Scholarlink

<p>Camellia oil is highly susceptible to adulteration with cheap oils such as maize oil, soybean oil, and rapeseed oil due to its high nutritional and economic value. It is urgent to develop non-destructive detection techniques with high accuracy and reliability. This study employed near-infrared spectroscopy combined with deep learning algorithms for the quantitative analysis of binary adulteration in camellia oil. The results showed that long short-term memory models exhibited the best overall performance, achieving coefficient of determination for prediction values of 0.9964 and 0.9934 and residual predictive deviation of validation values of 6.91 and 11.68 for camellia-maize oil mixture and camellia oil adulterated with soybean oil samples, respectively. The findings indicated that the long short-term memory model achieved an excellent performance, outperforming all traditional models and highlighting its superior ability to model the nonlinear relationships inherent in spectral data. This study has confirmed the feasibility of combining near-infrared spectroscopy with deep learning algorithms for the quantitative analysis of binary adulteration in camellia oil. As mentioned above, the proposed approach has offered an efficient and non-destructive solution for edible oil quality control and supported significant practical value for ensuring food safety.</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

Deep learning algorithms for near-infrared spectroscopy-based quantitative analysis of binary adulteration in camellia oil

  • 期刊论文

  • 发表日期：

    2026年3月7日

查看全文
  • Wenxuan Deng,
  • Ruoni Wang,
  • Yifei Du,
  • Yue Yu,
  • Guanghui Shen,
  • Zhongyang Ren,
  • Zhanming Li

摘要

Camellia oil is highly susceptible to adulteration with cheap oils such as maize oil, soybean oil, and rapeseed oil due to its high nutritional and economic value. It is urgent to develop non-destructive detection techniques with high accuracy and reliability. This study employed near-infrared spectroscopy combined with deep learning algorithms for the quantitative analysis of binary adulteration in camellia oil. The results showed that long short-term memory models exhibited the best overall performance, achieving coefficient of determination for prediction values of 0.9964 and 0.9934 and residual predictive deviation of validation values of 6.91 and 11.68 for camellia-maize oil mixture and camellia oil adulterated with soybean oil samples, respectively. The findings indicated that the long short-term memory model achieved an excellent performance, outperforming all traditional models and highlighting its superior ability to model the nonlinear relationships inherent in spectral data. This study has confirmed the feasibility of combining near-infrared spectroscopy with deep learning algorithms for the quantitative analysis of binary adulteration in camellia oil. As mentioned above, the proposed approach has offered an efficient and non-destructive solution for edible oil quality control and supported significant practical value for ensuring food safety.