Design a Robot System for Tomato Picking Based on YOLO v5

Abstract

Tomato picking robots can save labor and improve production efficiency, which is of great significance for facility tomato planting. The picking mode of the Tomato Picking Robot has an important impact on fruit picking quality and efficiency. At present, in the process of fruit picking, the rough recognition of fruit positioning needs compensation accuracy. The picking mode is usually to capture the fruit in a large range and then separate the fruit peduncles in a specific position. In the process of grasping and pulling the fruit, it may cause damage to the fruit and plant, which will reduce the fruit quality. Moreover, The retention length of peduncles is also difficult to control, resulting in difficulties in transportation and storage. Therefore, the prediction, location and segmentation of separation points on tomato images are an important guarantee for efficient and lossless harvest. In this study considering the changeable conditions such as light change and branches and leaves interference, YOLO V5 is used to confirm the position relationship between tomatoes and peduncles. The region of interest for peduncles picking is reduced according to the growth characteristics of the fruit. Then, taking the boundary boxes center of the peduncles as the picking points, the corresponding depth information is obtained and the robot is controlled to complete the picking task. The experimental results show that this method can recognize and locate tomato picking points under complex near-color backgrounds. The average recognition time of a single frame image is 104 ms, which meets the real-time requirements of automatic picking. Compared with the SSD algorithm, it has obvious advantages.