Marine Image Classification Using EfficientNetV2, Focal Loss, and XAI
DOI:
10.59888/ajosh.v1i03.787Published:
2025-03-25Downloads
Abstract
Background: Marine ecosystems require fast and accurate monitoring systems, yet the analysis of underwater images faces complex environmental conditions and class imbalance.
Objective: This study proposes the integration of an EfficientNetV2-based deep learning classifier with Adaptive Underwater Image Enhancement (AUIE), focal loss, and Explainable Artificial Intelligence (XAI) to improve marine image classification.
Method: The dataset consists of 22 classes of marine organisms divided into training, validation, and test sets with a ratio of 80:10:10. AUIE combines Contrast Limited Adaptive Histogram Equalization on the lightness channel with colour balance correction, and focal loss with ? = 2.0 replaces standard cross-entropy. This study used a quantitative experimental approach with four experimental scenarios.
Results: An ablation study of four scenarios showed that the baseline reached 79.47% accuracy, whereas the combination of AUIE and focal loss produced a validation loss of 0.5503 and an accuracy of 76.32%, recovering part of the accuracy lost when AUIE was used with cross-entropy. Confusion matrix inspection indicated fewer misclassifications of camouflaged minority classes, and Grad-CAM showed attention concentrated on the morphological features of the organisms after enhancement.
Conclusion: The study contributes to robust and transparent marine biodiversity monitoring systems, while longer training and per-class evaluation are still required.
Keywords:
marine images class imbalance EfficientNetV2 focal loss XAILicense
Copyright (c) 2022 Wahyu Bontot
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
Authors who publish with this journal agree to the following terms:
- Authors retain copyright and grant the journal right of first publication with the work simultaneously licensed under a Creative Commons Attribution-ShareAlike 4.0 International. that allows others to share the work with an acknowledgement of the work's authorship and initial publication in this journal.
- Authors are able to enter into separate, additional contractual arrangements for the non-exclusive distribution of the journal's published version of the work (e.g., post it to an institutional repository or publish it in a book), with an acknowledgement of its initial publication in this journal.
- Authors are permitted and encouraged to post their work online (e.g., in institutional repositories or on their website) prior to and during the submission process, as it can lead to productive exchanges, as well as earlier and greater citation of published work.
