Cross-Domain Generalization with Noise-Augmented Loss Function for Burn Area Detection

Abstract In recent years, domain adaptation has emerged as a critical challenge in the context of burn region identification, particularly when transferring models trained on source domains to unseen target domains. One key difficulty in this area arises from domain-specific disparities in image features, which often lead to suboptimal performance when directly applying models across different datasets. In this work, we address the problem of domain adaptation for burn region recognition by introducing a novel approach that integrates feature alignment techniques to reduce the domain gap. Specifically, we propose a method that leverages adversarial training to align the feature distributions between source and target domains, thereby improving the model’s ability to generalize to new, unseen data. Our approach encourages the model to learn domain-invariant features while maintaining discriminative power for burn region identification. Extensive experiments on benchmark datasets demonstrate the effectiveness of our method, showing improved performance in identifying burn regions across varied domain conditions. Our findings highlight the potential of domain adaptation techniques in enhancing the robustness and accuracy of burn region detection models, making them more applicable in real-world scenarios with diverse input data