Application of Improved SSD Algorithm in Surface Defect Detection of Workpieces
A MobileNetV3-SSD object detection algorithm is proposed to efficiently detect surface defects in workpieces.This algorithm improves the skeleton network of SSD by using MobileNetV3-Large,effec-tively reducing the network parameters and computational complexity.By combining a bottom-up feature pyramid network and an improved algorithm,the network is able to capture target information of different scales and shapes.Furthermore,the algorithm incorporates semantic segmentation as an auxiliary task and fuses multi-level features to further enhance the detection performance.The method is validated using the PASCAL VOC dataset and the Kolektor industrial defect dataset,achieving an mAP of 77.3%on the PAS-CAL VOC dataset,which is a 0.4%improvement over the SSD algorithm.The FPS on the Kolektor dataset reaches 105,which is a 112%improvement over the SSD algorithm.These results validate that the Mobile-NetV3-SSD algorithm outperforms the traditional SSD network in terms of detection accuracy and speed.
NETL
NSTL
万方数据
