International Conference on Computer and Knowledge Engineering

Home / 14th International Conference on Computer and Knowledge Engineering

Adaptive Multi-Scale Attentional Network for Semantic Segmentation of Remote Sensing Images

Authors :

Melika Zare¹ Sattar Hashemi²

1- Comp. Sci. & Eng. Dept, Shiraz University, Iran 2- Comp. Sci. & Eng. Dept, Shiraz University, Iran

Keywords :

semantic segmentation،remote sensing،residual structure،attention mechanism،convolutional neural network

Abstract :

High-resolution remote sensing image segmentation is challenging due to the complex and diverse nature of ground objects, varying scales, and the presence of small targets that are difficult to detect. Traditional methods often fall short in capturing multi-scale contextual information and accurately segmenting small objects amidst large, cluttered backgrounds. In this paper, we propose the Adaptive Multi-Scale Attentional Network (AMSANet), which addresses these challenges by leveraging a Hierarchical Multi-Scale Dilated Convolution Module and an Efficient Residual Atrous Spatial Pyramid Pooling (ER-ASPP) module to enhance feature extraction and refinement. The AMSANet architecture integrates these advanced modules to effectively capture and merge features at multiple scales, employing coordinate attention mechanisms for precise final refinement. We validate our approach on the ISPRS Vaihingen and Potsdam datasets, demonstrating that AMSANet significantly outperforms state-of-the-art methods in terms of mean intersection over union (mIoU), overall accuracy (OA), and F1 score. Our experimental results underscore the efficacy of AMSANet in improving segmentation accuracy and robustness, particularly in managing high-resolution remote sensing images with intricate details and heterogeneous landscapes.