Industrial Reactor Liquid State Recognition Based On HOG And Improved Color Histogram

The liquid separation of the industrial reactor is carried out according to different liquid status in the reactor,and the separated liquid is used for the production of industrial products.At present,the way to achieve liquid separation is mainly by manual operation.Because the method is low in efficiency and the production cost is too high,the paper proposes a liquid status recognition algorithm,which can be used for the liquid status in the industrial reactor.The improved algorithm are quickly and accurately identified for industrial automation production.The liquid data is divided into five categories by analyzing the liquid data in the industrial reactor.Firstly,when the HOG feature extraction algorithm is used to extract the liquid data,the feature dimension of the liquid data extracted is too high,so the PCA method is used to reduce the dimension of the HOG features.The classifier algorithm uses SVM classification algorithm,random forest classification algorithm and KNN classification algorithm.The research content of this paper is as follows:1.Research on PCA dimensionality reduction classification model based on HOG feature of liquid image data.When the HOG feature extraction algorithm is used to extract the liquid status features,the extracted feature dimension is too high.Therefore,this paper proposes to use the PCA method to reduce the dimension of HOG features,and then compare the features of dimensionality reduction with the features of LBP,LPQ and Haar in three classifiers.It is concluded that the recognition rate of PCA to HOG after dimension reduction in KNN classifier is up to 93.15%.Finally,the influence of PCA contribution rate on HOG feature extraction method is studied.2.Improvement of color histogram feature extraction algorithm based on adaptive threshold method.Firstly,a 256-dimensional feature is obtained by the traditional color histogram feature extraction algorithm,but the feature performs poorly in various classifiers.Then the above features are processed through thresholds to obtain a 256-dimensional binary feature.The recognition rate of the binary feature recognition is improved obviously by using the three classifier algorithms in this paper.However,setting the threshold size has a great influence on the recognition accuracy.Finally,the adaptive threshold method is proposed to form the binarization feature.The color histogram feature extraction algorithm based on adaptive threshold improvement has high recognition accuracy in random forest classification algorithm and SVM classifier algorithm.At the same time,through the experimental comparison,the improved color histogram feature extraction algorithm not only has high accuracy for all liquid state data recognition,but also maintains high recognition accuracy under the circumstance of less training data.