Tracking and Counting of Tomato at Different Growth Period Using an Improving YOLO-Deepsort Network for Inspection Robot

To realize tomato growth period monitoring and yield prediction of cultivation, our study proposes a visual object tracking network called YOLO-deepsort to identify count tomatoes in different periods. Based on the YOLOv5s model, model uses shufflenetv2, combined with CBAM attention mechanism, compress size from algorithm level. In neck part network, BiFPN multi-scale fusion structure is used improve accuracy network. When target detection completes bounding box target, Kalman filter predict target’s location next frame, which tracker this paper. Finally, calculate error between predicted output by update parameters repeat above steps achieve fruits flowers. After getting results, we use OpenCV create virtual line targets. Our achieved competitive result based methods: The mean average precision flower, green tomato, red was 93.1%, 96.4%, 97.9%. Moreover, demonstrate ability counting process Overall, could fulfill actual requirements forecast greenhouse scene, provide theoretical support for crop status forecast.