Research On Safety Detection System Of Rolling Equipment Based On Deep Learning

In the rolling industry,the accident that the machine stabs the human hand due to the careless operation of the worker often occurs.In response to such production safety accidents,this topic studies a method to quickly identify whether the worker's hand enters the dangerous area through a visual algorithm,and proposes a method.An improved lightweight hand detection algorithm MF-SSD based on SSD.The improved algorithm overcomes the shortcomings of the current algorithm and significantly improves the detection performance of the model on the hand.Secondly,based on the Jeston Xavier NX platform,this thesis develops a safety detection system for rolling equipment,and applies the improved MF-SSD algorithm to the embedded platform to realize the intelligent safety control of the equipment,so as to prevent the occurrence of accidents caused by machine stab wounds.Reduce labor risks and economic losses of enterprises.The main contents of this thesis are as follows:(1)The industrial demand in the rolling scene is analyzed,the key problems of this research are determined,the rolling equipment safety detection system based on the deep learning method is designed,and the hand data set suitable for the rolling scene is constructed.Collect the video of workers operating in the rolling production of the enterprise,and use Open CV to preprocess it to obtain 1115 data pictures with better quality.In order to expand the number of image samples and improve the richness of the dataset samples,the dataset was expanded to 4460 images by offline data enhancement,and the enhanced human hand dataset was annotated with the Label Img professional annotation tool.The VOC dataset on which the detection model was trained.(2)The MF-SSD hand detection algorithm is designed to achieve fast and accurate detection of workers' hands.Based on the SSD algorithm,the lightweight neural network Mobile Net is used to improve VGG16 as the base network of SSD,which reduces the computational and parametric quantities of the model and redesigns six feature layers for detecting hand targets at different scales;the structure of the Mobile Net network is fine-tuned,and the multilayer feature fusion module is designed with reference to the feature pyramid network to improve the The K-Means algorithm is used to cluster the real hand frame aspect ratio in the dataset,and the model a priori frame size is adjusted according to the clustering result to make the prediction of hand frame more accurate,and finally the migration learning strategy is introduced to accelerate the training and convergence of the model.(3)The algorithm is verified by the human hand dataset constructed in this thesis.Using the training methods of online data enhancement and transfer learning to compare the improved MF-SSD algorithm with eight algorithms such as SSD,FSSD,Mobile Net?SSD,Faster-RCNN,YOLOv3,YOLOv4,Retinanet and RFBnet,the average MF-SSD algorithm The detection accuracy rate is as high as 99.15%,the detection speed reaches 35 FPS,and the model size is only 25.7MB.The comprehensive detection performance is the best.Finally,the improvement points proposed in this thesis are verified and analyzed through ablation experiments.The experimental results further show the effectiveness of the algorithm improvement.(4)The safety inspection system of rolling equipment based on Jeston Xavier NX platform is developed and tested.The MF-SSD algorithm was applied to the Jeston Xavier NX hardware platform,and the human-computer interaction interface of the detection system was designed based on Py Qt5.At the same time,the method of detection and delineation of dangerous areas was studied.Finally,the real scene of the rolling site was simulated for testing.The results show that the system can perform accurate and real-time detection of the worker's hand in the delineated area.When the worker's hand is detected in the demarcated area,the detection system can send a real-time signal to automatically power off the machine,so as to realize the intelligent safety control of the equipment and protect the worker's hand safety.