Scholarlink

<p>Capsicum, a key agricultural commodity, is projected to reach a global market value of $6.8&#xa0;billion by 2030. Traditional methods for assessing harvest maturity are labor-intensive and inefficient. This study introduces an artificial intelligence-based system for automated capsicum maturity detection. Capsicum samples, grown under polyhouse conditions, were analyzed at different Weeks After Anthesis (WAA) for attributes such as firmness, pH, and Total Soluble Solids (TSS). A Support Vector Regression (SVR) model and Random Forest (RF) model were trained to predict WAA. The SVR model, with an R-squared value of 0.9938, outperformed RF and was integrated into CapsiMature, a web application classifying capsicum maturity into immature, mature, and overmature stages which was used for custom training the deep learning models, including Faster R-CNN, YOLOv5, and YOLOv8, were developed using a dataset of 485 images annotated with CapsiMature outputs. YOLOv8 exhibited superior performance, achieving a mean Average Precision (mAP) of 0.835, an F1-score of 0.81, and an inference time of 12.5 milliseconds per image, with a lightweight 22.5&#xa0;MB model size. This research demonstrates the integration of laboratory analysis, machine learning, and computer vision for real-time capsicum maturity detection, offering a scalable solution to enhance agricultural productivity.</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

Integrative approach for capsicum maturity detection: from laboratory analysis to machine learning assisted deep learning custom model development

  • 期刊论文

  • 发表日期：

    2026年5月26日

查看全文
  • Ayan Paul,
  • S. Ajay Nayak,
  • Rajendra Machavaram,
  • Pooja Verma

摘要

Capsicum, a key agricultural commodity, is projected to reach a global market value of $6.8 billion by 2030. Traditional methods for assessing harvest maturity are labor-intensive and inefficient. This study introduces an artificial intelligence-based system for automated capsicum maturity detection. Capsicum samples, grown under polyhouse conditions, were analyzed at different Weeks After Anthesis (WAA) for attributes such as firmness, pH, and Total Soluble Solids (TSS). A Support Vector Regression (SVR) model and Random Forest (RF) model were trained to predict WAA. The SVR model, with an R-squared value of 0.9938, outperformed RF and was integrated into CapsiMature, a web application classifying capsicum maturity into immature, mature, and overmature stages which was used for custom training the deep learning models, including Faster R-CNN, YOLOv5, and YOLOv8, were developed using a dataset of 485 images annotated with CapsiMature outputs. YOLOv8 exhibited superior performance, achieving a mean Average Precision (mAP) of 0.835, an F1-score of 0.81, and an inference time of 12.5 milliseconds per image, with a lightweight 22.5 MB model size. This research demonstrates the integration of laboratory analysis, machine learning, and computer vision for real-time capsicum maturity detection, offering a scalable solution to enhance agricultural productivity.