Research On Defective Recognition And Sorting Of Medical Band-aids Based On YOLO Framework

With the increasing social labor cost and automation,the research on automatic sorting of defective medical band-aids has important theoretical value and practical significance.Through combing and analyzing the relevant literature,automatic sorting of defective band-aids has following difficulties: the first is difficult to identify and locate medical band-aids effectively;the second is to classify defective medical band-aids effectively.This study combines improved YOLO-U network with secondary classification algorithm to solve difficulties and realize automatic sorting through experiments.The specific research works are as follows:First,for the problem that it is difficult to identify and locate band-aids accurately.This study takes the YOLO framework as the skeleton,proposes the YOLO-U network which is based on the YOLO v2-tiny network,and returns the position,category and confidence of the medical band-aids by YOLO-U network.In order to improve the recognition rate and locate the position of medical band-aids accurately,the YOLO-U network adopts the secondary feature extraction method to obtain more detailed features.That is,changing the architecture of YOLO v2-tiny network,adding multi-layer convolution layers to the original network with 3*3 size convolution core,and using batch normalization method to standardize the model in each convolution layer.The experimental results show that the error(Loss)of identification and location by YOLO-U network is lower than the YOLO v2-tiny network;the fitness(IOU)between the prediction box of YOLO-U network and the real box is higher than the YOLO v2-tiny network;the real-time performance of the YOLO-U network reached 19.0264 ms in the training set,which is 2.2236 ms lower than the YOLO v2-tiny network;the YOLO-U network achieves recognition rate of 98.3% in 1000 test samples,which is higher than the recognition rate of the YOLO v2-tiny network and the Faster RCNN network.Second,aiming at the problem that it is difficult to classify defective medical band-aids effectively,a secondary classification algorithm is proposed.The algorithm solves the problem of core leakage,core tilt and core contamination.A variable weight channel algorithm is proposed for the problem of drug core leakage.The algorithm changes the channel ratio when calculating the correlation,sets the threshold,and each sample is classified according to the confidence level.For the problem of core tilt and core contamination,a method of skew correction recognition and contour analysis based on mathematical knowledge of topology is proposed.The method uses the improved Hough transform principle to correct the sample,and then uses the contour algorithm based on topological analysis to return the coordinate points of drug core and obtain the core area.Samples are classified into the second core tilt and core contamination according to the threshold of contour tilt and core region.The experimental results show that the quadratic classification algorithm has higher classification accuracy than SVM algorithm,and the accuracy of classification in test samples is 94.67%.Third,in order to verify the accuracy of the algorithm and the practicability of automatic sorting of defective medical band-aids,this paper designs the overall scheme of sorting band-aids.The scheme includes the mechanical selection,simplification of sorting robot structure,positive motion solution,inverse motion solution,hardware selection,vision system calibration and sorting experiment.The calibration of robot vision system mainly determines the relative position of camera,sorting robot and medical band-aids.The sorting experiment uses three indicators of true error rate,false error rate and error rate to analyze the accuracy of sorting defective medical band-aids.The experimental data shows that the error rate is 6.3% when the speed of production line is 0.0167 m/s after fusing YOLO-U network and the secondary classification algorithm.