Scholarlink

<p>Color machine vision systems provide a rapid, non-destructive, and low-cost approach for food quality assessment. However, their effectiveness is limited for visually homogeneous liquids, where texture, shape, and geometric cues are weak or absent. This study proposes Multi-Space Histogram Features (MSHF), an interpretable color descriptor that fuses peak-based histogram information from five complementary color spaces: normalized RGB, HSV, I1I2I3, YCbCr, and CIELab. The proposed framework was evaluated for milk type classification, robustness under spoilage-induced temporal color drift, and freshness monitoring. The discriminative capability of MSHF was assessed against individual and ablated color-space descriptors using linear and nonlinear classifiers, repeated stratified cross-validation, PCA visualization, statistical comparisons, and group-permutation importance analysis. The results showed that MSHF achieved near-perfect cross-validated performance for milk type classification and maintained strong robustness in aged-milk classification. For freshness monitoring, performance varied across milk types, indicating that spoilage-related chromatic changes were more detectable in some samples than others. Group importance analysis further showed that the most informative color space was dataset-dependent, supporting the need for multi-space color fusion rather than reliance on a single color representation. Although color-only analysis was highly effective for milk discrimination, early spoilage detection remained challenging when fresh and deteriorated samples exhibited overlapping visual characteristics. Overall, the proposed CMVS-MSHF framework provides a compact, interpretable, and deployment-friendly approach for homogeneous liquid assessment, with potential application in dairy quality monitoring and other low-texture fluid food products.</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

Multi-space color histogram feature fusion for non-destructive milk quality assessment in visually homogeneous liquids

  • 期刊论文

  • 发表日期：

    2026年6月27日

查看全文
  • Mazlina Mamat,
  • Lorita Angeline,
  • Nurul Halimatul Asmak Ismail,
  • Noorakmar Ab Wahab

摘要

Color machine vision systems provide a rapid, non-destructive, and low-cost approach for food quality assessment. However, their effectiveness is limited for visually homogeneous liquids, where texture, shape, and geometric cues are weak or absent. This study proposes Multi-Space Histogram Features (MSHF), an interpretable color descriptor that fuses peak-based histogram information from five complementary color spaces: normalized RGB, HSV, I1I2I3, YCbCr, and CIELab. The proposed framework was evaluated for milk type classification, robustness under spoilage-induced temporal color drift, and freshness monitoring. The discriminative capability of MSHF was assessed against individual and ablated color-space descriptors using linear and nonlinear classifiers, repeated stratified cross-validation, PCA visualization, statistical comparisons, and group-permutation importance analysis. The results showed that MSHF achieved near-perfect cross-validated performance for milk type classification and maintained strong robustness in aged-milk classification. For freshness monitoring, performance varied across milk types, indicating that spoilage-related chromatic changes were more detectable in some samples than others. Group importance analysis further showed that the most informative color space was dataset-dependent, supporting the need for multi-space color fusion rather than reliance on a single color representation. Although color-only analysis was highly effective for milk discrimination, early spoilage detection remained challenging when fresh and deteriorated samples exhibited overlapping visual characteristics. Overall, the proposed CMVS-MSHF framework provides a compact, interpretable, and deployment-friendly approach for homogeneous liquid assessment, with potential application in dairy quality monitoring and other low-texture fluid food products.