The intersection of machine learning and agriculture has witnessed tremendous growth over the past few decades, evolving from rudimentary applications of traditional machine learning models to the sophisticated deep learning architectures employed today (Abbas I. et al., 2021). Early research in agricultural data analysis largely focused on conventional machine learning methods, including decision trees (Abdar et al., 2021b; Zhang et al., 2025b), support vector machines (Pérez et al., 2017), and k-nearest neighbors (Zhang et al., 2020), applied to relatively simple datasets such as crop disease classification or pest detection. While these models proved useful in certain contexts, their limitations in handling large, complex datasets became apparent as agriculture entered the era of big data.

With the advent of deep learning in the 2010s, a paradigm shift occurred in agricultural research. Convolutional Neural Networks (CNNs) and other deep learning models revolutionized the way crop monitoring tasks, such as disease identification and pest detection, were performed (Bao et al., 2024). These models excel at recognizing complex patterns in vast datasets, such as high-resolution images captured by drones and satellites, which traditional machine learning algorithms could not process as effectively (Abdulkadirov et al., 2023). The combination of deep learning with remote sensing technologies has been particularly impactful, allowing for high-resolution, real-time monitoring of crops over large areas (Abbas I. et al., 2021). UAVs, satellites, and ground-based sensors have enabled researchers and farmers alike to gather massive amounts of data, which can be processed to extract meaningful insights that were previously impossible to obtain (Jumaat et al., 2018).

The evolution of this research has been global in nature, with countries such as China, India, the United States, Japan, and various European nations, including the United Kingdom, Germany, and France, leading the charge in advancing AI-powered agricultural technologies (Alola et al., 2023; Dimyati et al., 2023; European Assoc Study Live et al., 2023; Guo et al., 2023; Abbas et al., 2024; Adger et al., 2024). These nations have invested heavily in AI-driven agricultural technologies, with numerous advancements made in crop health monitoring, pest management, and the optimization of resource use (Loeffler et al., 2024). In China, for instance, AI-powered systems have been employed for real-time pest surveillance, while India has seen significant progress in the use of deep learning for disease diagnosis in crops like rice and tomatoes (Xue et al., 2024). As AI techniques continue to mature, the scope of applications has expanded from image classification and disease detection to more sophisticated systems that can predict crop yields, monitor soil health, and even automate tasks like harvesting (Tassis et al., 2021). While the research is progressing rapidly, there remains much to be done to overcome challenges in scalability, data heterogeneity, and system integration, which are crucial for ensuring that these innovations can be widely adopted by farmers globally.

Despite the remarkable progress made in integrating machine learning and remote sensing technologies into agriculture, several significant challenges must be addressed for these technologies to reach their full potential:

This study seeks to address these challenges by employing a rigorous bibliometric analysis and knowledge graph methodology to map the evolution of research in agricultural information identification and monitoring (Chaw and Mokji, 2017). By systematically analyzing the body of literature on the application of machine learning and remote sensing technologies in agriculture, this paper identifies key trends, breakthroughs, and gaps in current research. The knowledge graph approach offers a visual representation of the interconnectedness of various research topics, providing insights into how machine learning and AI have been integrated into agricultural practices over time (Abubakar et al., 2022; Ahmed et al., 2023; Al-Shammary et al., 2024).

In addition to the bibliometric analysis, this paper presents a detailed examination of the technical aspects involved in the integration of AI and remote sensing technologies. Specifically, it explores how deep learning models, such as Convolutional Neural Networks (CNNs) and Transfer Learning, are being used to analyze crop images, detect diseases, and identify pests (Chen et al., 2020). Furthermore, the paper delves into the challenges associated with the use of remote sensing technologies, such as UAVs and satellite imagery, in large-scale agricultural monitoring (Zhang et al., 2025a). By synthesizing existing literature and examining case studies from different regions and crop types, this research provides a comprehensive overview of the current state of AI-driven agricultural monitoring systems and outlines the future directions necessary to overcome the existing barriers.

The motivation for this study stems from the increasing challenges faced by global agricultural systems, including climate change, resource scarcity, and the growing demand for food. Traditional farming methods are becoming insufficient to meet these demands, necessitating the integration of advanced technologies for sustainable, efficient agricultural practices. Machine learning, particularly deep learning, has demonstrated its potential to revolutionize agricultural monitoring by improving accuracy in crop disease detection, pest management, and yield forecasting. However, significant gaps remain, particularly in data quality, model generalization, and multi-source data integration (Seiya et al., 2020; Wei et al., 2025). This study aims to address these gaps by developing AI models that can adapt to diverse agricultural environments, improve crop monitoring accuracy, and promote sustainable farming practices, ultimately contributing to global food security.

