遥感
计算机科学
分割
人工智能
计算机视觉
图像分割
遥感应用
模式识别(心理学)
度量(数据仓库)
图像处理
作者
Qinglin Zhao,Jinming Li
出处
期刊:
[SPIE]
日期:2026-04-09
卷期号:: 41-41
摘要
High-resolution remote sensing images are characterized by significant inter-class scale variations, complex backgrounds, and occluded objects. These inherent challenges often lead to insufficient feature discriminability between different categories and high heterogeneity within the same category, thereby compromising the accuracy of semantic segmentation. To address these issues, this paper proposes a novel semantic segmentation network termed the Multi-Scale Frequency Enhancement Network (MSFENet). In the encoding stage, the network integrates Convolutional Neural Networks (CNNs) with a self-attention mechanism, effectively capturing both local detailed information and global contextual dependencies. In the decoding stage, a fast Fourier transform (FFT) is introduced to establish a dual-domain processing framework that synergistically utilizes both spatial and frequency domain information. This approach allows for more effective utilization of the multi-scale features extracted by the encoder and effectively suppresses interference from complex backgrounds. Extensive experiments on two public benchmarks demonstrate the superiority of our method. On the Vaihingen dataset, MSFENet achieves a macro-F1 (mF1) of 91.34%, an Overall Accuracy (OA) of 93.57%, and a mean Intersection over Union (mIoU) of 84.34%. On the Potsdam dataset, it obtains scores of 92.20%, 90.98%, and 85.75% on the respective metrics. These results collectively verify that the proposed model exhibits strong detail perception capabilities and possesses high discriminative power for segmenting small objects and handling objects with high inter-class similarity. The dual-domain design offers a robust and effective solution for processing high-resolution remote sensing imagery.
科研通智能强力驱动
Strongly Powered by AbleSci AI