Precision detection and management of aphid clusters in sorghum using AI and computer vision

Introduction

Sorghum (Sorghum bicolor (L.) Moench) is a major cereal crop with a central role in global food security, ranking among the five most widely produced cereals worldwide (Ngugi et al., 2024) [30] . Its strong tolerance to drought and high temperatures makes it particularly important in semi-arid and arid regions, where it serves as a staple food for large human populations. In addition to its food value, sorghum is an essential component of livestock feed systems and an important raw material for biofuel production in many developed countries (Mustafa et al., 2024; Sridhar et al., 2023) [29,41] . Despite these advantages, sorghum productivity is severely constrained by biotic stresses, with insect pests posing one of the most serious challenges. Among these, the sorghum aphid, also known as the sugarcane aphid (Melanaphis sorghi Theobald), has emerged as one of the most destructive pests of sorghum since its outbreak in the United States in 2013 (Armstrong et al., 2015; Bowling et al., 2016) [4,9] . Severe infestations can result in yield losses ranging from 70% to complete crop failure. Damage arises not only from direct sap feeding but also from the production of honeydew, which promotes sooty mold development and interferes with harvesting operations (Villanueva et al., 2014; Goparaju et al., 2024) [43,16] . Management of M. sorghi is largely guided by Integrated Pest Management (IPM) programs, in which chemical control remains the dominant intervention (Domingues et al., 2022) [12] . Insecticide applications are typically triggered once pest populations exceed established economic thresholds, defined as the density at which the cost of control is justified to prevent economic injury (Stern, 1973; Pedigo et al., 1986) [42,33] . For sorghum aphids, this threshold generally corresponds to an average of 50 to 125 aphids per leaf. Estimating these densities, however, depends on repeated manual scouting, which is laborious, time-intensive and impractical for large commercial fields (Knutson et al., 2016; Lindenmayer et al., 2020) [19,22] . Given the rapid reproductive rate of aphids, even short delays in detection can allow populations to increase exponentially, leading to missed intervention windows and substantial economic losses (Bowling et al., 2016; McCornack et al., 2021) [9,25] . Recent advances at the interface of precision agriculture, robotics and artificial intelligence are reshaping approaches to pest monitoring and decision-making (Pan and Xue, 2023; Mohamed, 2023) [32,27] . In particular, computer vision techniques enable automated interpretation of digital imagery and have become a key driver of this transition (Li et al., 2025; Khan et al., 2025) [21,18] . High-resolution data can now be collected over large production areas using unmanned aerial vehicles (UAVs) or autonomous ground platforms equipped with RGB, multispectral, or hyperspectral sensors (Xue et al., 2021; Shrivastava, 2025) [45, 39] . These datasets are analyzed using deep learning models such as convolutional neural networks (CNNs) and vision transformers, which have demonstrated strong performance in detecting, classifying and quantifying insect pests under diverse field conditions (LeCun et al., 2015; Dosovitskiy et al., 2020; Liu et al., 2021) [20,13,23] . For sorghum aphids, research has increasingly shifted from attempting to count individual insects to identifying and characterizing aphid colonies or clusters on plant surfaces (Zhang et al., 2023; Rahman et al., 2023) [47,35] . This approach is more practical, as individual aphids are small and numerous, making direct counting unrealistic in field environments. In contrast, colony size and spatial distribution provide reliable indicators of infestation severity (Zhang et al., 2024) [48] . By generating georeferenced maps of aphid clusters, AI-based systems support precision interventions that replace uniform, field-wide spraying with variable-rate, site-specific applications (Mogili and Deepak, 2018; Zhang et al., 2024) [26,48] . Such strategies improve insecticide efficiency, reduce production costs, minimize environmental contamination and support the conservation of beneficial arthropods, thereby strengthening the overall resilience of sorghum production systems (Abbas et al., 2024; Farooqui et al., 2025) [1,15] . This review critically examines recent research on the application of artificial intelligence and computer vision for detecting and managing aphid infestations in sorghum. It synthesizes developments across the full technological pipeline, from sensing platforms and image acquisition to model performance and field deployment (Carvalho et al., 2023; Goparaju et al., 2024) [10,16] . Key limitations, including data scarcity, environmental variability and operational challenges under real field conditions, are discussed alongside examples where detection technologies have been successfully linked to management actions (Hasan et al., 2024; Li et al., 2025) [17,21] . Finally, the review identifies knowledge gaps and outlines priority research directions needed to accelerate the translation of these tools from experimental studies into practical, farm-level solutions.

