Research On Target Detection Algorithm For Coal And Gangue

Coal gangue sorting has always occupied an important position in clean coal technology,and realizing efficient sorting of coal and gangue is conducive to improving the utilization rate of coal and reducing environmental pollution.In the complex coal gangue sorting environment,how to accurately detect coal and gangue targets is a key issue in the coal gangue sorting process.In recent years,with the rapid development of deep learning,Convolutional Neural Networks(CNN)are widely used in computer vision tasks such as natural scene classification and object detection because they can automatically extract high-level and robust feature representations of images.At present,some progress has been made in the research of coal gangue target detection using convolutional neural networks.The thesis studies some problems existing in the process of coal and gangue target detection,such as the noise,complex background and slow detection speed of coal and gangue images.In order to solve the above problems,the thesis proposes a coal and gangue image denoization method and two coal and gangue target detection methods.The main research work is as follows:(1)Aiming at the problem of noise in coal gangue images,a method for denoising coal gangue images based on Residual-Pix2Pix model is proposed.This method uses the encoderdecoder network combined with the residual block to construct the generative network,which can effectively extract the depth features and retain the detailed information of the image.At the same time,in order to enrich the extracted features without changing the size of the feature map,the convolutional layer in the residual block uses dilated convolution to increase the receptive field of the feature map by setting different dilation rates.In addition,the Smooth L1 loss function is used as the metric between the generated image and the real image,so as to obtain better coal gangue image denoising effect.(2)Aiming at the problem of low target detection accuracy caused by the complex background of coal gangue images,a coal gangue target detection method based on selfattention mechanism is proposed.This method takes the SSD target detection model as the basic framework,and improves the focus of the feature map on the foreground area by introducing a self-attention module in the shallow feature map,reducing the interference of background redundant information.At the same time,in order to optimize the detection performance of the model,GIoU Loss and Focal Loss are used to reconstruct the loss function.In addition,the K-means algorithm is used to cluster the anchor frame of the coal gangue data set,and the size of the anchor frame is redesigned to further improve the accuracy of coal gangue target detection.(3)Aiming at the problem of low detection speed of coal gangue target detection model based on deep learning due to the limitation of storage and computing resources of the current embedded hardware platform,a lightweight coal gangue target detection method based on anchor-free is proposed.In this method,the anchor-free target detection model FCOS(Full Convolutional One Stage,FCOS)is selected as the basic framework,and the extended GhostNet is used as the backbone network for feature extraction,which reduces the amount of parameters and computation of the model,thereby improving the detection speed of the model.In addition,BiFPN(Bi-directional Feature Pyramid Network,BiFPN)is used in the FCOS network structure to fuse multi-scale features to reduce the loss of semantic information in the process of feature transfer.At the same time,the feature map connection method in BiFPN is changed,and the feature map-guided strategy is used for connection to generate more representative fusion features,so as to obtain better detection results.