The current study provides several key contributions to advancing AI-driven agricultural practices, particularly in crop monitoring, pest detection, and disease diagnosis. First, this research addresses the challenges associated with integrating deep learning algorithms and remote sensing technologies, proposing a novel framework that combines these tools for more accurate, real-time crop monitoring. A critical contribution of this study is the development of an advanced deep learning model tailored to agricultural applications, showing significant improvements in pest detection and disease diagnosis compared to traditional methods. Furthermore, this study introduces a model capable of generalizing across diverse agricultural environments, which is essential for adapting to varying crops, pests, and climatic conditions without requiring extensive retraining. Additionally, by improving the precision of agricultural practices, the research contributes to more sustainable farming, optimizing resource use while enhancing crop yield predictions and pest control measures. These contributions lay the groundwork for future AI-driven agricultural systems, offering innovative solutions to some of the most pressing challenges in global food security and agricultural sustainability.

In conclusion, this study not only highlights the advancements in AI and remote sensing for agriculture but also provides a roadmap for future research, pointing out key areas that require further exploration, such as data standardization, model generalization, and multi-source data integration. The manuscript begins with a comprehensive background on the challenges faced by global agricultural systems and the potential of AI in addressing these issues. It then explores the integration of deep learning models with remote sensing technologies for crop monitoring. The subsequent sections focus on the methodology, data analysis, and key findings related to improving model generalization and enhancing agricultural sustainability. Finally, the manuscript concludes by discussing the implications of the study and the future directions for AI-driven agricultural systems. Thank you for your valuable feedback. Advancing the understanding of these technologies and addressing their limitations will unlock their full potential to transform agricultural practices, enhancing sustainability, efficiency, and the capacity to meet the growing demands of global food production.

The literature search was conducted using the Web of Science (WOS) and Engineering Index (EI) databases, two of the most authoritative academic databases for multidisciplinary research (Ma et al., 2023). WOS, developed by Clarivate (formerly Thomson Reuters), encompasses the Science Citation Index (SCI), Social Sciences Citation Index (SSCI), and Arts & Humanities Citation Index (A&HCI), renowned for its rigorous journal selection and citation indexing system that enables comprehensive literature retrieval through multiple search parameters including keywords, authors, and DOI. Similarly, EI, established in 1884 by the American Federation of Engineers, serves as a foundational database in engineering disciplines with global recognition. The search strategy employed the query TS = (“image recognition” AND “deep learning” AND “crops”) within the WOS Core Collection, covering all editions and publication years (2000–2025), which initially retrieved 785 records. A corresponding search in EI up to the present date yielded 142 records. After cross-database comparison and removal of 192 irrelevant publications and 85 consolidated items, 650 valid records were retained for analysis, including five highly representative studies that exemplify key advancements in deep learning-based image recognition for crop analysis (Figure 1). This systematic approach ensured a rigorous and comprehensive collection of relevant literature in the field.

In recent years, the application of deep learning technologies in agriculture, particularly for crop disease and pest detection, has seen significant advancements. With the modernization of agricultural production, traditional manual monitoring methods are no longer sufficient to meet the demand for fast and accurate detection. As a result, using deep learning models for disease and pest identification has become an essential tool for improving agricultural productivity. One study focuses on the automatic identification and diagnosis of rice diseases using deep convolutional neural networks (CNNs), providing a powerful framework for disease detection that enables early intervention to ensure crop health (Lu et al., 2017). Another research explores the use of deep learning, particularly Mask-RCNN, for strawberry harvesting robots, contributing to agricultural robotics by optimizing fruit detection, which increases efficiency and reduces labor costs (Yu et al., 2019). A third paper examines the use of deep transfer learning for plant disease diagnosis, demonstrating how pre-trained models can improve the accuracy and reliability of disease detection in varied environmental conditions, thus enhancing crop management practices (Chen et al., 2020). A fourth study applies transfer learning techniques to detect tomato plant diseases, highlighting the importance of deep learning in early disease detection and control, which supports sustainable agriculture practices (Abbas A. et al., 2021). Finally, the fifth paper introduces a method for classifying pests in crops using deep convolutional neural networks, offering a solution for pest management, which is crucial for protecting crops from harmful pests and ensuring better yields (Thenmozhi and Reddy, 2019). These contributions collectively highlight the pivotal role of deep learning in advancing agricultural technologies, focusing on disease and pest detection, and improving the efficiency and sustainability of crop management (Table 1).

