Real-time caterpillar detection and tracking in Jujube orchard with YOLO NAS and SORT

Caterpillars pose a significant threat to agriculture, devouring crop foliage and evading traditional pest control methods like sticky or pheromone traps. While chemical pesticides are effective, concerns arise over crop residues. This study aims to address these challenges by tracking and estimating caterpillar positions in orchards in real-time, leveraging the Intel Realsense D405 RGB-D camera. Training data comprises 2,000 images from a jujube orchard, capturing diverse conditions such as exposure, occlusion, and wind. Real-time inference yields promising results, even recognizing the smallest 2-cm caterpillar at 21x12 pixels from a distance of 35 cm. The transition from YOLOv7 to YOLO NAS and from DeepSORT to SORT enhances detection by 30%, surpassing 95% accuracy. This innovative approach not only offers improved pest detection but also holds promise for integration with various technologies. From employing robot arms for targeted caterpillar removal to implementing laser pest targeting, this breakthrough contributes significantly to sustainable agriculture. By addressing the critical need for effective and environmentally friendly pest control practices, it helps ensure the long-term viability of agricultural systems.