Research On Foreign Object Detection Of Mining Belt Conveyor Based On Anchor-free Algorithm

Foreign matter on the mine belt conveyor has always been one of the important factors affecting the safe production of mines.Accidents caused by foreign objects on mining belt conveyors occur frequently,and the reason is often that foreign objects on mining belt conveyors are not detected and removed.The existence of these foreign objects may cause damage to the belt and related machinery,thereby causing serious safety production accidents.This paper takes the foreign objects on the mining belt conveyor that may lead to safety production accidents as the research object,and discusses a new high-speed and accurate foreign object detection method to ensure the production safety of the mining belt conveyor.The specific work is reflected in the following aspects:(1)Using horizontal mirroring,vertical mirroring and affine transformation to expand the limited collection of foreign object image sample data sets of mining belt conveyors.Using adaptive histogram and median filter to enhance the expanded sample data.The construction of the foreign body sample data set of the mining belt conveyor is completed.(2)The original Center-Net algorithm is used as the basic model for foreign object detection.In order to realize the accurate detection of foreign objects on the mining belt conveyor,the CIOU loss function is used to replace the loss function in the model to speed up the convergence speed of the model.On this basis,an attention mechanism module is added to the Hourglass-104 feature extraction network to improve the ability to extract foreign objects in mining belt conveyors.After the accuracy of the foreign object detection model is improved,the depth separable convolution is used to replace the ordinary convolution in the residual structure of the mining belt transmission foreign object detection model to speed up the model operation.(3)For the detection of foreign objects in the mining belt conveyor proposed in this paper,the corresponding evaluation indicators are established.The recognition accuracy and model running speed of the foreign object detection model for mining belt conveyors were evaluated by using precision(precision),average precision(AP),average average precision(MAP)and model inference speed(FPS).Using the Mine-Center-Net proposed in this paper,the recognition accuracy of long,polygonal and circular foreign objects reaches 0.800,0.950 and 0.911,respectively.Compared with Center-Net,SSD,YOLO V3,and Faster R-CNN,the average accuracy is increased by 8.30%,8.30%,22.30%,and 13.70%,respectively.A new method is provided for the accurate detection of foreign objects in mining belt conveyors.