A Skin Cancer Classification Method Based on Discrete Wavelet Down-Sampling Feature Reconstruction

Aiming at the problems of feature information loss during down-sampling, insufficient characterization ability and low utilization channel in skin cancer diagnosis melanoma, a pathological mirror classification method based on discrete wavelet down-sampling reconstruction is proposed this paper. The given first, multichannel attention mechanism introduced to realize high-frequency low-frequency components, which reduces due effectively utilizes information. A model given, using combination depth-separable convolution 3×3 standard as input backbone ensure perceptual field while reducing number parameters; residual module optimized Hard-Swish activation function enhance representation capability model. network weight parameters are initialized ImageNet transfer learning then debugged augmentation HAM10000 dataset. experimental results show that accuracy for significantly improved, reaching 95.84%. Compared with existing methods, not only has higher but also accelerates speed enhances noise immunity. paper provides new some practical value.