Revolutionizing Acne Diagnosis With Hybrid Deep Learning Model Integrating CBAM, and Capsule Network
Krishna Veni Paluri, Ashish Gupta
Abstract
Acne is the eighth most prevalent global health concern and affects around 9.4% of the global population. The accurate classification of diverse acne categories is crucial for promptly creating efficient treatment plans. Although various deep learning (DL) models are available in the literature, achieving high classification accuracy remains challenging. This study introduces a unique hybrid DL model for acne classification, incorporating the attention mechanism from the convolutional block attention module (CBAM) for optimal feature selection, VGG16, and CapsNet. The proposed hybrid model autonomously captures significant features from VGG16 using CBAM and subsequently utilizes CapsNet to categorize these features. The efficacy of our proposed model is assessed on various datasets, including both augmented and un-augmented datasets. The outcomes are very remarkable, with high levels of precision (100%), F1 score (99%), recall (100%), accuracy (99%), specificity (100%), and kappa score (97.87%) attained throughout all the datasets.