Research On Cross-path Detection Algorithm Based On Feature Extraction And Convolutional Neural Network

China's production and demand for cloth is very large.Compared with the high automation in the production process,the inspection of grey cloth is still mainly artificial,and the test results are affected by the individual differences of the fabrication workers.Therefore,more and more researchers are researching and implementing automated detection of cloth defects,most of which can achieve good detection results.As wide range of defects,Motion Marks have inconspicuous and unrepairable features.Because of these features,Mation Marks have become a difficult point in defect detection and an important target for detection.In this paper,the structure of the automatic inspection machine and some components of each module are briefly introduced,and a large number of grey cloth images collected by this equipment are filtered and processed to establish a Motion Marks' data set.In this paper,Hough transform,autocorrelated coefficient detection,edge detection and Radon transform are implemented.Then based on the analysis of Radon transform effect,an auto-adaptive Radon detection algorithm is proposed: Correlation coefficient and the variance value is calculated according to the texture of the grey fabric,and we use the wavelet transform to eliminate the noise influence.We extract the feature points of the Motion Marks.This algorithm can effectively reduce the artificial parameter settings and have more accurate positioning of the Motions Marks.Then,this paper proposes a convolutional network detection algorithm based on Radon feature.Firstly,two classical networks are implemented and the structure is adjusted according to the situation of this paper.The classification result of the two networks is trained and compared.Then the features extracted by Radon transform are used as the network input to realize the Motions Marks' detection algorithm of the classification firstly and positioning laterly.As a result,the accuracy of this algorithm is better than that of the previous adaptive Radon transform.The simulated Motions Marks' detection system realized by the code implemented by MFC shows that the real-time performance of the algorithm can also meet the requirements.