Bibliometric analysis is a systematic, quantitative approach used to summarize and evaluate research publications. It involves four primary steps: sampling, recording entries, cataloguing, and metrics. The first three steps focus on standardizing data, including titles, authors, keywords, publication dates, methodologies, and citation counts. The final step, metrics, extracts valuable insights to guide future research. This process is greatly facilitated by software tools such as VOSviewer (version, 1.6.16) and CiteSpace (version, 6.3.R1). VOSviewer, developed by the Centre for Science and Technology Studies (CWTS) at Leiden University, specializes in visualizing scientific knowledge and analyzing undirected networks. CiteSpace, created by Professor Chen Chaomei at Drexel University, is used for citation analysis and identifying potential knowledge within scientific research. Both tools help uncover trends, research hotspots, and the evolution of scientific fields (Bouguettaya et al., 2022).

The knowledge graph for this study was constructed using VOSviewer and CiteSpace, two prominent tools in bibliometric analysis. VOSviewer was used for visualizing the relationships between key terms, authors, and journals. The tool applied co-occurrence analysis to generate term maps, where terms frequently appearing together in the dataset were grouped into clusters. These clusters represent research topics or themes within the field of AI and agriculture. The visual representation provided by VOSviewer highlighted the major research areas in crop monitoring, such as disease detection and pest management, as well as the dominant methodologies like Convolutional Neural Networks (CNNs) and remote sensing integration. CiteSpace was employed to conduct citation analysis, identifying influential papers, key authors, and significant trends in the field over time. CiteSpace’s citation burst detection feature was particularly useful in identifying rapidly emerging topics, allowing for the identification of key shifts in research focus. The software also provided a visualization of the collaborative networks between authors and institutions, enabling us to examine how knowledge flows and research collaborations have evolved in the AI-driven agricultural field.

By combining these two tools, we were able to not only extract key research themes and identify gaps in the literature but also understand the temporal evolution and collaborative dynamics within the field. The methodology allowed us to construct a comprehensive knowledge graph that reveals how AI and deep learning technologies are shaping agricultural research and practice.

This paper employs two primary research methods: bibliometric analysis and knowledge mapping analysis, to explore the analysis of agricultural information identification and monitoring. Various bibliometric tools, including WOS, Scopus, VOSviewer, and SCImago Graphica, are utilized to process data and analyze key aspects such as publication times, contributions, and highly cited works. VOSviewer is particularly useful for co-occurrence analysis and the visualization of research hotspots, while CiteSpace is applied to examine research cooperation, identify emerging keywords, and predict future trends (Iyortsuun et al., 2023; Rubbens et al., 2023). By combining these software tools, the study maximizes their respective strengths, offering a comprehensive view of the research field’s development and providing valuable insights into the ongoing evolution of scientific knowledge in this area.

In the realm of machine learning applications for crop image recognition, several authors have made pivotal contributions, shaping the field with their innovative approaches. Zhang (Y) has significantly advanced the automation of crop ripeness detection, particularly in jujube fruits, by developing AGHRNet: An attention ghost-HRNet for confirmation of jujube fruit ripening stage. This attention-based deep learning model, with its exceptional performance, has greatly enhanced precision in crop management systems, contributing extensively to the recognition of fruit ripeness and disease detection (Zheng et al., 2023). Following closely, Chen (C) has made important strides in improving crop disease recognition by addressing the challenges posed by occluded images. His work on HOB-CNN: Hallucination of occluded branches with convolutional neural networks introduced methods to handle imperfect data, making crop disease detection more reliable even in the presence of visual obstructions (Chen et al., 2023). Meanwhile, Saraswat (D) has focused on optimizing deep learning models for large-scale agricultural settings, particularly in crop disease detection. His contributions, including Performance of deep learning models for classification and detection of crop diseases, have greatly improved detection accuracy, making real-time agricultural monitoring more efficient and actionable (Ahmad et al., 2021). He (Y) has significantly advanced crop image recognition through his work on Object detection with attribute tagging task: A novel approach for crop image recognition. His research integrates object detection with attribute tagging, improving crop classification accuracy and enabling systems to capture additional contextual information, such as growth stages. This approach enhances real-time crop monitoring, contributing to more efficient and scalable agricultural management systems (Liu D. et al., 2025). Lastly, Fu (LS) has contributed to the field by introducing new segmentation techniques for crop disease detection. In Deep learning based segmentation for automated crop disease detection, Fu’s innovative methods have improved the precision of disease detection, particularly in tree fruits and vegetables, pushing the boundaries of crop image recognition (Majeed et al., 2020). Together, these authors have significantly refined machine learning models for crop recognition and disease management, with each making unique contributions that continue to propel the field forward, offering solutions for more accurate, real-time, and scalable crop monitoring in agriculture. The work of these five authors shares a common focus on advancing machine learning models for crop image recognition, particularly in disease detection and classification, yet each brings a unique perspective, with some emphasizing deep learning segmentation techniques, others integrating multi-source data, and a few focusing on the automation and real-time applications of crop monitoring systems. The top 5 authors and their representative contributions are listed in Table 2.

