Enhanced ResNet-18 for Fish Disease Detection: Integrating SE-Net and CBAM to Advance Aquaculture Through Artificial Intelligence

Gohar Ali; Niranjan KR; Miguel Villagómez-Galindo; Ajay Kumar; Saiprasad Potharaju; Jagadeesh B N; Siddharth Arora

doi:10.37965/jait.2026.1156

Enhanced ResNet-18 for Fish Disease Detection: Integrating SE-Net and CBAM to Advance Aquaculture Through Artificial Intelligence

Authors

Gohar Ali College of Information Technology, Kingdom University, Bahrain
Niranjan KR Department of Medical Electronics Engineering, B. M. S. College of Engineering, Bangalore, Karnataka, India
Miguel Villagómez-Galindo Universidad Michoacana de San Nicolás de Hidalgo, Facultad de Ingeniería Mecánica, Morelia, México, Santiago Tapia 403, 58000
Ajay Kumar Bharati Vidyapeeth (Deemed to be University) Institute of Management and Research, New Delhi, India
Saiprasad Potharaju Symbiosis Institute of Technology, Symbiosis International (Deemed University), Pune, India
Jagadeesh B N Department: CSE (AI & ML), Global Academy of Technology, Rajarajeshwari Nagar, Bangalore-560096, India
Siddharth Arora Department of Computer Science & Engineering, University Institute of Engineering, Chandigarh University, Mohali, Punjab, India

DOI:

https://doi.org/10.37965/jait.2026.1156

Keywords:

aquaculture automation, attention mechanisms (SE, CBAM), deep learning models, machine learning in aquaculture, ResNet-18

Abstract

The growing demand for fish protein and challenges in aquaculture disease management highlight the need for advanced technologies. This research proposes an automatic fish disease detection framework using ResNet-18 (Residual Network with 18 layers) enhanced with attention mechanisms like Squeeze-and-Excitation (SE) and Convolutional Block Attention Module (CBAM). Traditional convolutional neural networks (CNNs), though effective, often lack accuracy due to insufficient focus on critical features. By incorporating SE and CBAM into ResNet-18, feature extraction and classification accuracy are significantly improved. Evaluation on a seven-class fish disease dataset shows test accuracies of 63.83% for CNN, 67.72% for ResNet-18, 94.67% for SE + ResNet-18, and 96.62% for CBAM + ResNet-18. The CBAM + ResNet-18 model demonstrated superior performance, effectively addressing generalization and overfitting issues. This study connects recent deep learning advances with practical aquaculture needs. It provides a scalable and accurate framework for fish disease detection. Integrating attention mechanisms into deep learning models offers promise for sustainable aquaculture practices