Zero-shot multi-class semantic segmentation of remote sensing images using SAM 2 with prior database information

Frequency and Edge-Guided Segment Anything Model for Remote Sensing Image Semantic Segmentation

Remote sensing image semantic segmentation (RSISS) has attracted significant attention due to the growing demand for fine-grained land cover information. The Segment Anything Model (SAM), proposed as a foundation vision model, offers strong segmentation performance and generalization capabilities for RSISS tasks. However, existing SAM-based approaches face two limitations: (1) Insufficient adaptation of SAM's features to the diverse characteristics of land cover types. (2) Semantic ambiguity at object boundaries, which hinders accurate delineation. To address these limitations, we propose Frequency and Edge-guided SAM (FE-SAM), a scalable and efficient framework for RSISS. Specifically, we introduce a Frequency-Modulated Adapter (FMA) that adaptively decomposes and modulates frequency-domain features based on the input data. It selectively enhances informative high- and low-frequency components corresponding to different land cover types. Furthermore, to improve SAM's ability to capture fine-grained details, we design EGRefiner, which integrates multi-scale edge-enhanced information extracted from the input image. Extensive experiments on three benchmark datasets demonstrate that FE-SAM outperforms state-of-the-art methods. The source codes are available at: https://github.com/oucailab/FE-SAM.

Multi-Model Evaluation of Semantic Segmentation Techniques for Building Footprint Extraction

In the present generation of increasing geospatial data, accurate and automated extraction of building footprints from high-resolution aerial and satellite imagery has become crucial for various applications such as urban planning, infrastructure development, disaster management, and GIS database maintenance, as manual tracing is time-consuming and unstable for large-scale mapping. This study compares conventional image processing techniques such as thresholding, edge detection, morphological operations through a machine learning approach using Random Forest (RF), and deep learning-based semantic segmentation models, namely U-Net and DeepLabV3+, along with the Segment Anything Model (SAM) using a pre-trained prompt-based setup. All methods are tested on the same set of data, and a standardized data preprocessing is performed for fair comparison. The overall results indicate that the application of DeepLabV3+ is best, with an IoU of 82% and an F1 score of 90%. U-Net achieves second high IoU and F1 scores of 74% and 84% respectively, while Random Forest shows a high IoU of 60% and an F1-score of 72%. SAM has the lowest scores with an IoU of 50% and an F1 score of 51%.

Land Cover Classification of Multi-Source Airborne Data using Conventional and Deep-Learning-Based Unsupervised Domain Adaptation

Abstract. For an increasing number of applications, land cover maps can be generated from remote sensing imagery using conventional and deep-learning-based semantic segmentation models. Relying on a large pool of training data, the networks struggle with the spatial-temporal-spectral heterogeneity in the complex and diverse remote sensing imageries, leading to a significant number of errors in the model predictions. This paper presents a workflow comprising domain adaptation and classification. In particular, we analyze two domain adaptation techniques: First, a conventional histogram-matching method, which has turned out to be a surprisingly fast and reliable tool in a previous study, and second, a CycleGAN, which we applied both in its standard form and with the perceptual loss, thereby penalizing style inconsistencies on deeper layers. By applying the workflow to three remote sensing datasets and six directions of domain adaptation, we show that there is “no free lunch” in the sense that all domain adaptation methods have their advantages. Depending on the dataset, classification method, and especially on the availability of 3D data, the performance gap can be reduced to up to 1.5% of the mean F1 score, demonstrating the soundness of the proposed method.

A plug and play attention block for accurate multi scenario remote sensing image segmentation.

In this paper, we propose SegRSNet for the task of feature extraction from remote sensing images. To address the challenges of complexity and variability of target objects, obvious occlusion effects and rich multilevel semantic information in high-resolution remote sensing images, we design a plug-and-play network structure, SegRS Block, which consists of a series of key components responsible for: efficiently aggregating channel features, accurately capturing spatial location information, fusing feature maps across layers and deepening the fine modeling of channel dimensions. The experimental results show that SegRSNet achieves state-of-the-art (SOTA) performance on multiple benchmark datasets for both building and road feature extraction, which not only outperforms the existing best methods under the same parameter scale, but also shows high adaptability and accuracy for all kinds of remote sensing image analysis tasks. In addition, although the Transformer architecture has advantages in dealing with remote dependencies, it usually requires large-scale training data and a large number of parameters. In contrast, our study shows that the combination of convolutional neural network and a specially designed attention module can reduce the training cost while improving the receptive field to effectively deal with the remote sensing image segmentation problem, thus realizing the detailed and accurate recognition and parsing of multi-scenario remote sensing data.

Rcl-sam: a parameter-efficient segment anything model for high-resolution remote sensing cultivated land extraction

Semantic Segmentation of Remote Sensing Images Based on RS3mamba and Wavelet Transform

The semantic interpretation of remote sensing imagery through segmentation has become indispensable for a wide range of applications, including resource exploration, environmental assessment, and land-use analysis. Yet, accurate parsing of such images remains challenging because complex object boundaries and large scale differences often weaken the ability of conventional Convolutional Neural Network (CNN)-based methods to preserve local details. In response, this study constructs a segmentation framework that couples wavelet convolution with the Mamba architecture. To strengthen feature learning in the intermediate stages, an Auxiliary Segmentation Module (ASM) is employed to provide additional supervisory guidance, which supports optimization and encourages the representation of subtle semantic details. Wavelet-transform convolution is also introduced into the downsampling path, enabling spatial cues and frequency-related information to be exploited in a more coordinated manner for finer boundary and texture modeling. Experiments on public remote sensing datasets and mining area imagery further confirm the effectiveness of the method. Compared with several existing segmentation approaches, the proposed model delivers better overall performance in mIoU, F1-score, and recognition accuracy, particularly in scenes where multiple land-cover categories are heavily interlaced. Moreover, these gains are obtained with relatively low model complexity, suggesting good potential for practical deployment in land monitoring and ecological management.