Research On Classification Of Skin Pathological Images Based On Multi-decision Fusion And Attention Mechanism

With the development of artificial intelligence technology,deep learning has made a great breakthrough in the field of image classification.The key to its success is to have large-scale training data.However,large labeled medical data sets are difficult to obtain.How to use a small number of medical data samples to make the classification model quickly converge has become a research hotspot.At present,the problem of sample shortage is mostly improved by data enhancement,which ignores the impact of the amount of information carried by pathological images on the classification performance,resulting in model training not only requiring massive data,but also consuming more time.Therefore,it is of great significance to study how to obtain faster and better pathological image classification effects through limited medical data.In this paper,skin pathological image is taken as the research object,and a multidecision fusion method based on active learning is proposed.At the same time,in order to improve the classification accuracy of skin pathological images,a classification method incorporating the attention mechanism is further proposed.The main contents of this paper are as follows:(1)The algorithm of hair removal based on morphological black hat operation.Aiming at the problem that the skin surface hair in the skin pathological image has a great influence on feature extraction,the 8-bit grayscale image is obtained by converting the image color space of the skin pathological images,and the effects of each filter in feature conversion are compared.On the basis of choosing the appropriate filter for hair removal,the black hat operation is further used for morphological filtering of pathological images,and combine the threshold processing to get the skin surface hair contour image.At the same time,the fast marching algorithm is used to repair the image,and a hair removal algorithm based on morphological black hat operation is proposed.The experimental results show that this method can effectively remove the hair noise while ensuring the focus characteristics of pathological images.(2)Research on multi-decision fusion method based on active learning.Aiming at the problem that the traditional supervised learning training process is difficult to converge due to the unbalanced distribution of samples in pathological image data sets,a multi-decision fusion method based on active learning is proposed.According to the complexity score of image information obtained by the multi-decision evaluation method,images with higher priority are selected to train the classification model.On this basis,less samples are used to complete the rapid convergence of the model and realize the optimization of the training method based on active learning.(3)Research on classification of skin pathological images based on the mechanism of attention fusion.Aiming at the problem that the classification model has weak ability of feature expression on skin pathological images,a feature extraction method incorporating attention mechanism is proposed.Feature mapping under the attention mechanism and adaptive learning of features are performed in two different dimensions of space and channel.And design the attention mechanism fusion strategy,select the optimal fusion scheme through the comparative analysis of the feature extraction effect,improve the ability to represent the specific area of the pathological image,and improve the classification effect of the model.