Research On Segmentation And Recognition Algorithm Of Coke Optical Tissue Microscopic Image

Coke is one of the important materials and raw materials for blast furnace smelting.Its quality and physical and chemical properties during thermal reaction are directly related to its own microscopic optical structure.The automatic analysis and identification of coke optical structure can effectively identify the quality and characteristics of coke,so as to ensure the stable operation of the metallurgical process and the rational utilization of resources,which has important practical significance and development prospects for the development of the metallurgical industry.According to the difference in color and texture of different optical tissues,this paper extracts and fuses the color and texture features of superpixels in the main body of optical tissue images.The extracted high-dimensional features were dimensionally reduced,and the improved Kmeans clustering was performed on the dimensionality-reduced features to achieve complete segmentation of each optical tissue.On this basis,the fusion features of different tissues are compressed by DNN network,and then the GA-SVM-based algorithm was used to classify the compressed features of different tissues,and finally the purpose of optical tissue recognition is achieved.The main research contents of this paper are as follows:(1)Collect and organize optical tissue images,and propose an edge morphology extension method for the segmentation of the main body.Through the contour extraction of the main body of the image after adaptive threshold segmentation and morphological filtering,the main body of the color of the microscopic image can be extracted to the greatest extent.(2)A multi-feature fusion clustering method based on superpixels was proposed for the optical tissue segmentation of the main body.First,perform superpixel segmentation on the main body,then extract color and texture features for each superpixel,and perform fusion dimension reduction on the extracted multi-features;the label-free segmentation of different tissues in optical tissue images is achieved by performing an improved adaptive Kmeans clustering on the multi-features after fusion dimension reduction.(3)Extract the color features of the data set tissue image and the texture features in multiple directions in the grayscale co-occurrence matrix for each optical tissue,then use the DNN network to compress the selected features or feature combinations.Finally,the compressed features are trained using the GA-SVM model to achieve the classification of different optical tissues.(4)A coke optical tissue component analysis system platform was built,which integrates and visualizes the extraction algorithm of the main part of the optical tissue,the tissue segmentation algorithm of the main body,and the identification algorithm of each tissue component,and realizes the proportion of each tissue component in a single or single batch of optical tissue images.The experimental results show that the segmentation method proposed in this paper can segment different optical tissues to a greater extent;the recognition accuracy is relatively high,and it has the advantages of fast segmentation and recognition speed,high accuracy,and adaptability.It has the advantages of strong performance,etc.,which can better meet the actual industrial application.