Research And Implementation Of Metallographic Image Analysis System For Aluminum Alloy

Compared with other alloys,aluminum and its alloys have great advantages in manufacturing cost,mechanical properties and sustainable development.It is very important for the safety of mechanical equipment to judge the performance of metal materials by studying the microstructure of its metallographic image.Therefore,in this paper,the typical semi continuous casting Al-12.7Si-0.7Mg alloy is taken as an example to study the metallographic image preprocessing,feature extraction and microstructure segmentation method in depth.On this basis,the metallographic image analysis system is implemented.The specific work is shown below:Combined with the characteristics of aluminum alloy,the metallographic image of aluminum alloy was proceeded targeted preprocessing and the effective metallographic features were extracted.In the aspect of preprocessing,firstly,the homomorphic filtering algorithm is used to correct the shadow area of the image,then bilateral filtering is used to suppress the noise of image edge,and finally use CLAHE to enhanced the contrast of the tissue.In the aspect of feature extraction method,five effective features are extracted from the pre-processing metallographic structure,including gray level inversion,bottom-hat transformation,Weber’s local descriptor,rotation invariant LBP and Gaussian matched filter.The extracted features are fused in a simple series way and the feature vector of the metallographic structure is constructed to realize the accurate representation of the metallographic structure and lay the foundation for the subsequent tissue segmentation.Based on the constructed metallographic features,an unsupervised metallographic image segmentation algorithm based on SLIC and GMM is proposed.Firstly,SLIC super-pixel is used to coarse segment the image,and the extracted metallographic structure features are extended with super-pixel information(including Gaussian statistical features and topological features),so as to construct the feature matrix representing the super-pixel image.On this basis,using Mahalanobis Distance as the distance measurement standard,the super-pixel of image are fused by GMM clustering,so that the metallographic structure can be segmented without relying on the labeled metallographic data.The validity of the method is verified on the collected metallographic image set by taking the expert mark image as the gold standard.A clustering algorithm based on subset partition model is designed.Firstly,several new datasets are generated by resampling method,and initial clustering results with diversity are generated by SLIC and improved GMM.Then,the problem of clustering integration is transformed into the problem of solving the improved model of subset division.By introducing the similarity measurement function and discriminant constraints between sub clusters,the optimal clustering result is selected,which effectively improves the precision of microstructure segmentation of metallographic image.Design and implement the "aluminum alloy metallographic image analysis system" which integrates the functions of metallographic image preprocessing,microstructure segmentation,parameter statistics and screening,and can provide accurate analysis data for aluminum alloy metallographic analysis and improve the analysis efficiency of professionals.