Ultrasound waves travel through mediums, reflecting and transmitting at the boundaries where the material properties of the medium change. This behavior is determined by the acoustic impedance, which is defined by Equation 2:

The density ρ and speed of sound c determine the proportion of an ultrasound wave that is reflected vs. transmitted at a boundary (Pusppanathan, 2017). A significant impedance mismatch causes strong reflections. For example, when an ultrasound wave traveling in a tissue encounters air, the impedance difference causes nearly total reflection, resulting in minimal transmitted energy (Grager et al., 2018). Medical ultrasound imaging maps the reflection from the tissue interfaces of differing impedance. For normal incidence, the reflection coefficient R, is defined as the ratio of reflected incident pressure amplitude defined as Equation 3:

As (3) shows, a greater contrast in acoustic impedance Z results in a stronger echo (Yared, 2011). This is essential in both medical and industrial applications. In medical applications, echoes form an image and in industrial applications, echoes help detect flaws (Martynenko and Ermachenko, 2021; Zhong et al., 2022). The intensity of ultrasound waves is attenuated as they travel through a medium. This attenuation or loss occurs due to absorption, scattering, and reflection, and is defined exponentially as Equation 4 (Zheng et al., 2024):

It means that the intensity decreases with distance. In soft tissues, attenuation typically increases linearly with the frequency (often 0.5 dBcm⁻¹ MHz⁻¹) as a rule of thumb. Thus, higher-frequency waves penetrate a shorter distance while offering higher resolution. By contrast, a lossless medium like water has negligible attenuation, and many solids (metals) have low intrinsic absorption. However, scattering from the microstructure can attenuate the wave. Gases cause very high attenuation of ultrasound, especially at high frequencies (Gudra, 2008; Su et al., 2017; Wells and Liang, 2011).

The transducer is the most critical component in an ultrasound system. It converts electrical energy into acoustic waves and vice versa. Most ultrasound transducers are piezoelectric. A driving voltage pulse causes the piezoelectric crystal to vibrate mechanically, launching an acoustic wave (He et al., 2023). Upon receiving an echo, the pressure wave deforms the crystal, generating an electrical signal. Transducers are engineered with several key components: a piezoelectric element, a backing material, and one or more matching layers on the emitting face. The backing dampens the vibration duration (producing a broadband pulse) and reduces ringing. The acoustic matching layer is a critical design feature to efficiently transfer energy into the load medium, typically a quarter-wavelength thick layer with acoustic impedance intermediate between the high-impedance crystal and the lower-impedance medium (Toffessi Siewe et al., 2023; Barakat, 2023; Lu et al., 2023). A schematic illustration of an ultrasound transducer is shown in Figure 2 .

Transducer frequency is selected based on the application. Medical probes for abdominal imaging operate around 1–5 MHz, whereas intravascular ultrasound or ophthalmic probes may use 20–60 MHz for fine resolution (Li et al., 2010; Luca et al., 2021). In industrial testing, lower frequencies (0.1–5 MHz) are standard for thicker or more attenuating materials, whereas higher frequencies (10–20 MHz) are used for fine-grained materials or thin parts (Krautkrämer and Krautkrämer, 1990). Because ultrasound does not travel efficiently through air, a coupling medium is typically required for contact transducers. In medical imaging, a gel is applied between the probe and the skin to displace air. In industrial inspections, liquid couplants (e.g., glycerin, oils) or water-immersion setups are used (Cheng et al., 2022; Koulountzios et al., 2019). On the other hand, non-contact methods (discussed below) avoid liquid couplant by generating or detecting ultrasound through air or electromagnetic waves.

Contact ultrasound is the default approach in medical diagnostics and many NDT applications since it allows efficient acoustic coupling. A couplant (gel or liquid) is applied to minimize the air gap and reduce reflection losses at the interface. This yields a strong signal and high signal-to-noise ratio. Non-contact ultrasound is required where the direct coupling is infeasible or undesirable (e.g., hot surfaces, moving parts, large-scale scanning). The most common approach is air-coupled ultrasound. However, due to the severe impedance mismatch between typical transducer materials and air, only a tiny fraction of energy couples into air. Specialized air-coupled transducers are designed with lower frequencies to mitigate attenuation, high driving voltages and sensitive detection (Liu and Abdulla, 2023; Bente et al., 2023).

