Skin-LiteNet: A Lightweight ConvolutionalModule for Skin Disease Image Classification

Skin cancer is one of the most common malignancies worldwide. The deep learning algorithms, such as convolutional neural networks (CNNs), plays a crucialrole in the automation of skin image analysis. However, the high complexityand large parameter sizes of CNNs limit their deployment on mobile or embedded systems. To address this problem, we propose Skin-LiteNet, a lightweightCNN model designed for efficient skin lesion classification. To enhance the feature extraction, the model use a multi branch DBConv module. Moreover, themodel introduces an ECSA attention mechanism to focus on critical lesion regionswhile maintaining low computational overhead. Experiments on the HAM10000dataset show that Skin-LiteNet achieves an accuracy of 98.97% and an F1 scoreof 98.89%, using only 284.44K parameters and 48.11M Flops. Compared withexisting models such as ResNet, MobileNetV2, ConvNeXt, and ShuffleNetV2,Skin-LiteNet offers better accuracy with reducing resource requirements significantly. These results indicate a strong potential for Skin-LiteNet in real-time andresource-constrained clinical applications.