DMAT: An End-to-End Framework for Joint Atmospheric Turbulence Mitigation and Object Detection

The recent introduction of DMAT, an end-to-end framework for joint atmospheric turbulence mitigation and object detection, marks a significant advancement in the field of surveillance imagery. Atmospheric turbulence has long been a barrier to clear and accurate object detection, as it degrades image quality and complicates classification tasks. Traditional deep learning methods have struggled with these distortions, leading to the development of DMAT, which utilizes a 3D Mamba-based structure to effectively manage spatio-temporal displacements and blurring. By integrating knowledge from both the AT mitigator and the object detector, DMAT not only enhances visualization but also improves detection accuracy. The framework has demonstrated a performance improvement of up to 15% on datasets affected by turbulence, surpassing existing state-of-the-art systems. This innovation is vital for enhancing surveillance capabilities, particularly in environments where atmospheric conditions can hind…