In the field of medical image analysis, traditional deep learning models commonly face the challenge of accurately localizing and extracting key features from lesion regions when performing medical image classification tasks. This bottleneck is particularly pronounced in the classification of skin wart images, where subtle morphological differences between various types of warts impose higher demands on the model’s feature extraction capabilities. To address this challenge, this paper proposes an enhanced ResNet18 model based on the Convolutional Block Attention Module (CBAM). By integrating CBAM after each residual block in ResNet18, the model can adaptively focus on lesion regions in skin wart images, significantly improving the precision and specificity of feature extraction. Experimental results on a dataset containing four types of warts (common_wart, flat_wart, normal, and plantar_wart) demonstrate substantial improvements over the baseline ResNet18 model. Key metrics such as classification accuracy, precision, recall, and F1-score all exhibit superior performance, with accuracy increasing from 93.16% to 94.98%—a 1.82% improvement. These findings strongly indicate that the enhanced model possesses stronger recognition capabilities and generalization performance in skin wart image classification tasks. This work provides an effective solution for the automated diagnosis of skin lesions and holds significant clinical value.