Abstract
The pre-extracted target object is always distorted due to many parts spots, surface reflection and other factors in traditional visual inspection. Therefore, a detection process based on improved YOLOv5 is proposed. In the early stage, for the optimization of light source, the circular light is generally used to successfully extract the region of interest, so as to reduce the scratches caused by image pretreatment, resulting in detection omission and distortion. Adding the ACON (All Class in One Network, ACON) activation function can improve the network feature extraction ability and the accuracy of the detection algorithm. Bounding box loss, means of target detection loss and classification loss are all significantly improved by the improved algorithm, far exceeding the performance of traditional detection methods, and all have a certain degree of reduction.
Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
