Scholarlink

<p>The differentiation of organic and non-organic produce is vital for trust and compliance. This research proposes a simple, non-destructive pipeline that combines Tristimulus Colorimetry with Machine Learning using the Portable Multispectral Sensor (VIS–NIR) to classify organic vs non-organic Tomatoes. Spectral reflectance from tomatoes (Organic, Non-Organic) collected under D65 was converted to CIE X, Y &amp; Z, chromaticity (x, y), dominant wavelength, purity, NDVI, PRI, and two CIE-plane descriptors relative to a standard color anchor i.e. Radial shift (Δ r) and angular shift (Δ θ). Supported by Vector Machines, Random Forests, and a lightweight CNN were trained with grid search and ten (10) fold cross-validation. The CNN performed best (Result: greater accuracy and greater F1 value for organic produce) and generalized to an external set of various organic &amp; non-organic tomatoes (Result: greater accuracy for organic produce). Machine Learning Model interpretation indicated the Tristimulus features were most influential (~ 65%) Misclassifications were rare and occurred mainly at early ripening, where organics showed higher y-chromaticity (i.e. Greater chlorophyll retention) The method enables fast, low-cost field authentication without chemical analysis and is ready for sensor-embedded deployment.</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

Design and development of the multispectral sensor with CIE Tristimulus colorimetry and machine learning framework for organic/non-organic tomato classification

  • 期刊论文

  • 发表日期：

    2026年4月23日

查看全文
  • Tejashree A. Aher,
  • A. D. Shaligram,
  • Supriya S. Patil,
  • Varsha Atul Shukre

摘要

The differentiation of organic and non-organic produce is vital for trust and compliance. This research proposes a simple, non-destructive pipeline that combines Tristimulus Colorimetry with Machine Learning using the Portable Multispectral Sensor (VIS–NIR) to classify organic vs non-organic Tomatoes. Spectral reflectance from tomatoes (Organic, Non-Organic) collected under D65 was converted to CIE X, Y & Z, chromaticity (x, y), dominant wavelength, purity, NDVI, PRI, and two CIE-plane descriptors relative to a standard color anchor i.e. Radial shift (Δ r) and angular shift (Δ θ). Supported by Vector Machines, Random Forests, and a lightweight CNN were trained with grid search and ten (10) fold cross-validation. The CNN performed best (Result: greater accuracy and greater F1 value for organic produce) and generalized to an external set of various organic & non-organic tomatoes (Result: greater accuracy for organic produce). Machine Learning Model interpretation indicated the Tristimulus features were most influential (~ 65%) Misclassifications were rare and occurred mainly at early ripening, where organics showed higher y-chromaticity (i.e. Greater chlorophyll retention) The method enables fast, low-cost field authentication without chemical analysis and is ready for sensor-embedded deployment.