ABSTRACT Aiming at the problem that the identification of coal and gangue is easily affected by external factors, a coal gangue identification method based on multi-dimensional gradient features and gray texture features is proposed. Research shows that Sobel operators with different window sizes have different degrees of interest in image textures. The larger the window, the more interested they are in fine textures. The characteristic curves obtained by different Sobel operators are quite different, and the optimal gradient features can be obtained by using various Sobel operators. The separability of gradient features is higher than that of gray texture features, and the gradient features are less affected by light. The classification model established in this paper has a recognition rate of over 97% for different mining areas, and the detection time of a single image is less than 20.5 ms. When the light intensity changes from 600 lux to 1800 lux, the fluctuation range of the F 1-score in this method is only 0.26%. The fusion of multi-dimensional gradient features and gray texture features can significantly reduce the influence of external factors on the recognition model, and improve the recognition accuracy and anti-interference ability of the model.

Recognition of coal and gangue based on multi-dimensional gray gradient feature fusion

In order to improve the EEG recognition accuracy and real-time performance, a classification and recognition method for optimizing the penalty factor C and kernel parameter g of Support Vector Machine (SVM) based on Particle Swarm Optimization (PSO) algorithm is proposed in this paper. Firstly, the Regularization Common Spatial Pattern (R-CSP) was used for EEG feature extraction. Secondly, the penalty factor and the kernel function were optimized by the proposed PSO algorithm. Finally, the constructed SVM classifiers were trained and tested by the two class EEG data of right foot and right hand movements. The experimental results show that the recognition rate for EEG classification of the PSO-SVM is average 2.2% higher than the non-parameter-optimized SVM classifier, and it is significantly higher than the traditional LDA classifier, which proves the feasibility and higher accuracy of the algorithm.

Xue Li

Papers

EEG Characteristics Extraction and Classification Based on R-CSP and PSO-SVM