Other, non-piezoelectric, transducers such as capacitive micromachined ultrasonic transducers and optical/laser-based approaches have also advanced, permitting non-contact measurement with broader bandwidth or higher sensitivity (Yuan et al., 2023; Song et al., 2015). Electromagnetic acoustic transducers provide a non-contact option on conductive materials. They induce ultrasonic waves by electromagnetic forces in the test object (Jiang et al., 2023). However, EMATs efficiency can be lower than piezoelectric transducers, requiring powerful pulsed excitation. Another non-contact technique is laser ultrasound, where a pulsed laser generates ultrasonic waves via thermal expansion or ablation, and a separate interferometric laser detects surface displacements (Narumanchi et al., 2023). This is couplant-free and can operate at a standoff distance, but requires expensive, sensitive optical equipment. A comparison between contact and non-contact ultrasound is shown in Table 1 .

Depending on the application, various ultrasound modes, such as pulse-echo, through-transmission, and tomography, are utilized to gain in-depth information on the material under observation (Zhang and Cegla, 2022; Singh et al., 2022). Each technique differs in its operational concept, transducer arrangement, and data collection mechanism, providing distinct application advantages. In pulse-echo mode, a transducer sends short ultrasonic pulses ( Figure 3 ) into the material and receives the echoes that return. The system measures time-of-flight and echo amplitude to measure internal features (Yanez et al., 2022). This is fundamental to both medical B-mode imaging and nondestructive flaw detection. Modern array-based systems can perform beamforming by introducing electronic delays, improving lateral resolution, and enabling scanning without physically moving the probe (Foiret et al., 2022).

The through transmission mode places a transmitter on one side of the object and a receiver on the opposite. Instead of echoes, it measures how much ultrasound wave passes through, indicating attenuation or disruptions (like flaws) in the material, as shown in Figure 4 . This method requires access to both sides but can yield direct measurements of transmitted intensity (Lluveras Núñez et al., 2017). It is often used to detect large internal voids or regions of high attenuation. Depth information is not directly obtained unless combined with scanning or tomography.

Tomographic methods reconstruct 2D or 3D maps of acoustic properties by combining multiple measurements from different angles. In transmission tomography, one measures the time-of-flight and attenuation along numerous paths around the target, then uses inverse algorithms to compute spatial distributions of sound speed and attenuation (Wiskin et al., 2019). Reflective tomography collects echo data from multiple vantage points, similar to seismic imaging, to build a reflectivity map (Zhang et al., 2018). Hybrid methods capture both transmitted and reflected signals. A classic medical example is breast ultrasound computed tomography, where an array encircles the breast in a water tank, sequentially emitting pulses and recording transmitted/reflected waves in all directions. Industrially, ultrasonic tomography can be used for pipeline or structural inspections by placing multiple transducers around a test object and reconstructing internal features from the measured signals (Lyu et al., 2024).

