A study on rapid recognition of ripe tomato fruits in complex environment based on YOLOv5s

In recent years, with the tomato market demand continues to expand, China's tomato cultivation area increases year by year, its consumption of manpower, material and financial resources are also expanding, especially in the manpower link is in the tomato picking this link in the consumption of too much, and nowadays the aging population is serious, the lack of labor force to go to the picking, in order to solve the problem, this paper is based on the YOLOv5s model for the rapid recognition of ripe tomato fruit in a complex environment. In order to solve this problem, based on YOLOv5s model, this paper carries out an in-depth study on the rapid recognition of ripe tomato fruits in complex environments, so as to deploy the control module on the picking robot at a later stage. Firstly, fruit image datasets of tomatoes with different maturity (green and red) are collected, then labeled using labeling for deep neural network model training, and finally validated on a test set. The experimental results show that Mature tomatoes and Immature tomatoes precision rate is 97.1% and 98%, recall rate is 94.8% and 95.9%, and the average precision mean is 97.6% and 98.5%, respectively. The results demonstrate that the target detection algorithm can better meet the detection requirements of real picking tomatoes under the complex situations of fruit superposition, background interference, different densities, and foliage occlusion.