Research on micro-defect detection algorithm based on improved Faster R-CNN
Small target detection is a difficult problem in defect detection,with problems such as poor detection effect,slow speed and low positioning precision. Faster R-CNN model that fuses multi-scale information is proposed aiming at the problems of few feature information and low recognition accuracy of micro-defects feature on the surface of lithium battery tapes. Firstly,the residual network(ResNet)—50 and recursive feature pyramid are used as the feature extraction network to improve the feature extraction capability for micro-defects. Secondly,the error introduced by two quantization rounding in the original network localization is eliminated by the region of interest(ROI)calibration module to improve the defect localization precision. Finally,the loss function is optimized to solve the imbalance problem of learning ability between difficult and easy samples of the model. The improved Faster R-CNN model achieves an average precision of 97.9％ for the detection of micro-defects on the surface of lithium battery tapes,which is increased 2.1％ than the original Faster R-CNN.
万方数据
维普
