HBRNet: Boundary Enhancement Segmentation Network for Cropland Extraction in High-Resolution Remote Sensing Images

Abstract

Cropland extraction has great significance in crop area statistics, intelligent farm machinery operations, agricultural yield estimates, and so on. Semantic segmentation is widely applied to remote sensing image cropland extraction. Traditional semantic segmentation methods using convolutional networks result in a lack of contextual and boundary information when extracting large areas of cropland. In this paper, we propose a boundary enhancement segmentation network for cropland extraction in high-resolution remote sensing images (HBRNet). HBRNet uses Swin Transformer with the pyramidal hierarchy as the backbone to enhance the boundary details while obtaining context. We separate the boundary features and body features from the low-level features, and then perform a boundary detail enhancement module (BDE) on the high-level features. Endeavoring to fuse the boundary features and body features, the module for interaction between boundary information and body information (IBBM) is proposed. We select remote sensing images containing large-scale cropland in Yizheng City, Jiangsu Province as the Agricultural dataset for cropland extraction. Our algorithm is applied to the Agriculture dataset to extract cropland with mIoU of 79.61%, OA of 89.4%, and IoU of 84.59% for cropland. In addition, we conduct experiments on the DeepGlobe, which focuses on the rural areas and has a diversity of cropland cover types. The experimental results indicate that HBRNet improves the segmentation performance of the cropland.

BAFormer: A Novel Boundary-Aware Compensation UNet-like Transformer for High-Resolution Cropland Extraction

“…Feature fusion-based approaches [5,[17][18][19][20][21][22][23][24][25][26] enhance the representation of cropland information by supplementing additional feature information to the model. Considering the difficulty in labeling the existing high-resolution remote sensing image samples, [17] utilized the existing medium-resolution remote sensing images as a priori knowledge to provide cross-scale relocatable samples for HR images, thus obtaining more effective high-resolution farmland samples.…”

“…[21] proposed a fully convolutional neural network HRNet-CRF with improved contextual feature representation to optimize the initial semantic segmentation results by morphological postprocessing methods to obtain internally homogeneous farmland. [22] proposed a boundary-enhanced segmentation network, HBRNet, with Swing-Transformer as the backbone of the pyramid hierarchy to obtain contextual information while enhancing boundary details. Aiming at the different texture features of plots, [24] proposed a pyramid scene parsing network-statistical texture learning deep learning framework that combines high-level semantic feature extraction with low-level texture feature deep mining to achieve more accurate farmland recognition.…”

“…The dataset includes five foreground classes (impervious surfaces, buildings, low vegetation, trees, and cars) and one background class (clutter). For the specific experiments, we tested using image patches with IDs 2,4,6,8,10,12,14,16,20,22,24,27,29,31,33,35,and 38 for training, and used image patch with ID 30 for validation. The remaining 15 image patches were used for training.…”

“…However, most of them focus on specific geographic regions or a single cropland type, ignoring the regional differences of cropland parcels, and failing to achieve the purpose of generalized extraction. (2) Feature-based approaches [5,[17][18][19][20][21][22][23][24][25][26] enhance the representation and understanding of cropland information by supplementing the model with additional features. However, some redundant feature representations do not provide positive images to the model while increasing the computational burden of the model.…”

Vegetation Land Segmentation with Multi-Modal and Multi-Temporal Remote Sensing Images: A Temporal Learning Approach and a New Dataset

“…A series of optimization methods [31,32] have been proposed to improve the original VIT for better adaptation to visual tasks. HBRNet [33] performs best in field segmentation tasks by introducing the Swin transformer to obtain a larger receptive field. To balance the local and global information of remote sensing images, some models [34][35][36][37] that have combined unet and transformer have been proposed.…”