Research On Longitudinal Tear Detection Of Conveyor Belt Based On Deep Learning

Conveyor belt conveyors are important transmission equipment for transporting coal and ore in ports and mines.When the conveyor belt is running,during the operation of the conveyor belt,longitudinal tears will occur due to unexpected events.At this time,if an accident occurs,it is necessary to detect the tearing fault and shut down in time,otherwise the conveyor belt continues to run for a long distance will cause the conveyor belt to tear for a long distance,causing serious economic losses.When the conveyor is working,in order to quickly detect and accurately detect the longitudinal tearing accident of the conveyor belt,this project subject deeply studies the background of the longitudinal tearing of the conveyor belt and analyzes it in detail.The detection device analyzes the characteristic mechanism of the sample,first separates the foreground and background of the sample data,that is,divides the laser line area,calculates its center skeleton line in the laser line area,and finally analyzes the characteristics of the center line In order to indirectly determine whether the conveyor belt has a longitudinal tearing accident.For the segmentation algorithm between the laser line and the background,in order to separate the background and the foreground for more optimal segmentation,three methods are used to compare and select the optimal segmentation method.First,the Otsu method based on statistical criteria is used.In some scenes It has a good segmentation effect,but Otsu effect is not good in complex scenes,and it is prone to over-segmentation;then iterative method and search approximation method are introduced for image segmentation,which is superior to Otsu segmentation algorithm in some complex scenes.This algorithm The laser line is divided in the column direction.The advantage is that on each cross-section,their thresholds do not interfere with each other and do not affect each other,which plays a certain role in removing light reflection interference and various noises on the lower surface of the conveyor belt,but in some scenarios,Has the problem of under-segmentation;for this reason,it proposes a segmentation based on semantic segmentation network,segmenting the sample image,making a data set on the sample through the neural network,classifying training,optimizing the network model,and comparing with the real segmentation through the test of the test set.Get accurate segmentation results.In order to extract the centerline of the laser line accurately and at high speed and in real time,first use the centroid method to initially extract the centerline of the laser line,but the extraction effect is roughly inaccurate,and then calculate the normal of the light bar in this direction The Gaussian algorithm and curve fitting are used to achieve a more accurate extraction of the center line of the light.Afterwards,the distance relationship between the pixels adjacent to the center line of the light and the curvature change of the fitting of several consecutive pixels are analyzed to determine whether the light bar is torn or not,thereby verifying the correctness and accuracy of several algorithms.