma et al. (2023) developed a highly sensitive H₂ detection system by combining infrared absorption spectroscopy with photoacoustic/photothermal technologies to detect H₂ with weak absorption characteristics, achieving a detection limit of 45 ppm. Figure 7B illustrates the schematic of the H₂ sensor platform based on photoinduced thermomechanical spectroscopy. Hu et al. (2019) addressed the issue of high acoustic coupling strength and detection sensitivity in quartz-enhanced photoacoustic spectroscopy by optimizing the structural parameters of the acoustic micro-resonator in the embedded off-beam quartz-enhanced photoacoustic spectroscopy spectrometer, based on principles of solid-state heat transfer and mechanics. This optimization led to an improved signal-to-noise ratio gain. Compared to the configuration using a bare QTF, dual-channel detection was achieved by embedding two resonant tubes into the QTF, resulting in approximately 40 times enhancement in the photoacoustic signal. The schematic of the embedded off-beam quartz-enhanced photoacoustic spectroscopy spectrometer with one or two resonant tubes is shown in Figure 7C. Qiao et al. (2023) developed a sensitive light-induced thermomechanical spectroscopy trace gas sensor using an ultra-miniature QTF with a prong length of 3500 µm, prong width of 90 µm, and a resonant frequency of 6.5 kHz, designed for detecting acetylene (C₂H₂) gas. Compared to commercial QTFs, the signal improved by 1.64 times, achieving a minimum detection limit of 190 ppb with an integration time of 220s. The schematic diagram of the ultra-miniature QTF is shown in Figure 7D. Wu et al. (2022) developed an 8mm long clamped quartz tuning fork-enhanced photoacoustic spectroscopy system, which enhances the acoustic wave coupling efficiency by introducing a wavefront-shaped aperture with a diameter up to 1mm, while keeping the Q factor > 104, increasing the laser beam focusing area by more than ten times compared to the standard QTF, and achieving approximately 30 times signal-to-noise enhancement compared to the dual-tube micro-resonator in traditional Quartz enhanced photoacoustic spectroscopy systems. Wang et al. (2025) designed a quadrupole quartz tuning fork with more prongs to increase stress concentration for better acoustic detection. This also generated stronger signals, with a minimum detection limit of 96 ppb. Simon et al. (2025) coherent control technology was introduced in quartz-enhanced photoacoustic spectroscopy. The main approach was to adjust the laser pulse timing to match half the oscillation period of the tuning fork while keeping the laser output frequency constant. This adjustment ensures that the laser pulse reaches the gas molecules between the tuning fork arms, causing the gas to heat and expand, producing a reverse force on the tuning fork’s motion. A proof-of-concept platform was successfully established for analyzing methane gas using a quartz-enhanced photoacoustic spectroscopy chamber, showing that the coherent control method keeps spectral fingerprints clear and stable.

Several trace gas sensors using infrared absorption spectroscopy have been reported in the literature. These sensors can monitor agricultural production and detect specific gases in marine environments. For instance, the presence of endogenous methane gas is a key indicator of whether wheat experiences high-temperature and UV stress during the tillering stage. Zhang et al. (2025) developed an early recognition and prediction model by combining hyperspectral technology with infrared absorption spectroscopy to predict endogenous methane gas produced by wheat under heat stress (HS) and ultraviolet-B stress, achieving an accuracy of 98.88%. The structure of the detection system is shown in Figure 8A. Li et al. (2023) developed a mid-infrared CO sensor for fire warnings in cotton harvesting. The Lod was 0.83 ppm, and the system worked well at -40 to 85 °C. The mid-infrared CO and CO₂ dual-gas sensor system for cotton harvesters is shown in Figure 8B. Dai et al. (2023) used infrared absorption spectroscopy and LITES technology to monitor changes in O₂ concentration and assess strawberry spoilage during storage. The strawberry gas detection system is shown in Figure 8C. Liu et al. (2020) developed a system for measuring dissolved CO₂ in seawater using a rare CO₂ sensor based on tunable laser absorption spectroscopy. This system used a 4319 nm laser and an optical multi-pass cell and was integrated with a gas-liquid separator. It was deployed in the South China Sea at 30 Torr, validating the sensor’s ability to analyze dissolved gases in situ. The deep-sea dissolved CO₂ detection system is shown in Figure 8D.