To enhance the visual representation of the geographic distribution of publications by country or region, this study first saves the merged and de-duplicated data from the WOS database in a standardized *.txt format. The word frequency threshold is then set to 10 in VOSviewer, allowing for the identification of countries or regions with significant contributions to the field. After processing, the data is saved in *.gml format and imported into Scimago to create a geographical visualization of document distribution, with appropriate map decorations. Figure 3 illustrates the distribution of countries leading globally in crop image recognition research. A total of 11 countries have published more than 10 documents. An analysis of the top five contributing countries reveals notable patterns in the global distribution of crop image recognition research. China leads the field with 208 publications, reflecting its dominant role and the active, extensive research community engaged in this area. India follows with 88 publications, marking its significant contribution to agricultural research, particularly in precision farming and remote sensing. The USA ranks third with 40 publications, demonstrating substantial involvement in crop image recognition, though at a smaller scale compared to China and India. Australia (26 publications) and Brazil (19 publications) round out the top five, indicating growing research efforts, especially in precision agriculture and related technologies. Overall, the top five countries emphasize the prominent positions of China and India in advancing crop image recognition research, with the USA and other nations contributing significantly, though on a relatively smaller scale.

After synthesizing the data obtained from WOS, a total of 35 journals were identified, with 12 publishing more than three documents. This suggests significant interest in the research area of crop image recognition. Figure 4 displays the top ten journals. The top three journals in terms of publication count are COMPUTERS AND ELECTRONICS IN AGRICULTURE (109 publications), IEEE ACCESS (29 publications), and SENSORS (27 publications), with corresponding h-index values of 12, 9, and 8, respectively. The top three journals in crop image recognition research, COMPUTERS AND ELECTRONICS IN AGRICULTURE (109 publications), IEEE ACCESS (29 publications), and SENSORS (27 publications), focus on different aspects of the field. COMPUTERS AND ELECTRONICS IN AGRICULTURE primarily publishes papers on computational methods and electronics in agriculture, with a strong emphasis on precision farming, remote sensing, and automation systems for crop monitoring. This journal has seen a consistent increase in publications, particularly from 2010 onwards, as studies combining machine learning and image processing in agricultural applications have grown. In contrast, IEEE ACCESS accepts a broader range of interdisciplinary papers, focusing on the integration of computer science, electronics, and agricultural engineering, especially in developing advanced algorithms for crop image analysis. This journal has experienced a notable rise in publications over the last 5 years, reflecting the growing influence of AI and big data in crop image recognition. Lastly, SENSORS publishes research centered around sensor technologies, particularly in remote sensing and the use of hyperspectral and multispectral imaging for crop health monitoring. Publications in this journal have gradually increased since 2015, as sensor technologies have advanced, enabling more widespread application in agricultural monitoring. Together, these journals reflect a strong shift toward AI, machine learning, and advanced sensor technologies in crop image recognition, with an upward trend in these areas in recent years.

These five papers have attracted significant attention in the field of crop image recognition due to their application of advanced machine learning techniques, particularly deep learning, which have significantly enhanced the accuracy and efficiency of agricultural monitoring systems. A common theme across these papers is the integration of cutting-edge neural network models, such as deep convolutional networks (CNNs) and improved versions of YOLO, to address the challenges of crop detection, disease identification, and fruit harvesting automation. The adoption of techniques like transfer learning has further contributed to the high impact of these papers, as it enables models to generalize well across different crops and environments, even with limited data. The focus on real-time applications, such as automated fruit harvesting and disease detection, resonates with the growing demand for precision agriculture, where efficient, cost-effective, and scalable solutions are paramount. Moreover, the timing of these papers aligns with advancements in sensor technologies, making them even more relevant as the agricultural industry increasingly turns to remote sensing and AI-based tools for monitoring crop health. This convergence of deep learning, data-driven solutions, and agricultural needs has made these papers essential references in the field, driving innovation and inspiring further research in crop image recognition and automation (Table 3). In summary, these papers are highly cited for addressing critical issues in modern agriculture—crop detection, disease identification, and automation—while utilizing advanced machine learning techniques such as deep learning and transfer learning, which enhance their relevance in the field of crop image recognition and precision agriculture.

With the advancement of agricultural technologies, crop recognition and monitoring have become essential areas of study in modern agricultural management. The following four subfields have seen widespread application in crop monitoring and recognition (Figure 12).

In the field of Crop Image Processing and Computer Vision, research is concentrated around two main directions: Crop Phenology and System Development. These studies leverage image processing technologies, computer vision models, and data collection methods to provide key technological support for precision agriculture and crop monitoring (Figure 13).

