Surface defect detection of cherry image based on YOLOv5 algorithm

Aiming at the problem of a wide variety of defects and complex defect morphology during the detection of surface defects of cherry images, this paper proposes a parameter optimized YOLOv5 algorithm, which optimizes and trains the learning rate, warm-up learning times, connection weights and other parameters of the YOLOv5 algorithm on the basis of feature enhancement and conversion label format of the cherry image dataset, and optimizes the important parameters during training to make the surface defect detection effect of cherry images better, and uses accuracy, recall, average accuracy, The confusion matrix evaluation criterion evaluates the defect detection effect, and verifies the detection rate of cherry surface defects in the image through detection experiments.