Scholarlink

<p>Accurate monitoring of paddy rice is imperative for sustainable human development due to its pivotal role in ensuring food supply. Despite India’s status as a major paddy rice producer with favorable climatic conditions, the intricate growth patterns of paddy rice present challenges in developing effective models. To tackle this issue, the research employs a GRU-based deep learning model that utilizes time-series Sentinel-1 data to create a comprehensive rice field map by learning the temporal pattern of pixels during the growth stages and employs binary spatial classification. The model’s efficacy was evaluated across diverse landscapes, demonstrating its suitability for both mixed land use and rice-dominant environments. The model is also compared with a CNN based model, 3DCNN to evaluate its performance. Specifically, the model was applied to delineate rice fields in Udupi Taluk, yielding successful predictions for a rice area of 24.21&#xa0;km² during the Kharif season of 2021. Overall, the proposed model exhibited an accuracy of 98.4%, effectively navigating complex cultivation conditions and accurately identifying fragmented paddy rice fields. This study introduces a promising technique for rice field mapping in tropical production regions, holding significant potential for advancing global sustainable development in food and environmental management.</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

Gated recurrent unit neural network for rice field mapping

  • 期刊论文

  • 发表日期：

    2025年6月25日

查看全文
  • A. Aishwarya Hegde,
  • Pranav Rao,
  • Pruthviraj Umesh,
  • Mohit P. Tahiliani

摘要

Accurate monitoring of paddy rice is imperative for sustainable human development due to its pivotal role in ensuring food supply. Despite India’s status as a major paddy rice producer with favorable climatic conditions, the intricate growth patterns of paddy rice present challenges in developing effective models. To tackle this issue, the research employs a GRU-based deep learning model that utilizes time-series Sentinel-1 data to create a comprehensive rice field map by learning the temporal pattern of pixels during the growth stages and employs binary spatial classification. The model’s efficacy was evaluated across diverse landscapes, demonstrating its suitability for both mixed land use and rice-dominant environments. The model is also compared with a CNN based model, 3DCNN to evaluate its performance. Specifically, the model was applied to delineate rice fields in Udupi Taluk, yielding successful predictions for a rice area of 24.21 km² during the Kharif season of 2021. Overall, the proposed model exhibited an accuracy of 98.4%, effectively navigating complex cultivation conditions and accurately identifying fragmented paddy rice fields. This study introduces a promising technique for rice field mapping in tropical production regions, holding significant potential for advancing global sustainable development in food and environmental management.