Crop Phenology research focuses on monitoring crop growth and development through imaging technologies, including seasonal changes, climate impacts, and other growth factors. Specific areas of research include ground-based imaging, satellite imagery, drone imaging, and remote sensing technologies for crop growth monitoring. For instance, research indicates that ground-based imaging technology can accurately capture the growth status of crops, thus optimizing crop management (Chaw and Mokji, 2017). Moreover, satellite imagery has been applied to monitor the overall growth status of crops, providing crucial data support for agricultural decision-making (Xue et al., 2024). Drone imaging and remote sensing technologies are widely used in crop monitoring, especially for real-time monitoring of large-scale agricultural fields (Abdullahi et al., 2015). In addition, modeling crop phenology has gained significant attention. By constructing climate impact models, crop growth stage models, and seasonal change prediction models, researchers are able to better understand crop growth patterns and predict future growth trends (Guo et al., 2024). Research has also explored the role of climate impact modeling on crop growth cycles, shedding light on the potential effects of climate change on agricultural production (Li et al., 2024). In terms of data collection, crop phenology relies heavily on advanced sensors and data collection technologies, particularly for soil moisture, environmental data, and field data collection. Soil moisture sensors and environmental data collection techniques are key components of the research, enabling precise monitoring of the environmental conditions affecting crop growth (Beck et al., 2021). Field data collection techniques also play a crucial role, providing valuable data for in-depth studies in crop phenology (Jiang et al., 2020).

On the System Development side, the focus is primarily on the development of computer vision models, leveraging deep learning and pattern recognition technologies to enhance crop image processing. For example, object detection and image segmentation techniques are widely used in crop disease detection and crop classification (Liu D. et al., 2025). Furthermore, deep learning has been applied to train image processing models, improving the accuracy of crop monitoring through efficient image recognition techniques (Ahmad et al., 2021). Research on crop detection systems is focused on integrating various image processing technologies and machine learning models to develop automated systems for precision agriculture (Dong et al., 2024). These systems are capable of real-time analysis of crop images, identifying different crop types or diseases. The study of sensor integration addresses the challenge of integrating multiple sensor data into crop monitoring systems to enhance the precision and efficiency of monitoring (Alahi et al., 2023).

In summary, research in Crop Image Processing and Computer Vision has provided crucial support for the advancement of precision agriculture and crop phenology, particularly in crop monitoring, environmental monitoring, data analysis, and system development. These efforts are driving the automation and intelligence of agricultural production processes.

In the field of agricultural image processing and computer vision, scholars have continually explored advanced techniques to improve crop assessment and pest monitoring efficiency. A computer vision-based system was developed for automated evaluation of corn crop quality, combining image processing with machine learning algorithms to assess aspects such as health, size, and ripeness. This system significantly enhances the accuracy and efficiency of crop quality monitoring, although it faces challenges in adapting to varying environmental conditions. To address these challenges, a method for pest detection in winter grains was introduced, utilizing image classification and machine learning algorithms. By strengthening image preprocessing, this approach improved classification accuracy under different lighting and background conditions, demonstrating its robustness in real-world applications (Pérez et al., 2017). In addition, a multimodal produce recognition system was proposed that fuses tactile sensor data with visual imagery. This data fusion technique enhanced classification accuracy, particularly in recognizing produce with complex physical attributes, thus paving the way for more reliable agricultural automation (Chaw and Mokji, 2017). Furthermore, a deep convolutional neural network (CNN) was applied to classify vegetables with high efficiency. The system excelled in handling diverse datasets under varied lighting and backgrounds, showcasing the power of deep learning in agricultural image classification (Zhu et al., 2018). Similarly, a multi-class fruit detection system based on region proposal networks (RPN) was developed, overcoming the limitations of traditional methods by accurately detecting multiple fruit classes even in cluttered and overlapping scenes. This method significantly outperformed prior techniques (Kuang et al., 2018). Finally, a CNN-based crop pest classification system was introduced, capable of accurately identifying and categorizing different pests, even in complex and cluttered environments. This model demonstrated high robustness and accuracy, making it an effective tool for pest monitoring (Thenmozhi and Reddy, 2019). Collectively, these studies illustrate the growing potential of computer vision and deep learning technologies to revolutionize agriculture, enhancing automation and precision in crop and pest management. Table 4 shows several typical cases in the documents.

