Detective Networks: Enhancing Disaster Recognition in Images Through Attention Shifting using Optimal Masking

Narongthat Thanyawet, Photchara Ratsamee, Yuki Uranishi, Haruo Takemura; Proceedings of the Winter Conference on Applications of Computer Vision (WACV), 2025, pp. 8723-8732

Aerial investigation is used for surveying damage and identifying post-disaster events through imagery data. However the challenge lies in detecting disaster-related areas within aerial or shipborne images as these can appear as minor regions making recognition difficult. To address this challenge we introduce the Detective Network (DeNet) designed to optimally mask images thereby shifting the attention of machine learning models towards these small yet crucial regions. Utilizing the concepts of patch and anchor box DeNet incorporates a masking candidate layer and a masking layer to facilitate optimal masking. Our experimental findings are compelling; by preprocessing images with DeNet before analysis using an image captioning model we achieved a remarkable accuracy of 92.91% in landslide detection from side-view image captions and 87.50% for shipborne view detection. The result demonstrates the efficacy of DeNet in enhancing the recognition of disaster-related areas in challenging imaging conditions.