Scholarlink

<p>Automating the critical task of agricultural field boundary delineation using remote sensing imagery serves as the foundation for effective implementation of various precision agriculture practices. The last few years have seen a rise in the adoption of Deep Learning (DL) models for the detection of field boundaries, but the accuracy of such models is centred both on the quantity and quality of labelled data. Traditional Machine Learning (ML) approaches have been extensively applied in single-crop and multi-crop classification using time series satellite data; however, its application in delineating agricultural field boundaries using unsupervised approaches remains underexplored. In this work, we have proposed an Ensemble Clustering technique to address this gap by leveraging traditional clustering models on various phenological, spectral, harmonic and textural features extracted from time-series data obtained using Sentinel-2 MultiSpectral Instrument (MSI) sensor for delineation of agricultural fields. The proposed approach was tested at three different sites in the province of Punjab, Pakistan. The extracted boundaries were compared with the ground data using both quantitative and qualitative assessment. This approach has high potential for accurate boundary delineation as the buffer overlay method with 3.0 m buffer width resulted into 85%-96% completeness, 86%-92% correctness and 74%-88% quality.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Ensemble based Time Series Clustering Framework for Field Boundary Delineation using Satellite Imagery

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Maleeha Najam,
  • Hasnat Khurshid

摘要

Automating the critical task of agricultural field boundary delineation using remote sensing imagery serves as the foundation for effective implementation of various precision agriculture practices. The last few years have seen a rise in the adoption of Deep Learning (DL) models for the detection of field boundaries, but the accuracy of such models is centred both on the quantity and quality of labelled data. Traditional Machine Learning (ML) approaches have been extensively applied in single-crop and multi-crop classification using time series satellite data; however, its application in delineating agricultural field boundaries using unsupervised approaches remains underexplored. In this work, we have proposed an Ensemble Clustering technique to address this gap by leveraging traditional clustering models on various phenological, spectral, harmonic and textural features extracted from time-series data obtained using Sentinel-2 MultiSpectral Instrument (MSI) sensor for delineation of agricultural fields. The proposed approach was tested at three different sites in the province of Punjab, Pakistan. The extracted boundaries were compared with the ground data using both quantitative and qualitative assessment. This approach has high potential for accurate boundary delineation as the buffer overlay method with 3.0 m buffer width resulted into 85%-96% completeness, 86%-92% correctness and 74%-88% quality.