Image Segmentation And Feature Extraction In Coal Gangue

The method of coal and gangue identification based on optical image has become the focus of modern coal and gangue identification due to its stable equipment,high safety factor,low cost and easy implementation.Among them,target image extraction and feature extraction are the premise of effective identification of gangue.At present,in the research on coal and gangue identification,the segmentation accuracy and time efficiency of extracting target images cannot be well considered,and the research on coal and gangue characteristics is not comprehensive and in-depth,So,the following improvements methods are proposed in this thesis.(1)In order to solve the problem of large amount of calculation and longtime-consuming of tradition two-dimensional entropy method,gray wolf optimization algorithm is introduced,and a gray wolf optimization two-dimensional entropy segmentation algorithm based on improved nonlinear control factor is proposed.This method uses the monotonicity of the positive semi-axis of the cubic function to improve the control factor in the gray wolf optimization algorithm,which is transformed from linear iteration to nonlinear iteration,so that the global search optimization ability and local search optimization ability of the algorithm can achieve a better balance.The effectiveness of the improved gray wolf optimization algorithm in search ability and iteration speed is verified by the standard test function.The accuracy of this algorithm in coal and gangue segmentation is 13.95%and 9.26%higher than that of two-dimensional entropy method,and the-segmentation time is 2.21 times and 1.17 times faster respectively.(2)In addition to the traditional gray level statistical features and gray level co-occurrence matrix(GLCM),the characteristics of local binary pattern(LBP)and histogram of oriented gradient(HOG)are further studied.Moreover,sample entropy,approximate entropy and permutation entropy are introduced to characterize signal complexity.(3)The signal feature and combined feature are combined with support vector machine to recognize coal and gangue.The experimental results show that the average recognize rate is improved by 13.79%,16.99%,and 1.12%after introducing HOG feature,LBP feature and sample entropy feature,respectively.The average recognize rate of combing multiple image features is 92.85%,which is increased by 17.9%.