Ultrasound waves are utilized in soil monitoring applications to assess soil properties such as moisture content, texture, porosity, and structural integrity. To determine the metal concentration in soil samples, (da Silva Medeiros et al., 2020) utilized ultrasound-assisted extractions. Cavitations induced by the ultrasound accelerated the disintegration of soil particles, promoting the release of trace metals like Aluminum, Cadmium, Copper, Nickel, and Zinc. This method is efficient, straightforward, and environmentally friendly compared to the traditional acid digestion in monitoring. In another study, (Wang et al., 2024b) studied the propagation mechanism of ultrasound waves at the transducer-soil interface. By examining excitation frequency and amplitude, this study revealed the mechanism of energy transport inside the soil. The result provided a foundation for creating ultrasonic soil sensors and enhancing in-situ soil evaluation. (Woo et al., 2022) used a noncontact ultrasound system to measure the moisture in the soil. Ultrasound obtained the variation in soil strength and moisture accurately across sand, silt, and clay. It also employed a machine learning approach to predict moisture levels. In another work, (Orhan et al., 2022) developed a digital ultrasound-based soil texture analyzer that estimates silt, sand, and clay content in a soil-water mixture. The system eliminates the need for soil analysis in labs and offers low-cost and portable solutions for texture analysis in agricultural applications. This was further improved by incorporating pH, electrical conductivity data, and machine learning (Kilinc and Orhan, 2025). Bradley and Ghimire (2024) designed a noncontact ultrasound system to measure the soil porosity. This approach determined the porosity with high precision, and the technique was validated for dry agricultural soils. (Xu et al., 2023) used ultrasound wave velocity measurements to access the efficacy of microbially induced carbonate precipitation in stabilizing shale soils. The wave velocity showed a correlation with unconfined compressive strength and CaCO₃ content. This technique facilitated rapid and non-destructive assessment of soil enhancement. (Choi et al., 2021) focused on using ultrasound-assisted soil washing to treat heavy metals such as Cu, Pb, and Zn. Desorption with ultrasound and mixing yielded much higher efficiency than the traditional techniques. The technique worked best on smaller particles and under milder chemical conditions. However, significant degradation of PFAS was not observed, likely due to cavitation interference from soil particles. The summary is presented in Table 2 .

In the past, numerous studies have been conducted to assess the seed quality using ultrasound. To identify slight cracks that compromise the germination in cottonseed, (Zhang et al., 2022) utilized an air-coupled ultrasound system. The technique captured ultrasound echo signals from the cottonseeds, transformed them into color-encoded images, and classified them through deep learning models. This technique detected undetectable subtle damages via optical or thermal approaches with an average identification accuracy of 90.7%, (Jin et al., 2016) utilized an air-coupled ultrasound system coupled with principal component analysis (PCA) and K-nearest neighbor (KNN) classification to identify the worm or manually induced damage in corn seed. To collect the most helpful information, the ultrasound signals from both sides of the corn seed were collected and then reduced and deionized via PCA to extract the essential features. These features were then classified by using multiple pattern recognition algorithms. Among all the algorithms, KNN achieves the highest accuracy of 100% for the intact seeds and 97% for the damaged ones. In another study, (Guo et al., 2019) used an acoustic signal processing method based on Gaussian modeling and an improved extreme learning machine to estimate the damage in wheat kernels. The signals from sprout damaged, insect damaged and undamaged kernels were processed using a short-term Fourier transform and the Gaussian parameters were extracted. The technique showed a detection accuracy of 92% for undamaged, 96% for insect damage and 95% for sprout damaged kernels. The summary is presented in Table 3 .

Numerous studies have been done in recent years on plant health monitoring. Yang et al. (2022) utilized a non-contact ultrasound method called bulk modulus elastography to determine a cactus plant’s drying behavior and health status. The study utilized novel air-coupled transducers to capture the changes in the elastic modulus of prickly pear cactus (Opuntia) pad (nopal) over 11 days. This method can capture deep tissue changes and is also efficient in the early detection of stress conditions in agricultural contexts. (Gómez Álvarez Arenas et al., 2016) also used non-contact ultrasound to monitor changes in the leaf’s mechanical properties and determine the hydration condition of plants. The technique measured the change in the frequency of the leaf’s ultrasonic thickness, which is closely related to the relative water content and water potential. In the field experiments on the common grapevine (Vitis vinifera) and Arabica coffee (Coffea arabica) leaves, the system showed high sensitivity and repeatability in determining the draught and stress conditions in the plants. Similarly, to detect northern leaf blight in maize, (Maginga et al., 2024) utilized a non-contact ultrasound-based anomaly detection system integrated with IoT sound sensors. The method used the ultrasound emissions from maize stems, taking advantage of the fact that disease-induced stress alters the plant’s physiological acoustic signature. By training Long Short-Term Memory (LSTM) models on ultrasound data from healthy plants, the system was able to detect small deviations signaling early disease. The approach effectively detected disease 4–5 days before the visual symptoms with 99.98% accuracy.

