摘要
皮肤覆盖肌肉、骨骼和身体的每个部分,是人体中最大的器官。由于其暴露于外界,所以感染更容易发生在皮肤上。皮肤病作为一种常见疾病,利用计算机技术对其进行辅助诊断,有助于减轻医生负担。针对常规卷积神经网络应用于皮肤病图像分类时由于不同种皮肤病图像之间的类间相似性以及同种皮肤病图像之间具有类内差异性导致分类困难的问题,提出一种改进双线性特征融合模型。使用经过剪枝的Inception-ResNet-v1和v2版本作为特征提取器并行提取图像特征,对特征进行双线性融合,获取更多阶数的特征信息可以提高模型对图像细节的敏感度。然后添加额外的软注意力模块,通过加权和的方式进行过滤或者加强,给图像每个位置给予不同的权重以达到对模型的加强效果。在skin-cancer-classesisic数据集上的7种皮肤病图像上进行训练,与S-CNN、MobileNet和Incremental CNN的对比证明了该模型的有效性,在Precision、Recall和F1-Score指标上该模型均为最优。
Skin covers muscles,bones and every part of the body,which is the largest organ in the human body.Because it is exposed to the outside world,the infection is more likely to occur on the skin.Skin disease is a common disease.Using computer technology to diagnose it will help to reduce the burden of doctors.In view of the difficulty of classification caused by the inter class similarity between different skin disease images and the intra class difference between the same skin disease images when the conventional convolutional neural network is applied to the classification of skin disease images,an improved bilinear feature fusion model is proposed.Using the pruned version of Inception-ResNet-v1 and v2 as the feature extractor to extract image features in parallel,bilinear fusion of features and obtain more order feature information can improve the sensitivity of the model to image details.Then add an additional soft attention module to filter or strengthen the image by weighted sum,and give different weights to each position of the image to achieve the effect of strengthening the model.Training on seven kinds of skin disease images on skin cancer classic dataset,the comparison with S-CNN,MobileNet and Incremental CNN proves the effectiveness of this model,which is the best in Precision,Recall and F1-Score.
作者
黄志伦
刘俊
郑萌
HUANG Zhi-lun;LIU Jun;ZHENG Meng(School of Computer Science and Technology,Wuhan University of Science and Technology,Wuhan 430065,China;Hubei Province Key Laboratory of Intelligent Information Processing and Real-time Industrial System(Wuhan University of Science and Technology),Wuhan 430065,China;School of Computer Science and Technology,Wuhan University of Technology,Wuhan 430070,China)
出处
《计算机技术与发展》
2023年第2期161-166,共6页
Computer Technology and Development
基金
教育部科技基础工作专项项目(2014FY110900)。
关键词
皮肤病
深度学习
双线性特征融合
注意力机制
图像分类
skin disease
deep learning
bilinear feature fusion
attention mechanism
image classification
