Hybrid Tunnel Entrance Detection via Multiple Morphological Component-Based Texture Enhancement

The utilization of remote sensing imagery for identifying tunnel entrances plays a significant role in monitoring and assessing tunnel conditions and ensuring public safety. Prompt detection of changes or damages enables timely implementation of appropriate maintenance and emergency measures. However, due to the relatively small size of tunnel entrances and their limited informational content, existing object detection techniques face challenges in achieving precise identification. To address this challenge, this paper proposes a Texture-Enhanced Detection Model (TEDM). Firstly, by applying multiple morphological component analysis for image decomposition, various texture information including content, coarseness, contrast, and directionality are extracted to obtain richer information from the image. Subsequently, images enhanced with detailed textural information are input into the object detection framework to achieve precise identification of tunnel entrances. Finally, a voting strategy is employed for postprocessing, aimed at improving the efficacy of tunnel entrance detection. Empirical analyses demonstrate that TEDM achieved an [email protected] of 97.5% when evaluated against a newly created tunnel dataset, outperforming existing object detection algorithms.