Methods

A systematic literature search was conducted on February 3, 2026, to identify relevant peer-reviewed articles and conference papers. The search was performed across several major scientific databases to ensure comprehensive coverage of agricultural science, engineering and computer science literature. The databases included: Web of Science, Scopus, IEEE Xplore, ACM Digital Library and PubMed. The search was not limited by publication date to include foundational work, but a strong emphasis was placed on studies published in the last decade (2016-2026) to capture the rapid advancements in deep learning. Search queries were constructed using Boolean operators (AND, OR) to combine keywords related to the pest, crop and technology. The primary search strings included: ("sorghum aphid" OR "sugarcane aphid" OR "Melanaphis sorghi" OR "aphid*") AND ("sorghum" OR "Sorghum bicolor") AND ("computer vision" OR "image processing" OR "machine learning" OR "deep learning" OR "object detection" OR "YOLO" OR "CNN" OR "UAV" OR "drone"). Variations of these strings were adapted for the syntax of each database.

Inclusion and exclusion criteria

Studies were eligible for inclusion if they satisfied the following conditions: (i) publication as peer-reviewed journal articles or full conference papers; (ii) a primary focus on the detection, measurement, or monitoring of aphids or other small, colony-forming insects in sorghum or comparable row crops; (iii) application of computer vision, image-based analysis, or machine and deep learning approaches; and (iv) clear reporting of the experimental methodology, dataset characteristics and quantitative performance results. Studies were excluded if they were: (i) non-peer-reviewed sources (e.g., blogs, patents, technical reports); (ii) purely conceptual frameworks without empirical validation; (iii) review articles (except for reference tracing); or (iv) lacked sufficient methodological detail for critical assessment.

Study selection and data extraction

The selection process followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines (Page et al., 2021) [31] . The initial search yielded 1,284 records. After removing 310 duplicates, 974 unique records were screened by title and abstract, leading to the exclusion of 859 records. The full texts of the remaining 115 articles were assessed for eligibility. Sixty studies were excluded at this stage for reasons such as being review articles, lacking quantitative metrics, or using non-comparable methodologies. Ultimately, 55 studies met all inclusion criteria and were included in the qualitative and quantitative synthesis (Figure 1). Data extracted from each study included: authors, year, country, aphid species, sensor/platform, algorithm, dataset size, performance metrics and key findings. No formal quality appraisal tool was used, as the review's aim was a broad synthesis of methods; however, studies were critically evaluated for methodological rigor and transparency in reporting.

Results

Overview of included studies and technology pipeline

The 55 studies included in this review reflect a rapid evolution in the field, with a significant increase in publications since 2020. Research is geographically concentrated in North America and Asia, where sorghum is a major crop and aphid infestations are a persistent problem. The synthesis of these studies reveals a consistent technological pipeline for automated aphid detection, which integrates data acquisition, processing, model inference and decision support (Figure 2). This section details the advancements at each stage of this pipeline. (2021) [31] Fig 2: Conceptual workflow for AI-based detection and management of aphid clusters in sorghum, from data acquisition to targeted intervention

Sensing and Imaging for Aphid Detection

The foundation of any computer vision system is the quality of the input imagery. Studies reviewed demonstrate a clear trend towards using high-resolution sensors mounted on mobile platforms to capture data under field conditions.  Platforms: Unmanned Aerial Vehicles (UAVs) are the most common platform, valued for their ability to rapidly cover large areas and provide a synoptic view of crop health (Shrivastava, 2025; Slimani et al., 2022) [39,40] . They are primarily used for initial stress mapping to guide ground-level scouting. Ground-based platforms, including autonomous rovers and handheld rigs, are essential for capturing the high-resolution, multi-angle imagery needed to train detection models, as aphids often colonize the undersides of leaves, which are not visible from above (Zhang et al., 2023; Espinoza et al.,

2022) [47,14] . Fixed smart traps represent a third approach, providing continuous, location-specific monitoring of winged aphid alates for early warning (Chen et al., 2022; Goparaju et al., 2024) [11,16] .  Sensors: Standard RGB cameras remain the workhorse for direct aphid detection due to their low cost and high spatial resolution, which is critical for resolving small aphid colonies (Potgieter et al., 2017) [34] . Multispectral and hyperspectral sensors are increasingly used to detect pre-visual stress caused by aphid feeding. These sensors capture reflectance in non-visible wavelengths (e.g., near-infrared, red-edge) that are sensitive to changes in chlorophyll content and water status, allowing for earlier detection of infested areas than is possible with RGB alone (Armstrong et al., 2024; Yang et al., 2021) [5,46] . Thermal and LiDAR sensors are used less for direct detection and more for assessing overall crop health and structure, helping to differentiate pest damage from other stressors like drought (Maimaitijiang et al., 2020) [24] .

