With the rapid development of digital intelligent agriculture, the accurate extraction of field information from remote sensing imagery to guide agricultural planning has become an important issue. In order to better extract fields, we analyze the scale characteristics of agricultural fields and incorporate the multi-scale idea into a Transformer. We subsequently propose an improved deep learning method named the Multi-Swin Mask Transformer (MSMTransformer), which is based on Mask2Former (an end-to-end instance segmentation framework). In order to prove the capability and effectiveness of our method, the iFLYTEK Challenge 2021 Cultivated Land Extraction competition dataset is used and the results are compared with Mask R-CNN, HTC, Mask2Former, etc. The experimental results show that the network has excellent performance, achieving a bbox_AP50 score of 0.749 and a segm_AP50 score of 0.758. Through comparative experiments, it is shown that the MSMTransformer network achieves the optimal values in all the COCO segmentation indexes, and can effectively alleviate the overlapping problem caused by the end-to-end instance segmentation network in dense scenes.

In the field of remote sensing image processing, semantic segmentation has always been a hot research topic. Currently, deep convolutional neural networks (DCNNs) are the mainstream methods for the semantic segmentation of RSI. There are two commonly used semantic segmentation methods based on DCNNs: multi-scale feature extraction based on deep-level features, and global modeling. The former can better extract object features of different scales in complex scenes. However, this method lacks sufficient spatial information, resulting in poor edge segmentation ability. The latter can effectively solve the problem of limited receptive field in DCNNs obtaining more comprehensive feature extraction results. Unfortunately, this method is prone to misclassification, resulting in incorrect predictions of local pixels. To address these issues, we propose the Dual-branch Hybrid Reinforcement Network (DHRNet) for more precise semantic segmentation of RSI. This model is a dual-branch parallel structure with a multi-scale feature extraction branch (MFEB) and a global context and detail enhancement branch (GEB). This structure decomposes the complex semantic segmentation task, allowing each branch to extract features with different emphases while retaining sufficient spatial information. The results of both branches are fused to obtain a more comprehensive segmentation result. After conducting extensive experiments on three publicly available RSI datasets, ISPRS Potsdam, ISPRS Vaihingen and LoveDA, DHRNet demonstrates excellent results with the mean intersection over union of 86.97%, 83.53% and 54.48% on the three datasets, respectively.