Chemometrics has shown unique application value in several fields, including tea flavor identification, food adulteration detection, and origin tracing (Fang et al., 2019; Wang et al., 2022). Pest species identification relies on gas molecular fingerprints as key discriminative information. Chemometrics, as an important tool linking infrared absorption spectroscopy signals with target attributes, plays a critical role in data-driven modelling. Following the sensitive detection of pest-induced VOCs by infrared spectroscopy, chemometrics is employed to extract spectral features, build discrimination models, and ultimately match the spectral patterns to specific pest species through statistical analysis. The basic process consists of data pre-processing, feature extraction, modelling analysis, and model validation. Common data preprocessing methods include standard normal variation transformation, multiplicative scatter correction, and first- and second-order derivative transformations, which help reduce spectral noise, correct baseline drift, and improve data quality and stability. During feature extraction, principal component analysis and linear discriminant analysis are extensively used for spectral dimensionality reduction and sample classification tasks (Monika et al., 2024). Partial least squares regression is often used to establish quantitative relationships between spectral variables and target attributes, especially for handling high-dimensional and multicollinear data. To meet the demand for nonlinear feature modelling, methods like artificial neural networks and support vector machines have been introduced in spectral analysis in recent years, significantly enhancing the model’s ability to express complex systems and improve generalization performance.

Wang et al. (2020) studied how to detect wood-boring pests hidden inside tree trunks. They analyzed pest sample gases using GC-MS and identified four key gases as molecular fingerprints for the MIP molecular template. They then developed a statistical model using data from the QCM sensor array to differentiate the gases emitted by cedar trunks infected by the pests, and uninfected trunks. PCA and LDA were used to reduce dimensions and visualize the data. K-nearest neighbors, probabilistic neural networks, and support vector machines were used to establish pattern recognition models. The results confirmed that chemometric pattern recognition methods, combined with the QCM sensor array, could effectively identify the infestation of wood-boring insects, such as the striped pine woodborer and cedar bark beetle, in cedar trunks. Cardoso et al. (2025) used chemometric methods with NIR, MIR spectroscopy and RP-HPLC to analyze 56 traditional Mina’s cheese samples from different regions in Minas Gerais, Brazil. They obtained fingerprint spectra for cheeses from different regions and successfully traced their origin. PCA was used to assess similarities and differences in WSP extracted from cheeses produced in different regions. PLS-DA was used to establish a model that classifies different regions using NIR and MIR spectral data. Thus, chemometric methods provide efficient and sensitive tools for identifying plant pest damage, supporting precision agriculture.

Multivariate statistical analysis plays a crucial role in pest scale determination methods. Plants under pest stress release specific VOCs, and temporal variations in the concentrations of these gases can be significant indicators when evaluating pest severity and trends in spatial spread (Ricardo et al., 2019). VOCs released by plants are typically collected over continuous time periods at fixed intervals. The concentration of target gases (such as alcohols, esters, terpenes, etc.) is dynamically monitored to analyze their variations over time. Based on the response amplitude of VOCs concentrations, release rates, and peak times, multivariate statistical methods such as cluster analysis, discriminant analysis, and principal component regression can be employed to create pest severity classification models (e.g., mild, moderate, severe). These maps can predict the spatial scale of infestations, enabling regional monitoring and providing a quantitative basis for precision pest control decisions.

Henderson et al. (2010) used a portable electronic nose to detect stink bugs in cotton. Their study found a strong correlation between pest density and sensor response. Zhou et al. (2022) used infrared spectroscopy to study grape storage spoilage detection by monitoring ethanol gas concentrations. Their research found that as storage time increased, the ethanol absorption peak intensity grew. PCA analysis revealed distinct clustering at different storage times, indicating significant changes in ethanol concentration over time. The qualitative model developed with PLS-DA accurately classified grape samples into “fresh,” “slightly spoiled,” or “severely spoiled,” achieving 100.00% accuracy. Schaub et al. (2010) used proton transfer reaction mass spectrometry and gas chromatography-mass spectrometry to monitor the release characteristics of VOCs from poplar trees after feeding by fall cankerworms in a field environment. Their results showed that VOCs concentrations and compositions were closely related to pest density. Cai et al. (2015) used solid-phase microextraction combined with GC-MS technology to analyze the dynamic changes in VOCs during aphid infestation in soybeans. They found that certain key compounds (such as green leaf volatiles and terpenes) increased significantly as the pest damage intensified, indicating that VOCs concentrations can be used as an indicator of pest intensity. Milton et al. (2022) further systematically compared VOCs emission profiles from different crop varieties under similar pest infestations. They suggested that changes in VOCs components could be used both for screening variety resistance and for developing pest severity recognition models. Collectively, these studies demonstrate that continuous monitoring of VOC dynamics, combined with statistical modelling, enables not only qualitative pest identification but also quantitative classification of pest intensity and spatial prediction. This provides a theoretical foundation and methodological framework for regional pest monitoring and precision control.