Research On The Detection Method Of Workpiece Surface Defects Based On Deep Learning

The rigorous inspection of the workpiece surface quality has always been an indispensable part of the workpiece production process.In the actual production process,the surface defect detection still relies on traditional manual visual inspection method,which has some shortcomings such as low detection precision,slow speed and cost.Therefore,to obtain a better surface quality of the workpiece product,it is necessary to carry out efficient and high-precision detection and recognition for surface defects.In recent years,with the rapid development of computer technology,deep learning technology has been increasingly applied on industrial detection.This technology is adopted for quality inspection of workpiece products on the automated production line,and the manual visual inspection will be gradually replaced,which are needed by the enterprises.A surface defect detection algorithm based on deep learning for the need of workpiece surface quality inspection is proposed in this thesis,the main research contents of this thesis are as follows:Firstly,the metal surface workpiece processed by milling machine has the characteristics that the defect area is not separated from the background.The traditional defect detection algorithm has limitations,a defect detection based on full convolutional neural network and morphological filtering was proposed to deal with the shortcomings of the traditional defect detection algorithm and the opening and closing operation of the morphological filtering was combined with the algorithm,which reduced the local noise,filled the internal small holes,and optimized the image of the initial segmentation by the full convolutional neural network,and the defect detection was effectively achieved.Secondly,after the presence or absence of detected workpiece surface defect,to facilitate subsequent processing,it was necessary to classify the defect.Based on the research of the convolutional neural network method,an improved convolutional neural network model was proposed to solve the low accuracy problem of workpiece surface defect classification by using the convolutional neural network.Thirdly,according to the problem of high false detection rate of existing surface defect detection algorithms,a decision-level fusion model of multi-cascade classifier based on convolutional neural network was proposed.The model was trained by multiple cascaded classifiers on the third layer of the fully-connected layer and the second layer fully-connected layer features extracted by the local binary pattern and histogram of oriented gradient.The output result was fitted to the posterior probability of the target category,and decision-level fusion was performed by using decision weights,and the result of the decision-level fusion was used as the new classification result.Finally,a workpiece surface defect detection system was designed to verify the effectiveness of the proposed algorithm.the results showed that the designed system could complete the workpiece surface defect detection requirements,and the proposed algorithm could effectively detect the workpiece surface defects.