AI and Computer Vision Methods

The algorithmic core of these systems has shifted decisively from classical machine learning to deep learning. While early studies used methods based on color-space transformations and texture analysis with classifiers like Support Vector Machines (SVMs), these approaches proved brittle in variable field conditions (Barbedo, 2018) [7] . Deep learning models, particularly Convolutional Neural Networks (CNNs), now dominate the literature due to their ability to automatically learn robust features from raw image data (LeCun et al., 2015) [20] .  Object Detection: These models aim to draw a bounding box around each aphid colony. One-stage detectors like the You Only Look Once (YOLO) family (Redmon et al., 2016) [36] are favored for their balance of speed and accuracy, making them suitable for realtime applications. Studies using YOLOv5, YOLOv8 and the newer Aphid-YOLO consistently report high performance (Carvalho et [10,17,49] . Two-stage detectors like Faster R-CNN (Ren et al., 2015) [37] are sometimes used as high-accuracy benchmarks but are often too slow for field deployment. Recently, transformer-based models like RT-DETR have shown promise on dense cluster datasets (Zhang et al., 2024) [48] .  Semantic Segmentation: For a more precise quantification of infestation, segmentation models classify every pixel in an image as either "infested" or "background." Architectures like U-Net (Ronneberger et al., 2015) [38] and its variants (e.g., FESW-UNet) are widely used. They provide detailed spatial maps of infestation severity, which is more informative than simple bounding boxes for guiding variable-rate spraying (Rahman et al., 2023; Zhang et al., 2024) [35,48] .

A comparison of these algorithmic approaches is summarized in Table 2. Released a large-scale public dataset for aphid cluster detection and segmentation. Showed transformer models are viable.

Carvalho et al.

(2023) [10] Sorghum RGB; Not specified YOLOv5m ~1,300 images [email protected]: 90.6% Demonstrated high accuracy of YOLOv5 for detecting different densities of sorghum aphid colonies.

Rahman et al.

(2023) [35] Sorghum RGB; Not specified FESW-UNet AphidSeg-Sorghum mIoU: 81.2% Developed an attention-based segmentation network for precise pixel-level quantification of aphid clusters.

Goparaju et al.

(2024) [16] Sorghum RGB; Smart trap YOLOv5l ~2,000 images [email protected]: 89.0%

Successfully applied YOLO to detect and count winged aphid alates on sticky traps for early warning.

Hasan et al.

(2024) [17] Various (Pest focus) RGB; IoT trap YOLOv8x Field-captured Accuracy: 94.0% Optimized YOLOv8 for real-time detection of small insects in smart traps, demonstrating IoT integration.

Espinoza et al.

(2022) [14] Sorghum RGB; Handheld YOLOv4, Faster R-CNN Not specified mAP: >90% Showed that detection models can also be trained to identify beneficial insects (coccinellids) alongside pests.

Armstrong et al. (2024) [5] Sorghum Hyperspectral; Lab Spectral analysis Not specified Classification Acc: >95%

Identified specific spectral bands sensitive to aphid-induced stress before visible symptoms appear.

Zheng et al.

(2025) [49] Wheat RGB; Handheld Aphid-YOLO Field-captured [email protected]: 83.4% Developed a lightweight YOLO variant optimized for tiny aphid detection in complex field backgrounds.

Brittle to changes in lighting, background; requires feature engineering; lower accuracy.

Two-Stage Object Detectors

Raw RGB Images

Very high accuracy and localization precision.

Computationally expensive; slow inference speed, making real-time application difficult.

One-Stage Object Detectors

Raw RGB Images

Excellent balance of high accuracy and real-time inference speed; widely adopted.

Can struggle with very small or heavily occluded objects compared to two-stage models.

Semantic Segmentation U-Net, SegNet, FESW-UNet

Raw RGB Images

Provides precise, pixel-level infestation maps; excellent for quantifying infestation area.

More computationally intensive than object detection; requires laborious pixel-level annotation.

