An improved pear recognition and localization algorithm based on yolov5 model

Abstract

Improving the visual recognition and positioning accuracy in the outdoor environment is an important way to improve the picking efficiency of fruit picking robots. With the rapid development of artificial intelligence, the convolutional neural network algorithm has gradually become an important research direction for machine recognition and localization. It can automatically extract target features, with high recognition accuracy, high speed and strong robustness. This paper takes pears as the research object, and proposes an improved pear recognition and localization algorithm based on the yolov5 model. The generalization ability of the model is improved by preprocessing and data enhancement of the data set, and an improved k-means clustering algorithm is proposed to realize the optimal calculation of the initial anchor frame. Compared with the original yolov5 model, the fitness and best recall rate of the improved algorithm in recognizing pears are increased by 6% and 9%, respectively.