基于集成学习与生成对抗网络的皮肤镜图像分类方法

皮肤镜是辅助皮肤科医生诊断的有效途径,但是人工分类高度依赖医生的临床经验,并且皮肤镜图像本身的复杂性给分类提出了巨大的挑战.为了解决皮肤镜图像分类问题,基于集成学习提出了一种集成投票块的皮肤镜图像分类方法.首先,针对ISIC 2019提供的皮肤镜图像进行预处理,为了缓解数据集较少且分布不均的问题,使用生成对抗网络和旋转...

Dermoscopy image classification method based on ensemble learning and generative adversarial networks

Dermoscopy is an effective way to assist dermatologists'' diagnosis, but artificial classification is highly dependent on the clinical experience of the doctor, and the complexity of the dermoscopy image itself poses a huge challenge to classification. In order to solve the problem of dermoscopy image classification, a dermoscopy image classification method based on voting block integration is proposed based on ensemble learning. First, the dermoscopy images provided by ISIC 2019 were pre-processed. In order to alleviate the problem of less data sets and uneven distribution, the data set is augmented by generating adversarial networks and rotating images. Then, based on the idea of transfer learning, multiple convolutional neural networks are trained, and multiple convolutional neural networks with better classification results are selected to form voting blocks. Through voting block integration, the dermoscopy image is classified. The experimental results show that the accuracy, sensitivity, and specificity of the method can reach 0.925, 0.563, and 0.927, respectively. Compared with a single convolutional neural network model, each evaluation criterion is improved, which provides an effective solution for dermoscopy image classification.