Research on Fruit Recognition Method Based on Improved YOLOv4 Algorithm

Aiming at the problems of low recognition rate, large storage space and slow detection speed of traditional image recognition algorithms and YOLOv4 algorithm, a fruit recognition target detection method based on improved YOLOv4 algorithm was proposed. The attention mechanism SE module is introduced into CSPResblock module. SPP module is improved to use pool core with different aspect ratio. The PANet structure was improved, input information was spliced at the corresponding feature level, and the three-layer output of the trunk network was mainly fused to improve the detection ability of fruit targets. The data set includes common fruits: bananas, grapes, apples, pears, mangoes, peaches, avocados and bananas. Each type of fruit is 1000 pieces, and the image size is uniformly processed into 608 pixels *608 pixels, a total of 8000 pieces. The results show that the MAP of the proposed method is 98.02%, the accuracy is 95.62%, and the frames per second (FPS) is 28.56. MAP improved by 16.6% compared to YOLOv4. The research can meet the requirements of high detection accuracy and detection speed, and has important reference value for the improvement of fruit recognition accuracy.