| Coal is an important energy source,which is very beneficial to the economic and social development of the country.At present,the production and consumption of coal in China are ranked among the top in the world.Therefore,the reasonable development and utilization of coal is of great significance to the industrial and economic development of China.In the process of coal development,gangue image recognition is the key technology to achieve automatic gangue sorting in the coal processing.For the current coal gangue image recognition method is sensitive to the acquisition of illuminance,the recognition effect is unstable,and the application environment requirements are strict,a coal gangue recognition method based on laser speckle images is proposed in this thesis.Firstly,the principle of coal gangue laser speckle recognition is analyzed,the acquisition scheme of coal gangue laser speckle image is designed,the coal gangue laser scattering laboratory acquisition system is built,and the coal and gangue samples from two mining areas are used as objects to collect data sets.The preprocessing method of laser speckle image is studied.First,according to the noise characteristics of speckle image,and the images are denoised by Gaussian filter.Second,the extraction algorithm of speckle region of interest is studied,and the Otsu threshold segmentation method based on Gaussian pyramid in Lab color space is proposed,which realizes the accurate segmentation of speckle target area in coal gangue laser speckle image.With the center of mass of the speckle region as the central coordinate,a 250 X 250 pixel image is extracted from the speckle target region as the final region of interest,which lays the data foundation for the subsequent coal and gangue feature extraction and recognition.Secondly,the method of feature extraction for coal gangue speckle image is studied,which contains grayscale features and texture features.Aiming at the coal gangue laser speckle image,a Gray Level Size Zone Matrix feature extraction algorithm based on speckle-like texture is proposed in this study,and 7-dimensional features are constructed based on this matrix.A 4-dimensional grayscale feature based on the grayscale histogram and a 4-dimensional texture feature based on the Grayscale Co-occurrence Matrix are also constructed,and a total of 15-dimensional coal gangue laser speckle fusion features are constructed.The extracted feature data were analyzed to explore their expressiveness in terms of inter-class differences and intra-class differences between coal and gangue.Finally,based on designed the feature extraction method,the support vector machine and K-nearest neighbor classifier are used as the recognition models,and the coal gangue recognition effects of the two recognition models are compared and analyzed,the experiments show that the recall,accuracy and recognition accuracy of SVM classifier are higher than those of KNN classifier,and the SVM-based recognition model of coal gangue laser speckle image is determined.The stability of laser speckle image for coal gangue recognition is further verified,the classification effect of coal gangue natural image and coal gangue speckle image under the different environmental illumination are compared,the experimental results show that the coal gangue recognition based on laser speckle image proposed in this thesis has higher higher stability to environmental illumination and the obtained image information is more reliable,which provides a new idea and method for industrial applications in the field of coal gangue recognition. |
