Key Technologies Research Of Multi-objective Simultaneous Classification Based On Deep Learning

Laboratories of universities and research institutes have strict requirements on the attire of experimenters.Those who do not dress according to the requirements affect the experimental results,or even cause safty accidents.To identify whether personnel have complied with the regulations,dress requires manual supervision,which will cause the waste of labor costs.Object detection is one of the research hotspots in the field of computer vision,and it is the “intelligence” of machine vision.With the continuous development of deep learning in recent years,the continuous deepening of convolutional neural network CNN research,especially the YOLO series in the field of object detection.YOLOv8 has come out since 2023.The method of deep learning is applied to real life,which is used to identify the wearing of masks and white coats of experimenters,and automatically identify whether the dress of personnel is standardized by object detection method,replacing manual supervision.A dataset containing 5250 images,as well as annotated masks and white coats,is established for training and testing in object detection.The training,validation,and testing sets are divided into 8:1:1parts.These dataset images contain various states of personnel wearing white coats and masks,and specifically,interference items such as white clothes and masks revealing their nose and mouth.In the same image data,it is necessary to simultaneously detect whether the face is wearing a mask and whether the body is wearing a white coat.The target detection network used also needs to synchronously classify and solve two classification problems.The research is based on YOLOv8 network and makes three improvement strategies for it.The improvement strategies include: 1.For the possibility of labeling low quality problem,the regression loss function CIoU is modified to WiseIoU;2.Because there are two target recognition tasks in the same image,in order to further refine the features extracted from different tasks and improve the accuracy of the training model,a self attention mechanism MHSA is added;3.In order to cope with unknown geometric structure transformations,the traditional convolution method is modified to deformable convolution v2.For the original YOLOv8 network and using each improvement strategy separately,as well as combining the first two strategies or three strategies,all trained using the same self-made dataset,experimental results were obtained.The final conclusion is that after using each strategy,selecting the value of F1 and mAP50 as the evaluation indicator and analyzing confusion matrix improves the accuracy compared to the original network.By using three improvement strategies simultaneously,the performance can be significantly increased.YOLOv8 is the latest network in the YOLO series,and there have been few scholars improving on it.The dissertation innovatively uses some improvement strategies and combines them to achieve good performance improvement.The final plan can effectively detect the wearing of masks and white coats,which has practical significance in engineering practice.