基于稀疏化神经网络的浮选泡沫图像特征选择

Aiming at the problem that the froth characteristics are complicated and not conducive to the modeling control, this paper proposes a bubble image feature selection method based on sparse neural networks. Compared with most sparse learning methods, the linear regression model is used as the loss function. The neural network model closer to the nonlinear actual industrial process is used as the loss function, and the L_{2, 1}-norm constraint condition is added to achieve the effect of feature selection. This method establishes a feature selection method based on the characteristics of the foam to solve the regression problem with the mineral level, and the optimal solution is calculated by the near-point gradient method. The comprehensive ranking of the first layer weights obtains the corresponding feature selection results. Finally, the support vector machine is used to detect the different feature combinations of the input samples, and the optimal feature combination of the flotation process is obtained. The industrial data simulation results show that the proposed method can effectively realize the dimensional reduction of the bubble image.