Research On Mine Target Recognition Model Based On Deep Learning

Mineral resources are an important material basis for human economic and social development.With the rapid development of human society and economy,the demand for mineral resources in various industries is increasing.In order to promote the rational development and utilization of mineral resources and reduce the environmental pollution and geological disasters generated in the development process,it is urgent to effectively supervise the mining.However,the current supervision of mine development still remains at the traditional level,i.e.,remote sensing visual interpretation and field field survey,which is not only a large workload but also a long identification period,making it difficult to achieve effective supervision of mine resources.In recent years,the rapid development of deep learning algorithms in the field of remote sensing has provided a new way for the supervision of mine resource development.Therefore,this paper takes the rapid identification of three targets in the process of mine development,such as quarry,industrial square and tailing pond,and realizes the rapid identification of mine development targets through the improvement of YOLOv4 model.The main research contents are as follows.In order to reduce the impact of adding new networks on the operating speed of the model,the structure of CBH-SE is optimized to reduce the number of convolutions.Second,in order to reduce the impact of adding new networks on the running speed of the model,the structure of the model is optimized to reduce the number of convolutions;finally,the residual structure is added to avoid the problem of learning ability degradation when the depth of the model is too deep.The CBH-SE proposed in this paper effectively improves the problem of poor recognition accuracy due to target distortion in the imaging process of mine images.To address the problem of insufficient multi-scale mine target feature extraction capability,this paper replaces the spatial pyramid pooling(SPP)network in YOLOv4 with the empty space convolution pooling pyramid(ASPP).By analyzing the format of the input ASPP feature map and setting the cavity rate suitable for the current feature scale,the model is enhanced with the advantages of cavity convolution and five-layer structure for multi-scale mine target feature extraction.Based on this,this paper developed an automatic mine image cropping tool using Arc GIS to produce a typical mine target dataset.By comparing the improved model before and after the experiment,the results show that the improved model has higher average accuracy(m AP)than the original model,which verifies the effectiveness of the improved model;finally,this paper develops software for mine target identification using Python and the improved YOLOv4 model to achieve effective supervision of mine development targets.