Research On Feature Selection Method Of Foam Image In Flotation Process

Foam flotation is a mineral separation method that uses the difference in physical and chemical properties of mineral particles to achieve effective separation of different minerals.In recent years,with the continuous development of image processing technology and computer technology,machine vision technology has been widely used in mineral foam flotation processes.Through image segmentation,color extraction and ripple analysis,a large number of mineral flotation foam features can be extracted.But because there are many methods for extracting the features of the foam image,the same feature can be described by multiple machine features,and there is a non-linear relationship between the features.In the models such as soft measurement and fault diagnosis,it is easy to cause information redundancy.Therefore,dimensionality reduction is urgently needed to reduce the complexity of foam characteristics and identify key foam characteristics.Previous research focused on judging the key features of practical experience,but this article attempts to solve this problem in the theory of artificial intelligence through machine learning and industrial big data.Aiming at the above problems,this paper proposes a method for feature selection of foam images based on sparse neural network to solve the feature selection of the mineral item level regression problem.Aiming at the problems of measurement disturbance and artificial errors in complex industrial processes,a robust sparse neural network is established.At the same time,the support vector machine is used to test the effect of different feature combinations on the input samples,and the feature subsets are compared to obtain the optimal feature combination for the flotation process.The specific research content is as follows:1.Aiming at the problem that the flotation industrial process is complicated and not conducive to direct measurement,the relevant characteristics of the flotation index are selected based on the experience of foam flotation experts,and machine vision technology is used to extract the features of the collected foam image data.The analysis software performs correlation analysis on the extracted feature data sets to provide data support for the next step of identifying key features.2.Because there are many existing foam image feature extraction methods and the extracted features are also different,the process of modeling using foam image features is prone to cause information redundancy.The usual practice is to use expert experience to perform a subjective analysis,and then extract the required features.This paper proposes a sparse neural network method,using a neural network model that is closer to the nonlinear characteristics of actual industrial processes as a loss function,and adding norm constraints to achieve the effect of feature selection.The best solution is to obtain the feature selection subset by comprehensively ranking the weights of the first layer.Finally,the support vector machine is used to test the effect of different feature combinations on the input samples.3.Due to measurement disturbances and artificial errors,industrial measurement data is often doped with some undesirable disturbances.Therefore,based on the original sparse neural network,a robust constraint is added,and at the same time,for the problem that the sparse term is a non-smooth function is not conducive to solving,the smoothing term is introduced to optimize the objective function solution and reduce the algorithm complexity.Finally,the original sparse neural network and robust sparse neural network are compared to verify the antiinterference ability of the method.In order to reduce the complexity of establishing the regression function model of the flotation foam image and improve the prediction accuracy of the model,based on the sparse model,this paper proposes a sparse neural network feature selection method for nonlinear regression problems.After real data simulation experiments,the method can effectively deal with nonlinear problems,and better retain the data characteristics.The method can more effectively deal with the feature selection of the actual industrial process,which is beneficial to the application of image processing and machine vision in industrial practice.