A generalized framework for agricultural field delineation from high-resolution satellite imageries

Yu ZhuYaozhong PanTangao HuDujuan ZhangChuanwu ZhaoYuan Gaoa State Key Laboratory of Remote Sensing Science,Faculty of Geographical Science,Beijing Normal University,Beijing,People's Republic of Chinab Key Laboratory of Environmental Change and Natural Disaster of Ministry of Education,Faculty of Geographical Sciences,Beijing Normal University,Beijing,People's Republic of Chinac Academy of Plateau Science and Sustainability,Qinghai Normal University,Xining,People's Republic of Chinad Zhejiang Provincial Key Laboratory of Urban Wetlands and Regional Change,Hangzhou Normal University,Hangzhou,People's Republic of Chinae National Supercomputing Center in Zhengzhou,Zhengzhou University,Zhengzhou,People's Republic of China

DOI: https://doi.org/10.1080/17538947.2023.2297947

IF: 4.606

2024-01-12

International Journal of Digital Earth

Abstract:Accurate digital data on agricultural fields are crucial for various agriculture applications. While deep learning methods have shown promise in delineating fields from high-resolution imagery, there is a lack of research evaluating key techniques for field boundary detection. Challenges also exist in converting detection results into high-quality fields. This study addresses these issues and proposes a generalized framework for agricultural field delineation (GF-AFD). First, we identify three key techniques for field boundary detection and apply them to MPSPNet model. Ablation and comparison experiments demonstrate significant performance enhancements due to techniques. They also prove effective for DeeplabV3+ model, which shares a similar architecture. The modified models outperform U-Net-based models and approach the state-of-the-art models. Second, we show that performing region segmentation on boundary results yields improved field shapes. To address weak boundary loss and unstable parameters issues in existing segmentation-based methods, we introduce the OWT method to enhance weak boundaries directionally before segmentation. We also develop a hierarchical merging method, leveraging the observational hierarchy of fields, resulting in stable parameters across regions and models. The proposed GF-AFD framework was validated cross three diverse Chinese counties. The results demonstrate the framework's robust performance, providing a valuable solution for delineating agricultural fields.

geography, physical,remote sensing

What problem does this paper attempt to address?

The paper attempts to address the following issues:

1. **High-precision agricultural plot data acquisition**: Existing high-precision agricultural plot data mainly relies on field surveys and manual visual delineation, which are time-consuming and costly. Therefore, an efficient and automated method is needed to extract agricultural plot data from high-resolution satellite images.

2. **Challenges in boundary detection and region generation**: Existing agricultural plot division methods are divided into region-based approaches (RBA) and edge-based approaches (EBA). RBA methods tend to over-segment plots with high internal variability and under-segment small adjacent plots. Although EBA methods can detect detailed boundary information, they face issues such as weak boundary loss and parameter instability when converting detection results into high-quality plots.

3. **Limitations of existing models**: Current deep learning models perform well in detecting plot boundaries but lack the ability to effectively distinguish boundary types. Additionally, while existing multi-task models improve detection performance through additional regional task signals, their design has not been deeply analyzed based on the characteristics of plot boundaries, making it difficult to assess the impact of key technologies on detection accuracy.

To address these issues, this study proposes a General Framework for Agricultural Field Division (GF-AFD), aiming to:
- Summarize and evaluate three key technologies in plot boundary detection: multi-task processing (region and boundary), balancing low-level detail features with high-level category features, and emphasizing boundary detection training.
- Enhance weak boundary parts through Oriented Watershed Transform (OWT) to improve region segmentation effectiveness.
- Develop a hierarchical merging method to reduce manual intervention and ensure parameter stability, thereby obtaining high-quality plot objects.

In summary, this study proposes the GF-AFD framework to address the shortcomings of existing methods in high-precision agricultural plot data acquisition and high-quality plot generation, providing valuable solutions for agricultural resource management and precision agriculture.