Remote sensing scene classification (RSSC) is essential in Earth observation, with applications in land use, environmental status, urban development, and disaster risk assessment. However, redundant background interference, varying feature scales, and high inter-class similarity in remote sensing images present significant challenges for RSSC. To address these challenges, this paper proposes a novel hierarchical graph-enhanced transformer network (HGTNet) for RSSC. Initially, we introduce a dual attention (DA) module, which extracts key feature information from both the channel and spatial domains, effectively suppressing background noise. Subsequently, we meticulously design a three-stage hierarchical transformer extractor, incorporating a DA module at the bottleneck of each stage to facilitate information exchange between different stages, in conjunction with the Swin transformer block (STB) module to capture multi-scale global visual information. Moreover, we develop a fine-grained graph neural network (GNN) extractor that constructs the spatial topological relationships of pixel-level scene images, thereby aiding in the discrimination of similar complex scene categories. Finally, the visual features and spatial structural features are fully integrated and input into the classifier by employing skip connections. HGTNet achieves classification accuracies of 98.47%, 95.75%, and 96.33% on the aerial image, NWPU-RESISC45, and OPTIMAL-31 datasets, respectively, demonstrating superior performance compared to other state-of-the-art models. Extensive experimental results indicate that our proposed method effectively learns critical multi-scale visual features and distinguishes between similar complex scenes, thereby significantly enhancing the accuracy of RSSC.

The accurate extraction of agricultural parcels from remote sensing images is crucial for advanced agricultural management and monitoring systems. Existing methods primarily emphasize regional accuracy over boundary quality, often resulting in fragmented outputs due to uniform crop types, diverse agricultural practices, and environmental variations. To address these issues, this paper proposes DSTBA-Net, an end-to-end encoder–decoder architecture. Initially, we introduce a Dual-Stream Feature Extraction (DSFE) mechanism within the encoder, which consists of Residual Blocks and Boundary Feature Guidance (BFG) to separately process image and boundary data. The extracted features are then fused in the Global Feature Fusion Module (GFFM), utilizing Transformer technology to further integrate global and detailed information. In the decoder, we employ Feature Compensation Recovery (FCR) to restore critical information lost during the encoding process. Additionally, the network is optimized using a boundary-aware weighted loss strategy. DSTBA-Net aims to achieve high precision in agricultural parcel segmentation and accurate boundary extraction. To evaluate the model’s effectiveness, we conducted experiments on agricultural parcel extraction in Denmark (Europe) and Shandong (Asia). Both quantitative and qualitative analyses show that DSTBA-Net outperforms comparative methods, offering significant advantages in agricultural parcel extraction.