Khait et al. (2023) used ultrasonic acoustic monitoring combined with machine learning to monitor the physiological conditions of tomato and tobacco plants. The experiments were conducted in the acoustic chamber and a greenhouse under drought and mechanical stress. The study revealed that the stressed plants emit species-specific and condition-specific airborne ultrasound sounds in the range of 20–100 Khz, which could be detected from 3–5 meter distance. Using a convolutional neural network (CNN) and support vector machine (SVM), the study successfully distinguished drought-stressed from the control plant with 84% accuracy and identified the dehydration levels with 81% accuracy. Khait et al. (2023) monitored changes in the thickness resonances of plant leaves using non-contact, air-coupled ultrasonic spectroscopy to determine the physiological responses of plants to environmental stimuli. In response to sudden watering after a drought, diurnal cycles, and rapid changes in light intensity, the system detected shifts in the resonant frequency (150–900 kHz). Leaf turgor pressure, relative water content (RWC), and tissue elasticity correlated with these frequency shifts. Wang et al. (2011) detected xylem cavitation events in tomato plants using an ultrasonic acoustic emission (UAE) monitoring system to determine the degree of plant water stress and support precision irrigation. Piezoelectric transducers incorporated into a multi-parameter system that measured temperature, humidity, CO₂ concentration, light intensity, and transpiration rate were used to capture UAE signals in the 100 kHz to 1 MHz range. According to the study, the cumulative UAE signals showed clear diurnal patterns and a strong correlation with plant transpiration activity. Cavitation caused by water stress was reflected in the UAE peak’s slight lag the transpiration rate peak. Vergeynst et al. (2015) used broadband point-contact ultrasonic emission (UAE) sensing to investigate drought-induced cavitation and signal propagation in plant stems. The study looked at how ultrasonic signals travel from the source, such as xylem cavitation events, to the sensor by combining experimental detection of UAE signals in dehydrating branches with finite element modeling. This framework enables the differentiation of near- and far-field signals in woody species like Vitis vinifera and Fraxinus excelsior due to the AE source dynamics associated with abrupt xylem tension release. This method provides a high-resolution tool for monitoring xylem embolism under drought stress.

Charrier et al. (2015) used infrared thermography and UAE to study the ice nucleation and propagation in woody plants to investigate the response to freezing. The study found that the signals were strongest close to the nucleation point and decreased significantly with distance. This indicates a strong correlation with ice formation’s temporal and spatial dynamics. This work also concludes that the AEs were caused by cavitation events triggered by the tension at the ice–liquid interface, with the UAE source effectively following the moving ice front. This method is helpful for cold stress research and monitoring plant freezing resistance. To detect the internal and near-surface defect in trees, (Qiu et al., 2019) proposed a novel technique by integrating stress wave sensing with acoustics laser methods. Due to the limited propagation of surface waves, conventional sonic tomography has trouble identifying defects close to the bark. The authors solved this problem by combining data on stress wave transmission with measurements of surface vibration made with laser vibrometry, which measures variations in vibration amplitude caused by defects in the subsurface. The method was tested on Cinnamomum camphora trunks with 5–50 mm-deep artificial air holes and gaps. Internal (core) defects were clearly visible with conventional sonic tomography, but shallow defects (less than 25 mm deep) were not. Down to 5 mm from the surface, the integrated method successfully detected both internal holes and shallow defects, demonstrating significantly higher sensitivity and resolution. Bonisoli et al. (2025) used non-contact ultrasound to monitor plant drought stress in indoor and outdoor conditions. Tomato plants and pinto beans were subjected to water stress, and it was noticed that the stressed plants emit more ultrasound signals than hydrated controls. This study also considered the external noise factors, including wind, rain and insect chirping. The summary is presented in Table 4 .

