Research On Semantic Segmentation Method Of Coal And Rock Images Based On U-net

China’s coal industry has always been an important basic industry that supports national economic construction and ensures social development.It has always provided solid guarantees for the country’s energy security.Coal rock recognition is one of the key technologies to promote the intelligent construction of the coal industry.The current image semantic segmentation technology can distinguish between coal and rock parts in coal rock images compared to image recognition technology,which helps to comprehensively mine work surface recognition work.At present,there are few semantic segmentation based on coal rock images,and most of them improve semantic segmentation capabilities for the links of high-level context information of the image.This article takes the semantic segmentation of coal rock images as the main research goal.The U-Net network model is the main aspects:(1)Large number of pooling operations reduces the model feature extraction capability;(2)The shallow information of the image is not fully used.For the above two issues,the specific research work of this article is as follows:(1)Aiming at the problem of a large number of pooling operations in the U-NET network model that affects image feature extraction capabilities,design a coal-rock image semantic segmentation model Res Ne St-Unet.This model is based on the U-Net model structure.Based on the U-Net network,a sampling module with a combined attention mechanism and a sampling module with a channel attention mechanism are used to enhance the ability to extract the coal rock image by the network model.By improving the characteristics of the model’s feature extraction and the generalization ability of the model,the accuracy of semantic segmentation is further improved.Finally,two different evaluation indicators and standards were compared and analyzed by the experimental results.Experiments have shown that the pixel accuracy and intercourse of the network model are 96.62%and 94.83%,respectively.Compared with traditional network models such as U-Net and FCN and CA-Poly-Deep Labv3+ network models,U-Net ++ network models,etc.The models improve the accuracy of the recognition of coal rock images.It proves that this model has a certain advantage over the semantic segmentation ability of coal rock images compared to other network models.(2)Most of the semantic segmentation research is mostly based on the acquisition of highlevel information and ignores low-level information.It has designed a semantic segmentation model LST-Res Unet based on coal rock image texture analysis.Based on the Res Ne St-Unet model,texture quantitative counting is the calculation of QCO to quantify the texture information in the coal rock image and make full use of the low-level texture in the image in the image.Information,and then the LST module composed of the Texture Enhancement Module(TME)and the Pyramid Texture Feature Extraction Module(PTFEM)enhances and extracts the texture information in the coal rock image.The final network model is determined by experiments to verify the effectiveness of the LST module.At the same time,compare experiments with the CA-Poly-Deep Labv3+ network model,the U-Net ++ network model.The results showed that the pixel accuracy of the network model could reach 97.56%,and the average interchange and meal ratio was 87.93%.In the case of a darker environment and the existence of light source reflexes,they all had better segmentation results.