The bar chart presents the frequency of various deep learning algorithms used in crop monitoring and agricultural applications, as observed across multiple studies (Figure 14). Convolutional Neural Networks (CNN) and their variants, such as the traditional “Convolutional Neural Network” and “CNN,” were the most frequently employed algorithms, with the highest counts of 402 and 410, respectively. This widespread use of CNN is consistent with findings from various research, where CNNs have been shown to excel in image classification, plant disease detection, and crop segmentation tasks (Rakun et al., 2011; Zhu et al., 2018). Deep Neural Networks (DNN) followed with a count of 48, demonstrating a moderate level of usage for tasks requiring more complex feature extraction and representation learning (Ferreira et al., 2019; Majeed et al., 2020). Recurrent Neural Networks (RNN) were also used but with lower frequency (16 occurrences), indicating their more specialized application in time-series analysis for crop growth prediction (Shi et al., 2019; Hamidisepehr et al., 2020). Generative Adversarial Networks (GANs), Autoencoders, and other more specialized algorithms were less frequently mentioned, suggesting their emerging use in crop monitoring and further exploration in future research (Lu et al., 2015; Wu et al., 2020). The chart highlights the dominance of CNN in the field, reflecting its effectiveness in image-based tasks such as plant health assessment and pest detection, and its continued relevance in advancing agricultural automation through artificial intelligence.

In recent years, deep learning and artificial intelligence have emerged as pivotal tools for enhancing crop monitoring and agricultural automation. One significant challenge is the accurate classification of vegetables, which often varies under different environmental conditions. To address this, a deep learning-based approach using Convolutional Neural Networks (CNNs) was developed, demonstrating high classification performance even in complex lighting and background settings (Zhu et al., 2018). A related challenge involves the segmentation of plant structures for better understanding and analysis. Deep learning-based segmentation models have been applied to separate various plant parts, improving the precision of plant structure analysis and offering more accurate segmentation (Shi et al., 2019). Additionally, plant disease detection, traditionally reliant on labeled data, has been transformed through unsupervised deep learning techniques. These methods detect diseases without the need for labeled training data, enhancing scalability and efficiency in large-scale agricultural settings (Ferreira et al., 2019). Another area where deep learning has proven beneficial is in fruit detection, where deep learning-based segmentation networks have automated the identification and segmentation of fruits, paving the way for more efficient fruit harvesting (Chou et al., 2019). The detection of crop pests, a critical concern in agriculture, has also been improved through object detection methods based on deep learning. These models provide high accuracy, surpassing traditional methods in pest detection (Hamidisepehr et al., 2020). Furthermore, deep learning combined with remote sensing has been utilized to extract valuable apple tree crown information, assisting in precision agriculture by enabling more accurate crop monitoring and management (Wu et al., 2020). Collectively, these studies highlight the transformative potential of deep learning and artificial intelligence in agricultural automation, addressing key challenges in crop classification, disease detection, pest identification, and plant monitoring (Table 5).

One critical issue in orchard management is the detection of fruit tree pests, which can now be addressed using deep learning algorithms applied to remote sensing data, improving pest detection accuracy and efficiency (Chen et al., 2021). Similarly, rice lodging, a major issue affecting crop yield, has been tackled through the development of machine learning models that leverage remote sensing data, offering a more accurate and scalable solution for monitoring crop health and ensuring better management practices (Su et al., 2022). Moreover, maize tassel detection, vital for crop management, has been significantly improved by utilizing drone imagery combined with remote sensing techniques, allowing for precise identification of maize tassels and supporting better yield prediction (Zhang X. et al., 2023). In orchard management, the use of drone and remote sensing technologies has enabled the efficient mapping of individual fruit trees, facilitating better monitoring and resource management (Chou et al., 2019). Finally, rice seedling distribution, a crucial aspect of crop management, has been predicted with high precision using remote sensing, contributing to improved crop planning and yield optimization (Xiong et al., 2024). Collectively, these studies demonstrate the growing potential of remote sensing and drone monitoring technologies in revolutionizing agricultural practices, providing solutions that enhance both the accuracy and efficiency of crop monitoring and management (Anandakrishnan et al., 2025) (Table 6).

The pie chart presents the distribution of remote sensing technologies used in crop monitoring across the literature (Figure 15). Multispectral and hyperspectral technologies dominate, with multispectral sensing being widely applied for vegetation index calculation and pest detection, as shown in several studies (Hao et al., 2025). Hyperspectral sensing, known for its high spectral resolution, is primarily used for material identification and disease detection, contributing to more detailed crop analysis (Meng et al., 2022; Tang et al., 2022; Zhang et al., 2024). While less common, LiDAR technology is gaining traction in applications such as topographic mapping and biomass estimation, with significant potential for precision agriculture (Jayakumari et al., 2021; Wang et al., 2022). Thermal sensing, although mentioned in fewer studies, plays a crucial role in detecting temperature variations, such as water and heat stress in crops (Gill and Khehra, 2020). These technologies collectively offer a comprehensive suite of tools for enhancing crop monitoring, improving decision-making, and fostering precision agriculture practices.

