Ore-crushing Size Detection Based On The Improved Sparse Representation And Bipartite Graph Models

The granularity information of the broken ore can reflect the working condition of the crushing machinery,and the particle size distribution of the ore can be grasped by testing the granularity of the broken ore.Thus,it helps to adjust the width of the ore cutting edge of the crushing machine and adjust the particle size distribution of the broken ore;at the same time,it optimizes the implementation process of the crushing system,reduces the energy consumption and improves the utilization rate of mineral resources.However,the current segmentation algorithm is not effective enough to achieve effective segmentation of the adhesion mineral image,thus affecting the accuracy of particle size determination.Sparse representation can effectively guarantee the global characteristics and semantic segmentation results on the image,but it ignore the importance of the connection between adjacent super pixel blocks.Therefore,aiming at the defects of mined image characteristics and sparse representation algorithm,a bipartite graph segmentation algorithm based on improved sparse representation is proposed.The algorithm firstly pretreated the image by means of superpixel segmentation algorithm.Then,the color and texture feature of all super pixels are extracted and all the superpixel features adjacent to the reconstructed superpixel are made into a dictionary.After that,the reconstruction errors obtained by the dictionary are used to describe the similarity between superpixels.Finally,based on the bipartite segmentation principle,the bipartite model based on sparse representation is constructed,and the image is segmented by T-cut spectral clustering algorithm.The test results on BSDS300 database show that the main evaluation indexes of the improved algorithm are improved before the improvement,especially the PRI index increased from 0.8355 to 0.8361.In the improved algorithm framework,when the weight between pixels and superpixels is 0.045,the reconstruction sparsity is 5,and the LBP feature takes 64 dimensions,the algorithm has the highest PRI index on the standard color image library BSDS300 database.In order to extract the granularity of the ore,each region in the image obtained by the improved sparse representation of the bipartite graph segmentation algorithm is processed.Firstly,the algorithm determines whether the region is ore based on the average gray level of the region.Then Two-value processing and operation are performed on the particle size image.Finally,the contour of the ore is extracted,and the distribution of the particle size is calculated and counted.The experimental results show that this algorithm can effectively solve the segmentation of the ore particle image of adhesion,and effectively extract the grain size information of the ore.