Scholarlink

<p>The soil is essential in agriculture as it supports the environment for seed germination, plant establishment, and root development. Extensive research has been undertaken to predict soil structure utilizing diverse soil classification approaches. However, a system for predicting soil types with enhanced performance has yet to be designed. This study presents a novel architecture called “SoilQuadNet” designed for the classification of soil types, focusing on the enhancement of image processing techniques in order to improve the model’s accuracy. A proportion-based RGB-priority fusion approach for advanced feature engineering strategy is employed following comprehensive data augmentation, emphasizing enhanced performance resulting from image validation and filtration-based noise handling techniques. In addition, coordinate attention with a highly efficient functional CNN model is proposed in this study that embraces parallel architecture with diverse filters to emphasize both global and local features along with spatial location-based feature encoding. The evaluation procedure entails computing essential performance metrics for each model. The criteria are properly examined and contextualized to illustrate the efficacy of the proposed approach. Adopting this framework, the model achieved an excellent performance with accuracy value of 99.18 ± 0.15 and 98.12 ± 0.18 for dataset-1 and dataset-2, respectively, surpassing the baseline models. The improved precision enhances the model’s efficacy, making it appropriate for practical agricultural practices. This comprehensive approach expands the parameters of soil classification and highlights the need for accurate soil type identification to improve soil management and optimize crop productivity in agriculture.</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

SoilQuadNet: proportion based RGB-priority fusion with spatial feature attention encoded functional network for soil type classification

  • 期刊论文

  • 发表日期：

    2026年1月8日

查看全文
  • Sudipta Dash,
  • Aakanksha Sharaff

摘要

The soil is essential in agriculture as it supports the environment for seed germination, plant establishment, and root development. Extensive research has been undertaken to predict soil structure utilizing diverse soil classification approaches. However, a system for predicting soil types with enhanced performance has yet to be designed. This study presents a novel architecture called “SoilQuadNet” designed for the classification of soil types, focusing on the enhancement of image processing techniques in order to improve the model’s accuracy. A proportion-based RGB-priority fusion approach for advanced feature engineering strategy is employed following comprehensive data augmentation, emphasizing enhanced performance resulting from image validation and filtration-based noise handling techniques. In addition, coordinate attention with a highly efficient functional CNN model is proposed in this study that embraces parallel architecture with diverse filters to emphasize both global and local features along with spatial location-based feature encoding. The evaluation procedure entails computing essential performance metrics for each model. The criteria are properly examined and contextualized to illustrate the efficacy of the proposed approach. Adopting this framework, the model achieved an excellent performance with accuracy value of 99.18 ± 0.15 and 98.12 ± 0.18 for dataset-1 and dataset-2, respectively, surpassing the baseline models. The improved precision enhances the model’s efficacy, making it appropriate for practical agricultural practices. This comprehensive approach expands the parameters of soil classification and highlights the need for accurate soil type identification to improve soil management and optimize crop productivity in agriculture.