Research On Gangue Identification Based On Video Processing

Raw coal must be preliminarily drained before being selected.For +50mm raw coal,artifical gangue selected operation are generally required.Not only the intensity of labor is very high and the efficiency of manufacturing is low,but also missed elections and mischoosing will occur.Therefore,a gangue identification method based on video processing is proposed to simulate artificially selected gangue.Gangue identification based on video processing is to identify the raw coal image of the real-time monitoring video.In this paper,establishing a coal and gangue image identification model through machine learning,and the classification effect is identified based on experimental exploration.The raw coal moves with the belt,and the background coal differential is used to detect the raw coal from the video,and the window image is cut in the frame image for identification.The low-order moments of RGB space,HSV space,and gray-scale space are selected as the color features;the energy,contrast,correlation,entropy of the gray-level co-occurrence matrix,and the texture of Tamura Roughness,contrast,directionality are used as texture features.The 28 feature parameters describing image color and texture information were extracted.After the feature preliminary analysis,RGB spatial features were eliminated as redundancy,and the image feature parameters were reduced to 19.Calculating weights by Relief algorithm to characterize the contribution of features in the identification.Studying the identification effect of "k-nearest neighbor",support vector machine.The nearest neighbor number k is found by 5-fold cross validation in the kNN algorithm identification;the optimal parameters c and g are found by 5-fold cross validation and logarithmic coordinate grid search in the support vector machine algorithm;In the neural network identification,a three-layer BP neural network with 8-node hidden layer and S-shaped activation function is used.Studying the identification effect of the three algorithms,we find that the support vector machine algorithm is the best.Combining support vector machine algorithm and feature selection,feature recursive elimination based on weights of Relief algorithm is used to determine the optimal subset of features and improve the efficiency of identification model.The surface conditions of the raw coal are divided into four types: slurry-free and dry surface,slurry-free and wet surface,surface covered with dry slurry,surface covered with dry slurry.Based on image classification,the identification tests of the raw coal in Baidaigou Mine and Dafeng Mine were conducted.Gangue identification is divided into two types of identification of coal and gangue and three types of identification of coal,gangue and belt.The gangue identification of different surface types is studied in two kinds of mines.Weight-base feature recursive eliminating for the best performing SVM algorithms among the three identification models.Determing the optimal feature subset of the model.For two categories,the 5 times average of Baidaigou Mine identification rate in 4surface types is: 98%,96%,94%,96.5%;Dafeng Mine is: 97%,97%,95.5%,95.5%.For three categories,the 5 times average of Baidaigou Mine identification rate in 4surface types is: 92.33%,95.67%,96%,98.33%;Dafeng Mine is: 92.33%,96%,94.33%,96%.It has good identification effect.