Image-Based Phenotyping and Computer Vision in Agriculture

Phenotyping has long been recognized as a major bottleneck in crop improvement, limiting the effective translation of genomic advances into genetic gain. Traditional phenotyping methods relied heavily on manual measurements and visual scoring, which were labor intensive, and time-consuming. The integration of image-based phenotyping with computer vision (CV), machine learning (ML), and deep learning (DL) technologies has enabled high-throughput, non-destructive, objective, and scalable trait assessment across both the controlled and field environments. This chapter provides a comprehensive overview of imaging platforms including greenhouse-based systems, unmanned aerial vehicles (UAVs), tractor-mounted sensors, and ground phenocarts and imaging modalities such as RGB, hyperspectral, thermal, and LiDAR-based 3D systems. It describes the evolution of computational approaches from classical image processing to modern deep learning architectures such as Faster R-CNN, U-Net, and YOLO. Crop-specific applications in rice, wheat, maize, sorghum, soybean, barley, potato, tomato, cotton, and grapevine are discussed, highlighting advances in stress detection, yield prediction, organ counting, disease diagnosis, and structural trait extraction. The chapter further outlines standard phenotyping workflows, advantages for breeding programs, limitations, and emerging trends such as multimodal data fusion and AI-driven digital breeding systems.