Research On Visual Detection Method Of Conveyor Belt Deviation Based On Improved YOLACT Algorithm

In the process of transporting bulk cargoes such as grain,ore and coal,the conveyor belt is affected by the impact force of material loading,material gravity and the friction the belt and idlers,which leads to the unbalance force and deviation.Current visual detection methods of conveyor belt deviation have poor robustness and can not adapt for changeable carrying bulks and working environment.Most of them can only detect the operation states of conveyor belts qualitatively,instead of realizing the quantitative detection and risk assessment of conveyor belt deviation.Therefore,this research proposes a visual segmentation algorithm based on convolutional neural network to segment and extract the conveyor belt image,uses line structured light to assist in detecting the material attitude and conveyor belt position,from which a conveyor belt deviation risk detection and evaluation method is established based on the detection results,The method is verified by relevant experiments.The main contents and results of this dissertation are as follows:(1)In order to segment the image of conveyor belt quickly and accurately,an improved YOLACT conveyor belt image segmentation algorithm is proposed.By analyzing the module function and principle of convolutional neural network,YOLACT algorithm is selected as the basic algorithm of conveyor belt detection;Combined with the requirements of conveyor belt segmentation speed and accuracy,Res Ne Xt backbone feature extraction network is adopted and the bottom-up fusion path of feature pyramid is added to improve the feature fusion ability of YOLACT algorithm;The small convolution kernel optimizes the large convolution kernel and cavity convolution to enhance the receptive field of feature extraction of YOLACT algorithm,so as to realize the complete segmentation of conveyor belt edge image.(2)Based on the belt image segmented by the improved YOLACT algorithm,a quantitative detection method of belt deviation characteristics and deviation risk assessment method are proposed.Combined with the characteristics of the running environment of the conveyor belt,the linear structured light is used to enhance the edge image characteristics of the material and the conveyor belt,the quantitative detection model of the material attitude and the conveyor belt position is established,the centroid method is used to calculate the center of gravity position of the material,the proportion of the material bandwidth and the eccentricity of the material are obtained,and the deviation degree of the center line of the conveyor belt and the torsion degree of the conveyor belt are calculated by curve fitting and camera coordinate conversion;On this basis,the conveyor belt deviation index and its weight are determined,and the comprehensive evaluation method of conveyor belt deviation risk is constructed to prevent conveyor belt deviation accidents.(3)Carry out the experimental verification of the improved YOLACT algorithm and conveyor belt deviation detection method,and verify the image segmentation performance,material attitude detection and deviation detection effect of the improved YOLACT algorithm.The experimental results show that the image segmentation accuracy of the improved YOLACT algorithm is improved by 3.78%,and the detection speed is much higher than that of Mask R-CNN and MS-RCNN algorithms.The ablation experiment shows that the backbone network replacement,feature fusion,void convolution and small convolution kernel all play a good role in improving,and can be effectively used in the image segmentation of conveyor belt area.The average error of material offset quantitative detection method applied to soybean,resin particles,corn and rice is 1.895 mm,which can effectively identify the material offset on the conveyor belt.The average deviation of the three offset states of the conveyor belt is 0.73 mm,0.74 mm and 0.74 mm respectively,and the maximum deviation is less than 2 mm,which is 0.35% of the width of the conveyor belt,realizing the accurate measurement of the offset of the conveyor belt.Finally,the evaluation method of conveyor belt deviation is exemplified to verify its evaluation effect,which provides a new evaluation method for conveyor belt deviation risk prediction.