Smart Rating System Of Steel Scrap Based On Improved YOLOv3 Algorithm

With the rapid development of science and technology,people’s lives are becoming more and more intelligent.Industrial intelligence will become the general trend,China’s 13th five-year development plan clearly proposes to promote the application of artificial intelligence in key economic and social fields.The Intelligent Grading of scrap grade can realize the whole process monitoring of scrap recovery based on order.In the process of scrap recovery,due to the influence of weather conditions,light intensity and the inevitable occlusion between scrap and scrap in the process of discharging,it is difficult to achieve accurate positioning and classification of scrap objects in the image.Convolution neural network uses convolution operation to extract the features of the image to be detected.This paper mainly studies the series of Yolo algorithms,analyzes and compares the differences among the three algorithms:YOLOv1,YOLOv2 and YOLOv3,optimize the network structure of yolov3,visualization of scrap grade evaluation results,to realize the accurate identification and determination of scrap in the process of scrap recovery.The main research contents are as follows:Firstly,the object detection algorithm based on deep learning is studied,the advantages and disadvantages of various algorithms are analyzed.In this process,the working principle of convolutional neural network is analyzed,and the mechanism of feature extraction of convolutional neural network is understood.According to the practical problems of scrap grade evaluation:the detection of scrap steel should have good timeliness,in the case of occlusion,the object should be identified as accurately as possible,we choose yolov3 algorithm based on convolutional neural network,which is a regression based algorithm,it can directly determine the type of scrap in the input image,and does not need too long time in the detection process,according to the actual situation of scrap recovery,the network structure of yolov3 is improved,the image atomization convergence algorithm is optimized,and the non-maximum suppression algorithm is improved.With the help of VOC data set,the performance of yolov2,yolov3 and improved yolov3 algorithm is compared,the improved yolo3 algorithm can meet the requirements of scrap grade evaluation.Then,create the scrap data set,the images in the scrap data set come from the monitoring snapshot of the scrap unloading site,the scrap photos provided by scrap companies and some scrap images retrieved from the Internet.For the marking of scrap image,open source LabelImg software is selected.In order to improve the quality of the photos collected on site,all the input images to be inspected are processed by image noise reduction,due to the limited scale of the collected scrap image data set,we expand the scale of the scrap data set by means of horizontal flip,fixed angle rotation and multiple image superposition.In order to provide intelligent auxiliary reference service for scrap recovery enterprises and scrap providers,we use Microsoft Visual Studio 2017 programming environment and OpenCV open source library to design a concise scrap grade evaluation system.Finally,the results of scrap grade evaluation are visualized.In order to provide intelligent auxiliary reference service for scrap recovery enterprises and scrap providers,we use Microsoft Visual Studio 2017 programming environment and OpenCV open source library to design a concise scrap grade evaluation system.This system mainly includes image data import module,image preprocessing module,scrap grade evaluation module and detection information output module,it can detect the scrap in the input image,and visualize the detection results in the scrap grade evaluation system,which is convenient for users to query the scrap detection results in real time.