Based on the technologies identified in the diagram, the integration of algorithms, software technologies, hardware devices, and research objects forms the backbone of modern agricultural IoT and automation systems, driving advancements in precision agriculture (Figure 16). As highlighted in recent studies, the application of Convolutional Neural Networks (CNNs) for image-based analysis is pivotal in automating tasks such as fruit and vegetable detection, pest and disease identification, and crop health monitoring (Loss et al., 2009). These deep learning algorithms, particularly Deep Semantic Segmentation, enable highly accurate pixel-level classification in images captured by drones and autonomous robots, allowing for precise plant-part segmentation (Rakun et al., 2011). In terms of hardware devices, drones have emerged as a critical component in agricultural automation systems, providing mobility and the ability to capture high-resolution data through camera sensors such as RGB, thermal, and hyperspectral imaging (Chaivivatrakul and Dailey, 2014). Coupled with autonomous robots and robot arms for automated harvesting, these systems enhance the efficiency of tasks like fruit picking and pest monitoring, reducing human intervention (Lu et al., 2015). The implementation of multispectral imaging systems further improves the ability to monitor crop health and growth stages, providing a detailed analysis of soil and plant conditions (Zhu et al., 2018). On the software side, computer vision algorithms are employed to process the large amounts of data generated by these devices. Techniques such as data synthesis and augmentation tools improve model training, especially in environments with limited real-world data, ensuring the robustness of detection systems under various conditions (Xiong et al., 2024). Moreover, the integration of augmented reality (AR) into field visualization allows for enhanced interaction with real-time data, assisting farmers in decision-making processes (Loss et al., 2009). Finally, the overall research objects, including crop health monitoring, fruit and vegetable detection, and harvesting automation, rely on these integrated technologies to achieve greater accuracy and efficiency in modern agriculture. As these systems continue to evolve, an enhancement in productivity and sustainability across various agricultural practices is expected, aligning with broader trends toward smart farming (Raffik et al., 2024). The continued development of these technologies will likely lead to a more autonomous, data-driven agricultural ecosystem, where machines and AI systems work in tandem to improve agricultural outputs while minimizing resource use.

Recent advancements in the detection and classification of agricultural products have made significant strides, especially in fruit and bean inspection (Table 7). Researchers have leveraged convolutional neural networks (CNNs) to enhance real-time detection capabilities for various agricultural applications. A novel approach by integrating multi-scale feature fusion and lightweight networks has shown high-precision results in apple segmentation and defect detection (Kang and Chen, 2019). Further optimization was achieved through the combination of generative adversarial networks (GANs) and CNNs, demonstrating an improvement in defect detection accuracy for coffee beans (Chou et al., 2019). In another study, apple classification was enhanced by utilizing spatial pyramid matching (SPM) combined with support vector machines (SVM), achieving a classification accuracy of 98.15% (Ismail et al., 2018). Additional improvements in multi-class fruit detection utilized region-based convolutional neural networks, leading to a miss rate as low as 3.77% (Kuang et al., 2018). Moreover, the design of a shallow CNN structure for apple classification allowed for reduced parameter counts while maintaining competitive accuracy rates (Li et al., 2020). These efforts indicate a promising future for applying deep learning techniques to precision agriculture, improving both efficiency and accuracy in real-time monitoring and inspection systems (Zhang et al., 2020).

The application of machine learning, particularly deep learning models, has significantly advanced the detection of crop diseases and pests, offering more efficient and accurate methods compared to traditional agricultural monitoring practices. Over recent years, Convolutional Neural Networks (CNNs) have become the go-to tool for image-based recognition tasks in agriculture, such as disease diagnosis, pest detection, and even crop ripeness classification (Bao et al., 2024). These models can analyze vast amounts of visual data captured from various sources, including drones (Ab Rahman et al., 2019), satellites (Jumaat et al., 2018), and ground-level sensors (Loss et al., 2009), enabling the identification of diseases and pests in real-time with high accuracy.

The main advantage of machine learning techniques in this context is their ability to handle complex and high-dimensional data, enabling the detection of subtle patterns that human experts might miss. For example, deep learning models trained on large datasets of crop images can identify specific symptoms of diseases or pest infestations even under challenging field conditions, such as varying lighting or occlusions. This capability is particularly important for tasks like early disease detection, where timely intervention can prevent significant crop loss. Moreover, these models can be continually improved and adapted to new diseases and pest species, ensuring their long-term utility in agriculture.

However, there are still challenges in deploying these systems on a wide scale. One issue is the variability of environmental conditions, which can affect the performance of machine learning models trained on specific datasets. For instance, models that perform well in one geographical region may struggle when applied to crops grown in different climates or under different soil conditions. In addition, data quality remains a concern; high-quality annotated datasets are essential for training robust models, but such datasets are often limited in agricultural contexts. To address these limitations, recent research has focused on improving model generalization and developing strategies for training models with fewer labeled examples, such as transfer learning and semi-supervised learning. These efforts are crucial for expanding the applicability of AI-based crop monitoring systems across different farming environments, pushing the boundaries of what is possible in agricultural disease and pest management.