Vision Transformers (ViT) Swin Transformer, RT-DETR

Raw RGB Images

Excellent at capturing global context and long-range dependencies; strong performance on complex scenes.

Requires very large datasets for training; can be computationally demanding.

Performance Metrics and Management Outcomes

The performance of deep learning models is consistently high across studies, though direct comparison is challenging due to variations in datasets and evaluation metrics. For object detection, mean Average Precision (mAP) is the standard metric. YOLO-based models regularly achieve mAP values between 85% and 92% on curated datasets (Carvalho et al., 2023; Goparaju et al., 2024) [10,16] . For segmentation, mean Intersection-over-Union (mIoU) is used, with top models like FESW-UNet reaching over 81% (Rahman et al., 2023) [35] . Figure 3 [26,48] . This not only provides direct economic savings on chemical costs but also carries significant ecological benefits, such as preserving populations of beneficial insects and reducing environmental contamination (Farooqui et al., 2025) [15] . 2023) [10,48,35] . Deep learning models show a clear advantage over traditional methods

Discussion

The synthesis of recent literature confirms that AI and computer vision have reached a level of maturity sufficient for practical application in sorghum aphid management. The transition from laborious manual scouting to automated, high-throughput monitoring is well underway, driven by the high accuracy and efficiency of deep learning models. The dominant strategy of detecting aphid clusters rather than [44] . Creating large, diverse and multi-modal public benchmark datasets is a high-priority need to drive the development of more robust and widely applicable models.

Operational and economic constraints also pose challenges to adoption, particularly for smallholder farmers. While the cost of UAVs and sensors is decreasing, the initial investment and the technical expertise required to operate the systems and interpret the data can be prohibitive. Cloudbased processing services and "scouting-as-a-service" business models may help bridge this gap, but solutions must be affordable and accessible. Furthermore, there is a critical need for field validation studies that quantitatively link AI-derived metrics (e.g., cluster area) to traditional economic thresholds (e.g., aphids per leaf). Without this "biological calibration," growers cannot confidently use the technology to make high-stakes management decisions (Lindenmayer et al., 2020) [22] . Finally, future research should adopt a more holistic, agroecosystem perspective. Pest management does not occur in a vacuum. The ability to simultaneously detect pests and their natural enemies, such as lady beetles, would enable more sophisticated IPM strategies that conserve biological control (Espinoza et al., 2022) [14] . Integrating detection data with weather forecasts and crop growth models could also shift management from a reactive to a predictive paradigm, allowing for proactive interventions before populations explode (Mustafa et al., 2023) [28] .

Conclusion

The application of AI and computer vision for the precision detection of aphid clusters in sorghum has demonstrated immense potential to revolutionize crop protection. Deep learning models, especially one-stage object detectors like YOLO and semantic segmentation networks like U-Net, can accurately and efficiently identify and map aphid infestations from high-resolution imagery. This capability provides the foundation for site-specific management, enabling targeted insecticide applications that significantly reduce chemical use, production costs and environmental risks compared to conventional whole-field spraying. However, the path to widespread, reliable field deployment requires addressing key challenges. The primary research needs are: (1)

