Research On Image Extraction And Segmentation Method Of Al-Si Alloy Literature

Material genome advocates to give play to the role of material big data and uses machine learning to transform the culture of material research and development.Al-Si alloys(Al-Si)are widely used in the automotive,aerospace,and electronics industries due to their high specific strength,good wear resistance,as well as small coefficient of thermal expansion.The properties of Al-Si alloy are mainly determined by the shape and size of the primary Si phase in the alloy microstructure.At present,the acquisition of microstructure images is relatively complicated due to the high cost and time cost of the experiment,but it exists in the published literature.Therefore,in this paper,the deep learning method was adopted to extract illustrations and titles from the literature of AlSi alloys,screen out the microstructure images,and segment the shape of the primary crystal Si phase in the images,so as to acquire a large amount of relevant data,which is conducive to further optimizing the alloy properties.According to the characteristics of illustrations and titles in the scientific research literature,in this paper,an efficient FCENET(Figure Caption Extract Net)was proposed.Based on the Blend Mask,a horizontal and vertical attention module was added to the FCENET,and then the Blend Mask detection head was divided into two branches: illustration detection and title detection.Meanwhile,a multi-scale allocation strategy of length-width ratio was proposed to improve the final detection accuracy and speed.Finally,the performance of the model was proved by experiments.Compared with other existing models,the 8)of FCENet was improved by more than 8%.To avoid the interference of artificial subjective factors,in this paper,a new segmentation model CAU-Net(Class Attention U-Net)was proposed based on the existing deep learning model to segment the shape of the primary Si phase in the microstructure image of Al-Si alloy.This model is based on U-Net and combines with the attention mechanism,so it can effectively segment the microstructure.Compared with U-Net,the indexes of CAU-NET,including ?(8(8?were improved by about 10%.Moreover,compared with other semantic segmentation models,the proposed method was more suitable for processing the microstructure images of Al-Si alloy.In this paper,FCENet was used to collect and screen the related image data of AlSi alloy,and then the CAU-NET model was adopted to effectively segment the microstructure image of Al-Si alloy,which was conducive to the subsequent analysis and performance evaluation of microstructure image.