Ultrasound is conventionally defined as sound waves with frequencies above 20 kHz. However, in practical agricultural monitoring applications, systems operating just below this threshold, particularly in the 15–20 kHz range, are often included under ultrasound-based monitoring. This is because they utilize similar high-frequency sensor technologies, signal processing methods, and non-invasive approaches designed for detecting internal biological activity. Many insect species, especially larvae of soil- and wood-dwelling pests, produce informative signals such as stridulations and feeding sounds within this upper-audible to near-ultrasonic band. As a result, broadband acoustic systems spanning both audible and ultrasonic frequencies are commonly deployed. These systems are functionally aligned with ultrasound monitoring and are therefore considered relevant to the scope of this review (Mankin et al., 2011; Pinhas et al., 2008). In the past decade, many researchers have focused on the early detection of Rhynchophorus ferrugineus, also known as red palm weevil (RPW), a highly destructive pest of palm species. To analyze the ultrasound signals from RPW, (Martin et al., 2015) used an ultrasound-based acoustic technique to investigate the intensity and frequency of sounds emitted by RPW in coconut palms. The measurements were collected in both laboratory and field settings and further analyzed for distinctive spectral features. The system presented in this research did not give any real-time information, but the work laid the groundwork for the identification of pest-specific ultrasound patterns.

Hetzroni et al. (2016) used piezoelectric ultrasound sensors to detect larval chewing activity in date and canary palms. This study compared human and machine-assisted monitoring and demonstrated that both are feasible for detecting RPW in natural field conditions. The study also reported a human-assisted accuracy of 85% and a machine-assisted accuracy of 95%. Mankin et al. (2016) utilized advanced ultrasound signal processing to detect the RPW larval activity in commercial palm orchards. The study analyzed temporal and spectral ultrasound patterns to differentiate the environmental noise from the pest activity. The method was tested in the field and has shown capabilities in detecting multispecies of pests. To implement a scalable ultrasound-based monitoring system, (Ashry et al., 2022) introduced a distributed acoustic sensing (DAS) system that uses optical fiber to detect the ultrasound signals produced by RRW larvae. The system used phase-sensitive optical time domain reflectometry to detect the larval chewing activity in the 200–800 Hz frequency range. The system could capture RPW larval ultrasound emissions as early as 12 days post-infestation. The study also used a custom signal processing algorithm based on signal-to-noise ratio (SNR) to distinguish between infested vs healthy palm trees. The algorithm could detect 97 infestations in the infested trees, whereas only 9 in healthy trees. This system was further enhanced by deploying convolutional neural networks (CNN) in both controlled and outdoor farm environments. CNN was used to process the signals and classify them as infested’ or healthy. In the outdoor field trials the system showed an RPW larvae classification accuracy of 97%. Wang et al. (2021) also employed a fiber optic distributed acoustic sensing system integrated with machine learning to enhance early ultrasound-based detection of RPW in noisy farm environments. The system was used to record signals under various noisy conditions, and artificial neural networks (ANN) and CNN were trained to classify healthy vs infested trees. The study reported a classification accuracy of greater than 99% in combined noisy scenarios.

In addition to detecting RPW monitoring, researchers have also explored other agricultural pests. In coffee plantations, (Escola et al., 2020) developed a real-time acoustic detection system that utilized wavelet packet transform, bark scale filtering, and SVM classifications to detect and distinguish Quesada gigas cicadas signals from background noise. The study reported an accuracy of 96.41%. This work was further expanded by (de Souza et al., 2022) they integrated Paraconsistent Feature Engineering (PFE) and Empirical Mode Decomposition (EMD) for feature extraction, and the study reported classification accuracies of above 98% and reported the suitability of the system for smart farm deployment. In soil and root crop systems, (Görres and Chesmore, 2019) used acoustic monitoring system to detect the stridulation patterns of Melolontha and M. hippocastani larvae in soil. The study deployed a fractal dimension-based algorithm to distinguish stridulation events from the background noise. The automated analysis effectively detected the stridulation rates, which strongly correlated with the larval abundance. Nanda et al. (2023) used both acoustic and temperature signals to detect Coptotermes curvignathus in pine boards, which is considered one of the most damaging termites in Indonesia. The dual-sensing system used in the study was used to monitor real-time feeding, excavation, and alarm behavior. The study successfully detected the termite activity from the background noise over a 24-hour monitoring period. Additionally, the study also reported a rising temperature averaging 0.101 °C between healthy and infested samples. A regression model also confirms strong correlations between termite population and both temperature and acoustics signal duration. Table 5 summarizes the work reviewed in this section.

