Research On Coal-Rock Recognition Method Based On Convolutional Neural Network

Intelligent construction of coal mine is the only way to realize safe,efficient and green mining of coal mine,and coal-rock interface recognition technology is the key technology to realize intelligent mining of shearer,improve coal mining efficiency and reduce equipment failure rate.Some traditional coal-rock recognition technologies based on radar,infrared,cutting current and other sensing signals are vulnerable to the limitations of geological conditions and coal mining technology,and are difficult to be popularized in practical application.In recent years,with the development of machine vision,using machine vision image recognition algorithm to develop new coal-rock recognition technology has become the focus of research.At present,the research of this technology is still in the preliminary stage.Therefore,this paper takes convolution neural network as the main means and self-made coal-gangue data set and coal strata data set as the support to study the key technologies such as coal-gangue intelligent classification and coal-rock strata intelligent semantic segmentation.The main research work of this paper can be summarized into the following three parts.(1)Research on coal-rock image classification and recognition model based on deep learning.Data sets containing more than 2000 coal-rock block images were collected and made.After pre-processing such as fixed size,cutting,filtering and noise reduction,the morphology,color,texture and other features of coal and gangue images were extracted using machine vision technology,and the heterogeneity of coal and gangue targets was analyzed in target description.By building a VGG16 network of self-defined full connection layer,the coal-rock image features are extracted layer by layer and instantiated to obtain the coal-rock visual features extracted from each layer network.A coal block classification subsystem based on the improved VGG16 network is proposed.The experimental results show that the VGG16 network can distinguish the features of coal and rock pixels,and the accuracy of the VGG16 network can reach 97% in the coal block data set.It is applicable and feasible in the coal-rock block classification and identification task.(2)Research on superpixel semantic segmentation model of coal-rock strata based on cavity convolution.More than 3000 images of coal-rock strata were collected as semantic segmentation experimental data set.Deeplab v3+ semantic segmentation network was used as the reference network to carry out the segmentation and recognition experiment of coal-rock interface.The feature images output by the backbone network were segmented by the robustness of spatial pyramid pooling structure,and finally the prediction map of coal-rock domain segmentation was obtained after up-sampling.The experimental results show that the evaluation index m Io U value of the segmentation model under the optimal iterative training cycle reaches 53.91,which basically realizes the segmentation recognition of coal-rock domain.(3)The classification and recognition model of coal-gangue and the semantic segmentation model of coal-rock strata are improved and optimized,and the mixed model of coal-rock information recognition integrating multiple mechanisms is constructed.In view of the lack of full use and mining of convolutional network in existing coal-rock recognition studies,for example,when using convolutional network,feature learning is only carried out on the original image while ignoring the influence of possible artificial design features on classification results.In this paper,a classification network model of coal-rock block fusion algorithm DEM＿LC＿VGG16 is constructed by adding data enhancement module(DEM)and image saliency processing module(LC algorithm)to the VGG16 classification network.For the coal-rock layer segmentation model,Channel Attention mechanism was embedded into the backbone network of Deeplab v3+ segmentation algorithm,and adaptive learning strategy was used to optimize network training,and a CA＿Poly＿Deeplab v3+ coal-rock layer semantic segmentation and fusion algorithm model was constructed.The experimental results show that the accuracy of the coal-rock block classification model optimized by the fusion algorithm is improved by 1.72%.Compared with the benchmark segmentation network,the multi-mechanism coal-rock segmentation recognition model improves the m Io U value by7.74 in the segmentation evaluation index,and achieves a good coal-rock domain segmentation effect in the mixed coal-rock strata data set.