Comparative study of agricultural parcel delineation deep learning methods using satellite images: Validation through parcels complexity
Accurate delineation of agricultural parcels is crucial for applications ranging from resource management to policy decisions, with a direct impact on agricultural productivity and sustainability. Parcel delineation is the subject of numerous studies, most of which focus on the development of more e...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-03-01
|
| Series: | Smart Agricultural Technology |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2772375525000668 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Accurate delineation of agricultural parcels is crucial for applications ranging from resource management to policy decisions, with a direct impact on agricultural productivity and sustainability. Parcel delineation is the subject of numerous studies, most of which focus on the development of more efficient methods or ones better adapted to specific cases. In addition, various methods exist in the literature for delineating agricultural fields from satellite images. Deep learning, in particular, has revolutionized the field. Many state-of-the-art methods now utilize deep learning, often incorporating segmentation and classification techniques to define agricultural parcel boundaries. While recent research has led to a surge in deep learning methods for this task, evaluating their effectiveness goes beyond simply comparing outputs. This paper emphasizes the critical role of parcel complexity as a powerful lens for assessing the performance of deep learning methods in agricultural parcel delineation. We categorize 14 evaluation metrics into three main groups, global, boundary, and structure metrics, respectively. Global metrics assess the overall accuracy of the delineated parcels, boundary metrics focus on the precision of the parcel boundaries, and structure metrics examine the topological relationships between the parcels. Our goal is to compare these deep learning methods based on these metrics and their performance across varying levels of parcel complexity. We systematically evaluate nine state-of-the-art methods using a public database, explicitly analyzing how their strengths and weaknesses are affected by different levels of parcel complexity. This approach ensures that future deep learning techniques are robust and accurate enough to meet the demands of accurately defining the agricultural landscape and provides important insights for the development and refinement of future deep learning techniques. |
|---|---|
| ISSN: | 2772-3755 |