Several studies in the literature highlight the usage of ultrasound in monitoring fruit ripeness. This section will explore studies done in the last decade. Miraei Ashtiani et al. (2016) used ultrasonic spectroscopy to determine postharvest changes in the chemical and mechanical properties of the persimmons fruits. The study used Teflon-coated ultrasonic transducers to monitor ultrasonic velocity and attenuation as the fruit ripened over 21 days. These parameters were further statistically modeled with the coefficient of determination (R ²) value greater than 0.82 to predict the modulus of elasticity, rupture force, and soluble solid content (SSC) that indicates the ripeness and quality of the fruit. Vasighi-Shojae et al. (2018) used a portable ultrasonic system to determine apples’ firmness, rupture energy, and elastic modulus. This study measured the velocity and attenuation through the entire fruit and developed multiple linear regression models to predict the properties of the fruit. The model demonstrated a promising predictive capability with R ² values up to 0.73, indicating moderate to high correlation. This technique was further enhanced by combining ultrasonic measurements with artificial neural networks (ANN) by (Vasighi-Shojae et al., 2020). Using the measured acoustic properties of apples, ANN models were developed to predict firmness, elastic modulus, and stiffness. The models developed in this work demonstrated a high predictive accuracy with R ² = 0.99 for all three properties.

Soltani Firouz et al. (2021) used a custom ultrasound system together with a support vector machine (SVM) classifier to determine the mechanical effects of freeze in oranges. Ultrasonic velocity and attenuation were measured through the fruit, and the impact of freezing was analyzed. The study found that the freezing altered the fruit’s internal properties, affecting ultrasound propagation, which can be detected immediately after freezing and before the visual symptoms appear. Using SVM classifiers, the system achieved an accuracy of 100% in distinguishing healthy, mildly damaged, and severely freeze-damaged oranges. Budiastra and Jannah (2022) utilized ultrasonic methods to determine the firmness and sweetness of soursop fruit. Velocity and attenuation were measured using the ultrasound system, which showed a strong correlation, with physicochemical parameters firmness having R ² = 0.884 and total soluble solids (TSS) having R ² = 0.81. In the study, two regression models were also developed to classify firmness and sweetness levels, with a classification accuracy of 100% and 95%, respectively, when validated against manual measurement on 20 randomly selected samples. Table 6 summarizes the work reviewed in this section.

Several specialized ultrasound-based systems are commercially available for in-field crop and tree monitoring. For the structural health of trees, sonic tomograph devices are used to detect internal decay or cavities non-invasively. Notably, the PiCUS Sonic Tomograph (Argus Electronic GmbH, Germany) (IML Electronic GmbH, 2021) and ArborSonic 3D (Fakopp Enterprise, Hungary) employ multiple contact piezoelectric sensors around a trunk to measure sound wave transit times and construct cross-sectional images of wood integrity (Fakopp Enterprise Bt, 2023). Similarly, the Arbotom system (Rinntech, Germany) (Rinntech GmbH, 2022) uses impulse sonic tomography to map tree trunk and limb quality for risk assessment. These systems are field-portable and designed for outdoor use by arborists or growers to evaluate trunk and canopy structural health (e.g., detecting hollows, rot, or cracks) without harming the tree.

Ultrasound sensing has also been applied to pest and stress monitoring. For example, IoTree sensors (Agrint, Israel) are wireless in-tree seismic/ultrasonic devices that detect vibrations from wood-boring insect larvae (such as red palm weevil) inside live palm (Agrint Ltd, 2024). This contact sensor network is deployed in orchards and plantations, enabling early pest infestation detection under real field conditions. In the domain of plant physiological stress, emerging solutions like Plense Technologies’ system (Netherlands) use high-frequency acoustic sensors to “listen” to plants. Plense’s ultrasound sensor employs small ultrasonic microphones and speakers near the plant to detect xylem cavitation sounds and measure water content, providing real-time monitoring of drought stress in greenhouse or field (Plense Technologies, 2024). These examples illustrate the range of commercial ultrasonic technologies now available for in-field agricultural monitoring, from tree health diagnostics to pest detection and crop stress sensing. Table 7 summarizes the commercially available devices.