Research On Defect Detection Technology Of Automobile Instrument Based On Machine Vision And Deep Learning

With the rapid development of Internet technology and electronic information technology,a new type of full LCD car instrument is gradually being used in the car cockpit.Compared with the traditional mechanical car instrument,this new type of full LCD car instrument is due to its own production process.And product accuracy puts forward higher requirements for inspection.How to accurately and efficiently detect automobile instrument defects and accurately classify them is an urgent problem in the field of automobile instrument inspection.Today’s automobile instrument defect detection is mainly based on manual visual inspection and classification.This method is subject to human subjectivity and is not only low in efficiency,but real-time and accuracy cannot be guaranteed.In response to the current problems,this paper proposes the detection of automobile instrument defects based on machine vision combined with deep learning.The main research contents are as follows:(1)First,the automobile instrument defect detection system was designed and built.According to the automobile instrument defect classification and system demand analysis,the hardware platform design,hardware equipment selection.(2)Use the built system platform to complete the collection of car instrument images through machine vision,and perform image preprocessing on the collected original images,so that better quality car instrument images with less interference information can be obtained.(3)Research and analyze the structure of the convolutional neural network and the training process and parameter indicators of the network.The structure and characteristics of the two widely used traditional convolutional neural networks,VGG19 and Inception-V4,are analyzed.Car instrument images are built to train the two kinds of networks,and the training results of the two networks are analyzed.(4)Aiming at the performance and convergence problems of the two traditional networks,a defect recognition and classification algorithm based on the improved Inception-V4 network model is designed.The sample set is used to train the improved Inception-V4 network proposed in this article.And compared with the training results of two traditional networks.Through the training curve of the training process and the use of test set experiments to verify the comparative analysis,the method proposed in this article not only has better performance,faster speed,and stronger convergence,but also in the accuracy rate,precision rate,recall rate,F1 Score,and the detection speed has been greatly improved.(5)Through the Labview development environment,MATLAB Deep Learning toolbox and the Tensor Flow deep learning framework,the automobile instrument defect detection software system will be developed.The performance test of the system shows that the accuracy of the system for the detection and classification of automobile instrument defects has reached 95%,the false detection rate and missed detection rate are above 2.5%,which has a good practical effect.