Defect identification and quantitative analysis method based on improved YOLO V5

The YOLO V5 algorithm was used to study the identification method of carbon fiber composite material preset defects. In order to improve the detection accuracy while ensuring the detection efficiency, the original algorithm was improved by adding channel attention mechanism and spatial attention module, using k-means++ to re-cluster the prior box and optimizing the loss function. Using the improved network training defect data set, the number of images processed per second reached more than 12 frames, the average accuracy reached 98. 8%, and the recall rate was 98. 1%. Compared with other algorithms, the detection accuracy and speed of this algorithm have been improved to a certain extent which can meet the real-time and accuracy requirements.