Extraction And Research On Characteristic Parameters Of Coal Flotation Foam Texture

In the process of foam flotation,workers adjust the flotation operation according to the visual state of the flotation foam,and the level of automation is low.Applying image recognition technology to the flotation process to improve the level of the flotation process and realize the intelligent operation and information display of the flotation process has important research significance and application value for the automated development of the flotation process.Extract visual features such as texture and size of flotation froth image,and recognize the flotation status by establishing a machine learning classification model.The main research work is as follows:A slime flotation experiment was performed on an XFD 1.5L flotation machine,and foam images were collected through a machine vision system built with cameras,lenses,and light sources.The foam image is subjected to operations such as homomorphic filtering,adaptive morphology operation,and threshold segmentation to extract the quantitative features and the pixel features of the bright spots.The gray features of pixel mean,variance and kurtosis value variance were collected based on gray matrix.The parameters of Tamura texture feature are coarseness,directionality and contrast.The feature data of 200 images of the three categories were extracted by the method of bubble size feature,gray scale feature and texture feature extraction,and the training set and test set for bubble condition recognition were established.The 150 foam images were trained and 50 images were used to verify the classification results.The support vector machine and multi-layer perceptron,were used for classification training and testing,and the classification recognition rate of the fusion of single feature and multi-foam feature was obtained.The results show that the classification recognition accuracy of multi-foam support vector machine and multi-layer perceptron is better than that of single image feature.Using the Alex Net network to classification and recognition image.Join dropout function to prevent over fitting and gradient disappearing,it is concluded that the 0.5 probability recognition effect is good,the machine learning SVM,MLP and CNN were used to identify the different sample size of bubble image contrast experiment,the results show that the SVM can be obtained under the condition of low sample size better classification results;With the increase of the number of samples,Alex Net's classification and recognition accuracy gradually increased.The depend degree of MLP on sample size in between.The classification performance of Alex Net network model is the best under the condition of the existing maximum image sample.Figure 41;Table 7;Reference 73.