Research On Longitudinal Tear Detection System Of Conveyor Belt Based On Machine Vision

The longitudinal tearing failure of the conveyor belt seriously threatens the safe operation of the belt conveyor,which is mainly checked by manual inspection at present.However,manual inspections have low detection efficiency,long detection cycles,low diagnosis rates and potential production safety hazards.In recent years,the detection methods for longitudinal tearing of conveyor belts based on machine vision technology have been continuously developed,but still have problems of poor recognition rate and stability under complex working conditions.To solve this problem,this article takes the conveyor belt device in laboratory as the research object,and conducts research about the machine-vision-based conveyor belt longitudinal tear fault detection method.The main research contents of this paper are as follows:(1)Due to the fast running speed of the conveyor belt,motion blur will occur in the captured conveyor belt image,which will affect the subsequent image recognition.To solve this problem,a weighted guided filter adaptive image restoration algorithm suitable for conveyor belt images is proposed.This method uses the Lucy-Richardson filtered image as the guide image,introduces a weighting factor into the guide filter model and adaptively change the model parameters relationship to the edge detection result of the Sobel gradient operator,and realizes image restoration with the help of the guide image.The results show that this method can not only perform adaptive restoration according to the tearing area in the image,but also effectively avoid the ringing effect of the non-tearing area during the restoration process,which lays the foundation for the subsequent accurate extraction of tearing image features.(2)A method of longitudinal tear detection for the conveyor belt based on artificially extracted image features is proposed.Firstly,the traditional pulse coupled neural network(PCNN)method is improved.By using Hebb learning rules to update the weight matrix of PCNN,the connection between similar pixels is strengthened,which realizing the complete segmentation of the torn area in the image,and guaranteeing the accuracy for the extraction of the crack shape feature.Secondly,extracting the texture feature of the conveyor belt tearing image and the shape feature of the crack after segmentation,the classification and recognition of the conveyor belt tearing image is realized by using a support vector machine(SVM)classifier based on mixed features.(3)A deep-learning-based conveyor belt longitudinal tear detection method is proposed,and which uses the mask region convolutional neural network(Mask RCNN)to automatically extract and recognize the features of the tear image.By the analysis of Mask RCNN network model parameters,according to the size of the actual data set and the characteristics of the conveyor belt tearing image,the feature extraction network of Mask RCNN is selected,and the anchor frame scale and proportion in the candidate region generation network are improved.The results show that,compared with the unimproved Mask RCNN,the optimized and improved Mask RCNN has effectively improved the recall rate,accuracy and recognition speed of crack target recognition.And the recognition effect of the conveyor belt tearing image is better than that of the SVM classifier based on mixed features.(4)Based on the method proposed in this paper,a real-time detection system for longitudinal tearing of an embedded conveyor belt based on Raspberry Pi is developed,which includes image acquisition module,image processing module and execution module.The experiments result shows that the system can realize the real-time detection of longitudinal tearing faults and the location of the tearing area.With high recognition rate and strong real-time performance,it has strong practical value and is of great significance for improving the safety of conveyor belt operation.