Automobile Fuse Box Detection Algorithm Based On Deep Learning

With the continuous development of computer vision,the importance of image processing is becoming more and more obvious.How to make full use of image information is becoming more and more important.Nowadays,the achievements of the development of computer vision field are gradually applied in people’s daily life.In the field of security,the security system reduces the patrol burden of security personnel;In the medical field,good medical image processing can help doctors better understand what is wrong with patients;In the field of traffic,the real-time monitoring and analysis of traffic conditions have effectively reduced the occurrence of traffic accidents.Fuse box as an important part of the automobile parts,the correct installation of fuse and relay for the importance of driving safety can not be ignored,so it is neccessary to ensure high precision in its detection.However,the traditional manual detection method has many problems such as low detection efficiency,error detection and leakage detection,which cannot meet the high quality development concept.In the field of object detection,convolutional neural networks are used to extract image features for subsequent detection.The application of deep learning in automobile fuse detection can greatly accelerate the detection speed,improve the detection accuracy,and provide help for the detection of fuse defects in the fuse box.Therefore,this paper studies the detection algorithm of fuses and relays in the automobile fuse box.The main research work includes the following three aspects:(1)In this paper,a small number of pictures of the fuse box were used as the initial data set,a floating window was set for the sample in the initial data set for clipping and sampling,and LabelImg was used for manual annotation.Finally,a data set containing 5000 samples was obtained through data augmentation to improve the generalization ability of model training.(2)Compared with the detection effect of traditional networks SSD and Faster R-CNN on the fuse box data set,Faster R-CNN network is chosen as the basic network in this paper.On the basis of this,aiming at different scales in the fuse box,F-Faster R-CNN network is proposed in this paper.The feature pyramid network structure is introduced in this network,and multi-scale feature extraction is realized by using different scale receptive fields.(3)In order to solve the problems of the F-Faster R-CNN network model,such as large number of model parameters and long prediction time,separable convolution and inverse residual structures are introduced.Meanwhile,the detection accuracy of MF-Faster R-CNN network,which is obtained by combining feature pyramid network with different input layers,was studied.On the basis of achieving a certain level of detection accuracy,the detection speed is significantly improved.