Abstract: ObjectiveTerraced fields serve as an effective engineering measure for soil and water loss control. Rapid and accurate acquisition of their spatial distribution information provides crucial data support for evaluating regional soil and water conservation benefits and planning agricultural production layouts. In response to the challenges of difficult boundary and field surface recognition, as well as morphological fragmentation and dispersion of terraced fields in high - resolution remote sensing images, this study constructed a dataset of terraced fields using high - resolution remote sensing images and proposed an AMSL - Net model for terraced field segmentation. MethodsIn the encoder stage of the AMSL - Net model, the lightweight MobileNetV3 is employed as the backbone network. The ASPP - CBAM module is introduced to extract multi - scale features of terraced fields, and long skip connections are utilized to establish long - range dependencies, enabling the fusion of features at different levels of a given image. In the decoder stage, deformable convolutions are applied to adapt to the shape variations of terraced fields, and the DySample dynamic upsampler is used to enhance classification accuracy. ResultsA comparative experiment was conducted on the same test set using five models, namely U-Net++, Seg-Net, PSP-Net, CPF-Net, and DeepLabV3. The experimental results show that the proposed model achieves excellent performance in multiple metrics: it reaches 96.18% in overall accuracy (OA), 90.97% in intersection over union (IoU), and 95.24% in F1-score, respectively. Compared with other methods, the accuracy has been significantly improved. In addition, the results of ablation experiments verify that each module has an obvious promoting effect on terrace recognition. ConclusionThis model exhibits high accuracy in segmenting terraced fields from high - resolution remote sensing images, offering a valuable reference for the refined monitoring and management of terraced fields.