Study On Classification Of Foreign Objects On The Coal Conveyor Belt Based On Fusion Residual Information

China is the world’s largest coal producer.Plenty of studies on the safe and efficient transportation of coal in mines reveal that the belts often suffer some hazards caused by foreign objects such as large gangue,bolts and other foreign bodies scratching,tearing the belt,and blocking the coal discharge point in the process of coal transportation.If the early warning,sorting and linkage control are not timely,it will seriously affect the coal transportation efficiency.Therefore,in order to ensure the safety of coal mine production,it is necessary to quickly and accurately discriminate and alarm the presence of foreign objects on the coal conveyor belt.With the progress of Internet technology and the development of computer hardware resources,deep learning plays an important role in various fields with its powerful feature extraction ability.Deep learning technology is used as a mainstream technology to classify and identify foreign objects present on coal conveying belts.The current image classification network’s classification performance in coal mines is not ideal due to the abominable environment and complex background.The classification accuracy and real-time performance of image classification networks have been put forward with high requirements due to the high risk of coal mine underground and the large limitations on hardware equipment.To overcome the problems of large parameter count,poor real-time performance,and low recognition accuracy in existing image classification networks,two lightweight image classification methods are proposed in the thesis.They are lightweight image classification network based on fused multi-channel residual information and lightweight image network classification based on UECBAM attention and multichannel feature fusion respectively.(1)A lightweight image classification network integrating multi-channel residual information is proposed because of the problems of low feature utilization,deep network depth,and large computational load.In the network,the degradation of the model is effectively restrained because the residual structure is used as the basic feature extraction unit.The problems of low feature utilization,feature redundancy and feature loss in the single-channel network have been solved because the multi-channel feature fusion network built by using the cross-learning mechanism has been used.The problem of increasing the loss value of the test set is effectively suppressed in the network training process,and the universality of the network is improved because the loss function is processed by threshold.The experiment shows that the network performs well on both public and mining data sets,and has the advantages of low computational complexity and high classification accuracy.It can quickly and accurately classify and identify foreign objects present on coal conveying belts.(2)A lightweight image classification network based on the fusion of UECBAM’s attention and multi-channel features is proposed to exist in the existing attention mechanism model because of the large number of parameters and the inability to consider both channel and spatial attention.The UECBAM attention mechanism model is constructed based on the CBAM construction idea and the coal mine image texture features as the main features,and based on ECA-Net and ULSAM attention modules.In the model,spatial features of different scales are captured,the dependence between channel features is considered,the correlation between spatial features is focused on,and the problem of difficulty in using features due to insufficient features is solved.The experiment shows that the parameter quantity of the attention mechanism model is reduced,the classification accuracy of the network is improved,and the recognition ability of foreign objects on the coal belt is effectively improved when combined with the multi-channel feature fusion network.The thesis has 54 figures,10 tables and 85 references.