Research On Visual Inspection Technology Of Cable Coaxiality Based On Convolutional Neural Network

Traditional cable coaxiality detection method is based on X-ray machine image feature detection method,the detection accuracy can not be guaranteed quantitatively,the application range is small,and the anti-interference ability is poor.In this paper,under the support of automatic image acquisition system,combined with convolutional neural network（CNN）and visual detection technology,a set of intelligent cable coaxiality detection method is proposed.Firstly,according to the requirements of cable image acquisition,the automatic image acquisition platform is built,and the corresponding equipment control and image acquisition program are compiled.A large number of image data sets are formed by collecting and sorting the image data of mineral insulated cable.Then,the collected images are analyzed,and the coaxiality detection scheme is designed.Canny edge detection and Hough transform line detection technology are used to detect cable images.According to different image quality,different detection parameters are selected for detection,and classification is made according to the detection parameters required by the images to form a labeled data set.Finally,on the basis of the visual inspection program,an intelligent detection method of cable coaxiality based on improved lightweight convolutional neural network model is proposed.The convolutional neural network model is built,and the cable image data are formed into effective training set and test set.The trained convolutional neural network model adjusts the threshold parameters of Canny operator edge detection and Hough transform line detection according to different image categories meet the detection requirements.This method makes full use of the intelligent advantages of machine learning algorithm and enhances the robustness of traditional visual inspection program.It is applied to the coaxiality detection of mineral insulated cable with complex production process.The classification success rate of convolution neural network model reaches 98.5%,and the success rate of coaxiality detection reaches 99.3%,which fully meets the requirements of real-time detection technology of enterprises.