🤖 AI Summary
To address the high computational complexity and redundancy inherent in self-attention mechanisms for remote sensing image semantic segmentation, this paper proposes a Threshold Attention Mechanism (TAM) and a lightweight, efficient network termed TANet. TANet comprises two key components: an Attention Feature Enhancement Module (AFEM) for shallow-layer global feature enhancement, and a Threshold Attention Pyramid Pooling module (TAPP) for deep-layer multi-scale contextual modeling. TAM dynamically filters salient attention regions via adaptive thresholding, preserving long-range dependency modeling while substantially reducing computational overhead. Evaluated on the ISPRS Vaihingen and Potsdam benchmarks, TANet achieves competitive accuracy against state-of-the-art methods, with ~22% faster inference speed and 40% lower GPU memory consumption—demonstrating a favorable balance among accuracy, efficiency, and deployment practicality.
📝 Abstract
Semantic segmentation of remote sensing images is essential for various applications, including vegetation monitoring, disaster management, and urban planning. Previous studies have demonstrated that the self-attention mechanism (SA) is an effective approach for designing segmentation networks that can capture long-range pixel dependencies. SA enables the network to model the global dependencies between the input features, resulting in improved segmentation outcomes. However, the high density of attentional feature maps used in this mechanism causes exponential increases in computational complexity. In addition, it introduces redundant information that negatively impacts the feature representation. Inspired by traditional threshold segmentation algorithms, we propose a novel threshold attention mechanism (TAM). This mechanism significantly reduces computational effort while also better modeling the correlation between different regions of the feature map. Based on TAM, we present a threshold attention network (TANet) for semantic segmentation. The TANet consists of an attentional feature enhancement module (AFEM) for global feature enhancement of shallow features and a threshold attention pyramid pooling (TAPP) module for acquiring feature information at different scales for deep features. We have conducted extensive experiments on the international society for photogrammetry and remote sensing (ISPRS) Vaihingen and Potsdam datasets. The results demonstrate the validity and superiority of our proposed TANet compared with most state-of-the-art models.