Scholarlink

<p>Food adulteration remains a significant challenge in the meat industry, particularly when it involves mixing beef with pork, which raises ethical, health, and religious concerns. As pork consumption is restricted in various cultures and religions, ensuring the authenticity of meat products is essential for consumer trust and regulatory compliance. This study proposes SE-CBNet, a lightweight 1D-CNN-BiLSTM hybrid model with Squeeze-and-Excitation (SE), for accurate and non-destructive detection of pork adulteration in beef. The hybrid model combines one-dimensional convolutional (1D-CNN) layers to extract features from E-nose sensor signals, Squeeze-and-Excitation (SE) blocks to enhance channel-wise feature representation, and bidirectional Long Short-Term Memory (BiLSTM) layers to capture temporal dependencies across the sequence. The model is evaluated using a publicly available dataset collected through an electronic nose (E-nose) system, which captures the volatile compounds emitted by meat samples. The dataset includes seven classes representing different levels of adulteration, ranging from pure beef to pure pork. To improve signal quality and enhance model performance, discrete wavelet transform (DWT) is applied for noise reduction. SE-CBNet achieves 99.71% cross-validation accuracy and 98.81% test accuracy, demonstrating its strong capability in identifying adulterated samples. These results highlight the potential of combining E-nose sensing technology with advanced deep learning methods for reliable food authenticity verification and meat quality assessment.</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

SE-CBNet: a lightweight CNN-BiLSTM model with squeeze-and-excitation for pork adulteration detection in beef using an E-Nose

  • 期刊论文

  • 发表日期：

    2026年5月21日

查看全文
  • Tajmal Hussain,
  • Jongwon Seok

摘要

Food adulteration remains a significant challenge in the meat industry, particularly when it involves mixing beef with pork, which raises ethical, health, and religious concerns. As pork consumption is restricted in various cultures and religions, ensuring the authenticity of meat products is essential for consumer trust and regulatory compliance. This study proposes SE-CBNet, a lightweight 1D-CNN-BiLSTM hybrid model with Squeeze-and-Excitation (SE), for accurate and non-destructive detection of pork adulteration in beef. The hybrid model combines one-dimensional convolutional (1D-CNN) layers to extract features from E-nose sensor signals, Squeeze-and-Excitation (SE) blocks to enhance channel-wise feature representation, and bidirectional Long Short-Term Memory (BiLSTM) layers to capture temporal dependencies across the sequence. The model is evaluated using a publicly available dataset collected through an electronic nose (E-nose) system, which captures the volatile compounds emitted by meat samples. The dataset includes seven classes representing different levels of adulteration, ranging from pure beef to pure pork. To improve signal quality and enhance model performance, discrete wavelet transform (DWT) is applied for noise reduction. SE-CBNet achieves 99.71% cross-validation accuracy and 98.81% test accuracy, demonstrating its strong capability in identifying adulterated samples. These results highlight the potential of combining E-nose sensing technology with advanced deep learning methods for reliable food authenticity verification and meat quality assessment.