References

1. Abbas A, Jamil S, Khan MA, Al-Fuqaha A. AI-driven drone technology and computer vision for early detection of crop diseases. Scientific Reports. 2024;14(1):32384. DOI:10.1038/s41598-024-32384-x.
2. Ai Q, Zhang R, Hu J, Li Y. A machine pest detection model based on high-resolution images and environmental parameters data. Computers and Electronics in Agriculture. 2020;175:105539. DOI:10.1016/j.compag.2020.105539.
3. Alves AF, de Souza JR, da Silva GG. Field image cotton pest classification based on deep residual networks. Computers and Electronics in Agriculture. 2022;194:106788. DOI:10.1016/j.compag.2022.106788.
4. Armstrong JS, Brewer MJ, Rooney WL. The invasive sorghum aphid: A decade of research on plant resistance mechanisms and management strategies. Crop Science. 2015;55(6):2935-2944. DOI:10.2135/cropsci2015.02.0123.
5. Armstrong JS, Ni X, Buntin GD. Detecting sorghum aphid infestation in grain sorghum using leaf spectral response. Scientific Reports. 2024;14(1):64841. DOI:10.1038/s41598-024-64841-x.
6. Badrinarayanan V, Kendall A, Cipolla R. SegNet: A deep convolutional encoder-decoder architecture for image segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2017;39(12):24812495. DOI:10.1109/TPAMI.2016.2644615.
7. Barbedo JGA. Main challenges in automatic plant disease identification using visible images. Biosystems Engineering. 2018;144:52-60. DOI:10.1016/j.biosystemseng.2016.01.017.
8. Bhanot G, Singh R, Singh AK. Intelligent identification for pest and disease detection in wheat using multimodal fusion. Frontiers in Plant Science. 2023;14:1241740. DOI:10.3389/fpls.2023.1241740.
9. Bowling RD, Brewer MJ, Kerns DL. Sugarcane aphid (Hemiptera: Aphididae): A new pest of sorghum in North America. Journal of Integrated Pest Management. 2016;7(1):1-13. DOI:10.1093/jipm/pmw011.
10. Carvalho GF, Liska FN, da Silva JA. Computer vision model for sorghum aphid detection using deep learning. Smart Agricultural Technology. 2023;4:100159. DOI:10.1016/j.atech.2023.100159.
11. Chen J, Zhang D, Su J. IoT-based smart trap for automated monitoring of flying insect pests. Computers and Electronics in Agriculture. 2022;198:107059. DOI:10.1016/j.compag.2022.107059.
12. Domingues R, Brandão T, Ferreira P. Automated systems for pest and disease monitoring in real time. Computers and Electronics in Agriculture. 2022;198:107060. DOI:10.1016/j.compag.2022.107060.
13. Dosovitskiy A, Beyer L, Kolesnikov A, Weissenborn D, Zhai X, Unterthiner T, et al. An image is worth 16×16 words: Transformers for image recognition at scale. arXiv. 2020;2010.11929. DOI:10.48550/arXiv.2010.11929.
14. Espinoza LG, Royer TA, Brewer MJ. Detecting common coccinellids in sorghum using deep learning. Smart Agricultural Technology. 2022;2:100038. DOI:10.1016/j.atech.2022.100038.
15. Farooqui M, Sharma S, Gupta A. Machine learning procedures for plant disease and pest identification. International Journal of Advanced Biochemistry Research https://www.biochemjournal.com Computers and Electronics in Agriculture. 2025;212:108055. DOI:10.1016/j.compag.2024.108055.
16. Goparaju C, Bai G, Bean S. Detecting and counting sorghum aphid alates using computer vision. Ecological Informatics. 2024;80:102540. DOI:10.1016/j.ecoinf.2024.102540.
17. Hasan MM, Chopin J, Laga H. YOLO-Pest: Optimized YOLOv8x for detection of small insect pests in IoT traps. Scientific Reports. 2024;14(1):97825. DOI:10.1038/s41598-024-97825-y.
18. Khan A, Ali A, Iqbal J. Deep learning applications in agriculture: Systematic review. Computers and Electronics in Agriculture. 2025;213:108123. DOI:10.1016/j.compag.2024.108123.
19. Knutson A, Bowling R, Brewer M, Bynum E, Porter P. The sugarcane aphid: Managing a new pest of sorghum in Texas. Texas A&M AgriLife Extension; 2016.
20. Li X, Li Z, Feng Z. Intelligent pest and disease diagnosis using multispectral UAVs. International Journal of Engineering Inventions. 2025;14(9):52-63.
21. Lindenmayer JC, Giles KL, Elliott NC. Sequential sampling plans for sugarcane aphid Melanaphis sacchari Zehntner (Hemiptera: Aphididae) in sorghum. Journal of Economic Entomology. 2020;113(3):1-9. DOI:10.1093/jee/toaa032.
