Hybrid Efficientnetv2b0 with Attention Mechanism and Multi-Layer Perceptron for Optimized Multi-Class Skin Wound Classification

Hartini Damanik (Information Systems Study Program, STIKOM Tunas Bangsa, Indonesia)
Anjar Wanto (Master’s Program in Informatics, STIKOM Tunas Bangsa, Indonesia)
Surya Darma (Information Systems Study Program, STIKOM Tunas Bangsa, Indonesia)

Jurnal Teknik Informatika (JUTIF)
Vol. 7 No. 4 (2026): JUTIF Volume 7, Number 4, August 2026

Skin wound classification remains a challenging task in medical image analysis due to variations in visual appearance and limitations of conventional diagnostic methods. This study aims to develop an optimized deep learning framework for multi-class skin wound classification by integrating EfficientNetV2B0, Convolutional Block Attention Module (CBAM), and Multi-Layer Perceptron (MLP). EfficientNetV2B0 is used for feature extraction, while CBAM enhances feature representation by emphasizing relevant channel and spatial information, and MLP strengthens the classification process in multi-class scenarios. Experimental results show that the proposed hybrid model achieves an accuracy of 95% and consistently outperforms the baseline EfficientNetV2B0 architecture. The findings demonstrate that the integration of attention mechanisms and MLP significantly improves classification performance. This study contributes to the development of automated skin wound analysis systems and provides insights into hybrid deep learning architectures for medical image classification.