The integration of remote sensing technologies with artificial intelligence (AI) represents a transformative shift in the monitoring and management of agricultural systems (Alahi et al., 2023). Remote sensing tools, including unmanned aerial vehicles (UAVs) (Deng et al., 2022), satellites (Jumaat et al., 2018), and various sensor technologies (Kang and Chen, 2019), provide high-resolution (Jumaat et al., 2018), multispectral data that (Hao et al., 2025), when combined with AI, can offer unprecedented insights into crop health and environmental conditions (De Angelis et al., 2023). The ability to capture detailed images and sensor data from large-scale agricultural fields allows for continuous, real-time monitoring of crops, which was previously not possible with traditional manual methods (Guo et al., 2023).

Machine learning algorithms, especially deep learning models, are able to process and analyze the vast amounts of data generated by remote sensing platforms. For example, deep convolutional networks can classify crop types, detect diseases, monitor growth stages, and even assess soil conditions from images captured by UAVs. This fusion of AI and remote sensing technology significantly enhances the accuracy and efficiency of crop management practices, enabling farmers to make data-driven decisions that improve yields and reduce resource waste. Furthermore, AI-powered systems can detect patterns and correlations in the data that would otherwise be difficult to uncover, such as subtle changes in crop conditions that might indicate the onset of disease or pest infestation.

Despite the significant potential of these integrated systems, challenges remain. One of the key issues is the complexity of managing and processing the massive datasets generated by remote sensing technologies. These datasets require sophisticated data storage solutions, fast processing power, and highly advanced algorithms for timely analysis. Additionally, the environmental variability between regions—such as differences in climate, topography, and soil conditions—means that AI models often need to be retrained or fine-tuned to perform well in different settings. To address these issues, research is moving toward creating more adaptable and scalable systems that can be customized for different types of crops and farming conditions. Additionally, there is a growing need for models that can handle multi-source data, integrating information from satellite imagery, UAVs, soil sensors, and weather data to provide a more holistic view of crop health and field conditions.

While AI has the potential to revolutionize agriculture, its widespread adoption faces several significant challenges (Abdar et al., 2021a). A critical hurdle is the need for large, high-quality datasets to train machine learning models (Zhang et al., 2020). In agriculture, collecting sufficient labeled data can be particularly challenging due to the diverse and dynamic nature of farming environments (Xiong et al., 2024). Unlike other industries, where datasets can be standardized, agricultural data varies greatly depending on the crop type (Abdali et al., 2024), climate (Abbass et al., 2022), geographical location (Comber et al., 2023), and even specific farm practices (Sa et al., 2018). Furthermore, manual annotation of agricultural data—such as labeling images of crops, diseases, or pests—can be labor-intensive and expensive (Ferreira et al., 2019), making it difficult to create comprehensive datasets that capture all the potential variations in agricultural scenarios.

One challenge is the adaptability and scalability of AI models, particularly in the context of climate change. Models trained on specific conditions may struggle when applied to diverse environments. Barriers such as limited infrastructure and policy support in developing countries hinder technology deployment. However, government incentives and public-private partnerships can accelerate adoption, and current research focuses on developing adaptable models that require less retraining across various conditions. Another challenge lies in the adaptability and scalability of AI models. While deep learning models have demonstrated impressive performance in controlled conditions, their deployment in real-world agricultural environments can be less straightforward. Machine learning models trained on specific crops or conditions may struggle when applied to different environments, leading to issues with model generalization. This is particularly relevant in the context of climate change, which is altering growing conditions across the globe. Developing models that can adapt to a wide range of environmental conditions—without requiring extensive retraining—is a major focus of current research.

Looking to the future, the combination of AI with other emerging technologies, such as robotics and the Internet of Things (IoT), offers exciting possibilities for creating fully autonomous, data-driven agricultural systems. AI-powered robots, for example, could automate tasks like planting, harvesting, and pest control, while IoT devices could monitor environmental factors such as soil moisture, temperature, and nutrient levels in real-time. The integration of these technologies would allow for the creation of smart farming systems that can make real-time decisions based on data, significantly improving efficiency, reducing waste, and optimizing resource usage. However, to fully realize this vision, further advances in hardware, data processing, and system integration are necessary. AI systems must become more robust, scalable, and adaptable to handle the complexities of agricultural environments and meet the diverse needs of farmers worldwide.

In conclusion, while AI and machine learning technologies hold immense potential for transforming agricultural practices, addressing the challenges of data quality, model adaptability, and system integration will be key to their widespread adoption. As research continues to evolve, the convergence of AI, remote sensing, robotics, and IoT will likely usher in a new era of precision agriculture that improves both the productivity and sustainability of farming operations on a global scale.