Study On Coal Gangue Image Recognition Method Based On Convolutional Neural Network

Coal plays an important role in the development of national economy.In the process of coal mining and coal washing,coal gangue often comes into being.The existence of coal gangue makes the mined coal not pure and of low quality,which seriously affects the use of coal.Therefore,it is of great significance to remove gangue from coal.At present,there are many traditional separation methods,but most of them have low separation degree and high production cost,which can not meet the current separation of coal and gangue.With the advent of the era of big data and the rapid development of image recognition technology,the combination of coal and gangue separation technology and image recognition technology has become the field that researchers are gradually exploring.In this paper,based on convolution neural network,the coal and gangue images are identified,mainly including the following contents:Taking the recognition of coal and gangue as the research object,the collected coal and gangue images are screened,and then the number of samples is expanded by using data enhancement technology,mainly including random cutting and turning operation.Finally,the image of coal and gangue is divided into training set and testing set,which lays a foundation for the classification of coal and gangue.Aiming at the recognition of coal and gangue,a network basic module of multi branch feature fusion is proposed based on the multi branch convolution idea of googlenet,and then the classic AlexNet is improved.Finally,the performance of the improved network is evaluated on the image data set of coal gangue.The experimental results show that the accuracy of AlexNet-Branch model is 1.55%higher than that of fine-tuning AlexNet model,and the model size is smaller and the number of parameters is less.In order to solve the problem of coal and gangue identification in the environment with limited computing power,the fine-tuning lightweight SqueezeNet is used to identify coal and gangue,and the SqueezeNet-DS is constructed based on the SqueezeNet and the deep separable convolution.Finally,the experimental results of three models,SqueezeNet,SqueezeNet-Residual and SqueezeNet-DS,are compared and analyzed.The experimental results show that the accuracy of SqueezeNet-Residual model is 92.35%,and that of SqueezeNet-DS model is 89.14%.Compared with SqueezeNet,the size of SqueezeNet-DS network is about 1/2 of that of SqueezeNet,but the accuracy is only reduced by 2.02%.It shows that using deep separable convolution to improve SqueezeNet can not only reduce the network size,but also ensure a certain accuracy,which can be effectively used for coal gangue image recognition.Figure 30 table 10 reference 81...