Pepper Target Recognition and Detection Based on Improved YOLO v4

Abstract

In order to improve visual recognition accuracy of pepper and provide reliable technical support for agricultural production, an improved YOLOv4 algorithm for pepper target recognition and detection was proposed in this paper. By adding Mosaic data enhancement and CBAM (Conventional block attention module) attention mechanism to the primitive character extraction network, the method enhanced the learning ability of the target detection algorithm, made the network effectively suppress the interference features, and increased the attention to effective features. To improve the accuracy of identification. The improved network model was trained, verified and tested on the self-made data set. The results showed that the proposed algorithm could effectively improve the accuracy of pepper recognition under natural light, and finally improved the mean Average Precision (mAP) of the existing YOLOv4 algorithm from 88.95% to 98.36%.