MLANet: multilayer attention network for airport runway segmentation in remote sensing images

As the core infrastructure in the aviation sector, airports act as a vital link connecting air traffic and ground transportation. The effective segmentation of airport runways and taxiways is essential for ensuring the smooth operation of airports and the efficiency and safety of aircraft operations during takeoff and landing. However, most existing research focuses on runway segmentation but overlooks the segmentation of taxiways adjacent to runways; moreover, efficient and accurate segmentation under multi-perspective conditions has not been adequately addressed. Consequently, we present Mlanet: a novel semantic segmentation architecture for remote sensing images of airport runways. The model distinguishes runways from taxiways with remarkable precision thanks to its proprietary multi-layer attention module. Mlanet additionally utilizes a three-branch large kernel attention module to broaden the receptive field, thereby further enhancing its feature extraction capacity. Experimental results demonstrate that Mlanet achieves superior performance in key metrics including mIOU, mPA, and Accuracy compared to other models.