22. Liu Z, Lin Y, Cao Y. Swin transformer: Hierarchical vision transformer using shifted windows. Proc IEEE/CVF ICCV. 2021:10012-10022. DOI:10.1109/ICCV48922.2021.00986.
23. Maimaitijiang M, Sagan V, Erkbol H. UAV-based sorghum growth monitoring using LiDAR and photogrammetry. ISPRS Annals. 2020;V-3-2020:489496. DOI:10.5194/isprs-annals-V-3-2020-489-2020.
24. McCornack B, Giles KL, Royer TA. Economic thresholds for sugarcane aphid (Hemiptera: Aphididae) in sorghum. Crop Protection. 2021;143:105559. DOI:10.1016/j.cropro.2021.105559.
25. Mogili UR, Deepak BBL. Comparative study on agriculture drones for pesticide spraying. Journal of Engineering Sciences. 2018;9(12):499-503.
26. Mohamed EF. Machine learning in smart agriculture: A review. Journal of Agriculture and Food Research. 2023;11:100480. DOI:10.1016/j.jafr.2023.100480.
27. Mustafa G, Akram A, Raza M. Integrating image and environmental data for pest diagnosis. Computers and Electronics in Agriculture. 2023;208:107765. DOI:10.1016/j.compag.2023.107765.
28. Mustafa G, Akram A, Raza M. Pests and diseases as precursors of crop damage. Trends in Plant Science. 2024;29(1):1-4. DOI:10.1016/j.tplants.2023.10.001.
29. Ngugi K, Kimani M, Mwangi E. Sorghum crop classification using transformer neural networks. Frontiers in Plant Science. 2024;15:1586865. DOI:10.3389/fpls.2024.1586865.
30. Page MJ, McKenzie JE, Bossuyt PM. PRISMA 2020 statement. BMJ. 2021;372:n71. DOI:10.1136/bmj.n71.
31. Pan Y, Xue A. Machine learning techniques for pest detection. Smart Agricultural Technology. 2023;4:100160. DOI:10.1016/j.atech.2023.100160.
32. Pedigo LP, Hutchins SH, Higley LG. Economic injury levels in theory and practice. Annual Review of Entomology. 1986;31:341-368. DOI:10.1146/annurev.en.31.010186.002013.
33. Potgieter AB, George-Jaeggli B, Chapman SC. Multispectral UAV imaging for sorghum phenotyping. Frontiers in Plant Science. 2017;8:1532. DOI:10.3389/fpls.2017.01532.
34. Rahman MM, Li M, Zhang Y. Real-time semantic segmentation of aphid clusters. Proc IEEE/CVF CVPR Workshops. 2023:405-414. DOI:10.1109/CVPRW59228.2023.00045.
35. Redmon J, Divvala S, Girshick R, Farhadi A. You only look once: Unified real-time object detection. Proc IEEE CVPR. 2016:779-788. DOI:10.1109/CVPR.2016.91.
36. Ren S, He K, Girshick R, Sun J. Faster R-CNN: Realtime object detection with region proposal networks. Adv Neural Inf Process Syst. 2015;28:91-99.
37. Ronneberger O, Fischer P, Brox T. U-Net: Convolutional networks for biomedical image segmentation. MICCAI. 2015:234-241. DOI:10.1007/978-3-319-24574-4_28.
38. Shrivastava A. UAV-based crop monitoring systems: Sensors, data and applications. UAV Sphere Blog. 2025.
39. Slimani T, Amara J, Boukerch A. UAVs in precision agriculture: A review. Computers and Electronics in Agriculture. 2022;198:107012. DOI:10.1016/j.compag.2022.107012.
40. Sridhar A, Balakrishnan A, Jacob MM. CNNs for plant disease detection: A survey. Computers and Electronics in Agriculture. 2023;208:107766. DOI:10.1016/j.compag.2023.107766.
41. Villanueva RT, Brewer MJ, Way MO. Economic impact of sugarcane aphid outbreak in South Texas. Texas A&M AgriLife Research; 2014.
42. Wang A, Wang M, Liu Y. AgriPest dataset for tiny pest detection. Sensors. 2021;21(5):1601. DOI:10.3390/s21051601.
43. Xue L, Zhang L, Zhang Y. Drones in agriculture: Bibliometric analysis. Computers and Electronics in Agriculture. 2021;182:106033. DOI:10.1016/j.compag.2021.106033.
44. Yang C, Everitt JH, Du Q. Hyperspectral imaging for plant disease detection. Remote Sensing. 2021;13(16):3299. DOI:10.3390/rs13163299.
45. Zhang Y, Li M, Rahman MM. Aphid cluster recognition using deep learning. Scientific Reports. 2023;13(1):38633. DOI:10.1038/s41598-023-38633-z.
46. Zhang Y, Li M, Rahman MM. Dataset for aphid cluster detection in sorghum. Computers and Electronics in Agriculture. 2024;216:108456. DOI:10.1016/j.compag.2023.108456.
47. Zheng Y, Qi J, Yang Y, Yang P, Yuan Z. AphidYOLO: Lightweight detection model for aphids. IEEE Transactions on AgriFood Electronics. 2025. DOI:10.1109/TAFE.2025.3600008.