Research on Appearance Detection Based on Improved YOLOv Network
Appearance detection involves the accurate and efficient recognition and positioning of objects in images or videos.In order to solve the small-sized target detection on the surface of objects,the paper optimizes the YOLOv3 network model,introduces the multi-scale detection and depth separable convolution technology to improve the detection accuracy and efficiency of the model to enhance the recognition ability of small-sized targets,and then the depth separable convolution technology is used to reduce the a-mount of calculation and improve the training effect of the model.The experimental results show that the research model has achieved a significant improvement in the detection of small-scale objects on the surface.Compared with other metal surface damage detection algorithms,the optimized YOLOv3 achieves a detection accuracy of 71.52％,surpassing that of the Faster R-CNN by 6.83％.Al-though the Faster R-CNN is excellent in accuracy but slow in speed,the seed of SSD is faster but not as good as that of YOLOv2.While the speed of YOLOv2 is fast but low in accuracy.Compared with the accuracy of the original model,the average accuracy of the research algorithm has increased by 7.77％,reaching 79.21％.Although the increase in network depth slightly increases the amount of calculations and slightly reduces the detection rate,but the introduction of depth-separable convolution,the detection speed reaches 36.2 fps,which is only slightly lower than the original model by 2.4 fps.The optimized algorithm improves the accuracy and robust-ness of small-sized object detection,and promotes its wide application in the field of computer vision.
万方数据
维普
