EUVIMEDEUVIMED entdecken
Crossref

Research on Methods for Crop-Field Parcel Information Extraction Based on Deep Learning Models

Su YANG, Ying WANG, Luxi ZHOU, Yang GU, Chong BAO, Hongtao SHI, Qing WU

Acta Interdisciplinary Science · 2026

DOI: 10.48014/ais.20260210002 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Acta Interdisciplinary Science, 2026. Verfasst von Su YANG und 6 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate extraction of crop-field parcel boundary information is of great value for agricultural production management, crop growth monitoring, yield assessment, and disaster monitoring. By integrating deep learning techniques with high-resolution remote sensing imagery, automated extraction of crop-field parcel information can be achieved. U-Net and DeepLabV3 are two commonly used convolutional neural network (CNN) models for crop-field segmentation. However, U-Net performs poorly in segmenting parcels with irregular shapes or blurred boundaries. In addition, DeepLabV3 has a relatively large network depth and complex structure, which leads to overfitting and low crop-field parcel information extraction accuracy. To address the above issues, this study pioneers the introduction of the lightweight LinkNet and the multi-scale Feature Pyramid Network (FPN) into crop-field parcel information extraction, constructing endto- end pixel-level semantic segmentation network models and analyzing the performance differences of various network architectures in terms of metrics such as mIoU, pixel accuracy (PA) , F1-score, and recall. For dataset construction, a standardized semantic segmentation sample set was established based on multisource high-quality data fusion. Following the data selection criteria of “high resolution, high timeliness, and high-precision alignment, ” Sentinel-2 multispectral remote sensing imagery was fused with 2023 highprecision farmland vector data from Jiangsu Province. Through an optimized sliding-window cropping size (128 × 128 pixels with 10% overlap) and a sample quality screening mechanism (excluding samples with farmland pixel proportion

Zitieren

APA

YANG, S., WANG, Y., ZHOU, L., GU, Y., BAO, C., SHI, H., & WU, Q. (2026). Research on Methods for Crop-Field Parcel Information Extraction Based on Deep Learning Models. Acta Interdisciplinary Science. https://doi.org/10.48014/ais.20260210002

Vancouver

YANG S, WANG Y, ZHOU L, GU Y, BAO C, SHI H, et al. Research on Methods for Crop-Field Parcel Information Extraction Based on Deep Learning Models. Acta Interdisciplinary Science. 2026. doi:10.48014/ais.20260210002

BibTeX
@article{YANG2026,
  author = {YANG, Su and WANG, Ying and ZHOU, Luxi and GU, Yang and BAO, Chong and SHI, Hongtao and WU, Qing},
  title = {Research on Methods for Crop-Field Parcel Information Extraction Based on Deep Learning Models},
  journal = {Acta Interdisciplinary Science},
  year = {2026},
  doi = {10.48014/ais.20260210002}
}
Kostenlos starten →