Research On Auto Parts Detection Technology Based On Image Processing

In the process of installing buttons on the automobile hose,due to the production environment,installation process,and other reasons,problems such as insufficient number of buttons and incorrect installation positions often occur,which directly affects the quality of the connection between the automobile hose.In the past,manual inspection methods were mostly used for the qualified inspection of buttons.However,due to various human factors,product inspection efficiency and quality were low,which could not meet the requirements of production qualification inspection.Therefore,the detection method of button is deeply studied in this paper.In this paper,deep learning technology is proposed for button detection to solve the problem of low detection accuracy of traditional target detection algorithms and easy interference by external factors.By analyzing the current deep learning target detection technology,the overall design scheme of button detection is proposed.For the detection of the number of buttons,an improved Faster R-CNN network based on multi-scale fusion and attention mechanism is proposed to solve the difficulty of small target detection.The Faster R-CNN network is improved by selecting a suitable residual feature extraction network,introducing a multi-scale fusion feature pyramid structure and channel attention mechanism.It can effectively improve the network structure of Faster R-CNN.As a result,the ability of the network to detect buttons is improved,and the loss of target information is reduced.The test results show that the mean average precision of the improved Faster R-CNN has increased from 89.09% to97.89%.It improves the detection accuracy and proves the feasibility of improving the Faster R-CNN network for small target detection.For the position detection of the button,Support Vector Domain Description(SVDD)algorithm was proposed to find the abnormal data in the button position.First,this paper uses the disparity map to obtain the three-dimensional coordinates of the button and the white line as the input value of the SVDD algorithm.Then,through the training of the button position detection model and related parameters,the button position qualified detection is realized.The experimental results show that the AUC value is 98.761% when the one-class support vector machine SVDD is used for the qualified detection of the button position,indicating that the model has a good predictive effect on the positive case and can identify abnormal data.