Research On Image Enhancement And Classification Of Ccoal Gangue Based On Deep Learning

The gangue in the coal gangue is affected by the contamination of coal powder,which makes the coal gangue difficult to distinguish only from the image,which is one of the main factors affecting the recognition accuracy of coal gangue.In view of this problem,in order to increase the difference of coal gangue image and improve the identification accuracy,this thesis starts from the original coal gangue image input by the neural network,and identifies and categorizes the preprocessed coal gangue images.The preprocessing mainly includes coal gangue images.The initial preprocessing of coal gangue images and the enhancement processing of coal gangue images,combined with the convolutional neural network in deep learning,improve the Le Net network,and input the preprocessed coal gangue images into the improved Le Net network for coal gangue recognition and Classification,the main research contents include:(1)Preliminary preprocessing operation on the collected original coal gangue map.The image edge filling method is used to standardize the resolution of the coal gangue image,and then the Gaussian filter algorithm is used to smooth the coal gangue image,and then the Otsu algorithm is used to binarize the filtered and denoised coal gangue image,and finally the edge search is used.The method is used to search the edge of coal gangue object in coal gangue image and determine the edge position to accurately crop the image.(2)Perform image enhancement on the pre-processed coal gangue image,which includes edge enhancement and texture enhancement.Among them,for the edge enhancement of coal gangue map,the Canny edge detection algorithm is used to detect the edge of coal gangue and enhance it to highlight the edge contour of coal gangue.The fusion method performs edge fusion and enhancement processing on the coal gangue image.Regarding the texture enhancement of the image,the gray level co-occurrence matrix is used to extract the mean,variance,homogeneity,contrast,dissimilarity,entropy,energy and autocorrelation of the gangue image texture features,and the pixel-level image fusion and feature-level Combined with the image fusion method,the weighted and fusion method is used to enhance the texture fusion of coal gangue images.(3)Improve the convolutional neural network Le Net,increase the size of the input feature map to increase the amount of image data,change the average pooling to the maximum pooling method,reduce the deviation of the estimated mean caused by the parameter error of the convolutional layer,and improve the Much of the texture information of the coal gangue image is retained,and the Sigmoid activation function is changed to use the Relu function,which greatly improves the convergence speed and avoids the problem of gradient disappearance.(4)Finally,the coal gangue images after image preprocessing of various methods are input into the improved Le Net network for classification experiments.The experimental results show that the edge enhancement and texture enhancement of the image can significantly improve the recognition rate of coal gangue,and the recognition rate can reach 100%,which is higher than some current coal gangue recognition algorithms.The difference is that the fusion ratio of the original image and the edge-enhanced image has little effect on the recognition rate of coal gangue,while the fusion ratio of the original image and the texture-enhanced image will affect the recognition rate of coal gangue.(5)According to the comparative analysis of the experimental results,select the optimal image enhancement method,and use this method to develop coal gangue identification software.The results show that the coal gangue recognition and classification method based on edge enhancement and improved Le Net is better than other methods.Using this method,a GUI application program is written by Tkinter programming to develop coal gangue recognition software and test it.The test results show that the coal gangue recognition software developed using the coal gangue recognition algorithm in this thesis has a higher recognition rate for coal gangue,and is more convenient and fast.