Study On PCA And CNN Coupling Algorithm And Its Application In Pellet Phase Recognition

Convolutional Neural Network(CNN)is a kind of deep feedforward Neural network.As a powerful image recognition tool,CNN not only achieves good results,but also has some problems in practical application,such as weight optimization,overfitting and network degradation.Therefore,it is very important to find a suitable improvement scheme and integrate CNN into the new application model in a reasonable form to solve practical problems.Based on the problem of overfitting and feature fusion of CNN,this paper improved the traditional CNN,combined the advantages of principal Component Analysis(PCA)in image processing,established the coupling algorithm of PCA and CNN,and applies it to the problem of pellet phase recognition.Firstly,the basic structure and training process of CNN were analyzed,focusing on the theories of local sensing field,weight sharing and convolution computation,and two kinds of optimization problems,namely model overfitting problem and feature fusion problem,were proposed in the practical application of CNN algorithm.Secondly,aiming at the problem of overfitting,seven image processing methods were proposed,which were image scaling,image brightness and contrast adjustment,image flipping,image rotation,gaussian filtering,image segmentation and PCA dimensionality reduction.By improving the quality of sample data from the aspects of data enhancement and data dimensionality reduction,the problem of model overfitting was effectively avoided,and the model recognition accuracy is improved while the model calculation speed is improved.Aiming at the problem of feature fusion,the coupling algorithm of PCA and CNN was established.The features obtained by each convolution in the CNN model were processed by PCA dimensionality reduction,and the principal feature extraction of PCA was integrated into the deep learning of CNN.The two algorithms were effectively combined to realize the multi-layer feature fusion of the shallow information feature extracted by PCA and the deep feature extracted by CNN,thus enhancing the performance of the overall model on the image.Finally,based on the pre-treated pellet phase sample set,PCA and CNN coupling model was established to identify the position and alkalinity of pellet phase,and the recognition results were compared with those of traditional CNN algorithm.By comparison,it was found that the recognition accuracy of PCA and CNN coupling algorithm was 93.82% and 91.26%,respectively,which were both higher than the accuracy of traditional CNN algorithms(92.73% and 88.93%).Thus,the accuracy of PCA and CNN coupling model and its applicability in pellet phase recognition were verified.It extends the application scope of deep learning algorithm in the field of metallurgy,and is conducive to promoting the development of China's steel industry to the direction of intelligence.Figure 39;Table 6;Reference 83...