Scholarlink

<p>Food adulteration threatens the public health and requires strong detection methods using advanced techniques in machine learning and deep learning. This paper presents a new approach that combines quantum-inspired hybrid optimization with deep-learning models to achieve efficiency and accuracy in food adulteration detection. The data collection involves two publicly available datasets: the one called Food Adulteration Dataset and another from Kaggle titled Groundnut Oil Adulteration Dataset. Preprocessing techniques concerned dealing with missing values using Multiple Imputation by Chained Equations (MICE); then, outliers removal using the Interquartile Range (IQR) method; normalization using min–max scaling on the numerical features; encoding of text data using TF-IDF and Word2Vec embeddings; among others and for Generative Artificial Intelligent Data augmentation using Synthetic Minority Over-sampling Technique (SMOTE). The feature extraction computation will be assigned independently to numeric and textual data. The feature extraction of numeric data relies on statistical and ratio methods. The text features will be extracted based on N-grams, character N-grams processing, and part-of-speech tagging of parts of the sentence. Also, for further processing on text-based feature extraction, RoBERTa-based deep learning techniques will be used. For the optimization of feature selection, a Quantum-Inspired Hybrid Optimization Approach has been introduced that is coupled by Grey wolf Optimizer (GWO) and Quantum Walk Optimization Algorithm (QWOA) to gain efficiency in feature subset selection. The training of the model by Hybrid Deep Learning Quantum Multilayer Perceptron (MLP) and EfficientNet for numerical data and MOGNET (Mixture of Gaussians-based Neural Network) for textual classification is conducted. This experiment demonstrates the obtained results proving that this novel hybrid approach significantly increased the classification accuracy as it reduced the computational complexity. This proposed framework is also scalable in providing a high-performance solution for real-time food adulteration detection and has scope in cases of regulatory enforcement and quality assurance.</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

Food Adulteration Detection Using Quantum-Inspired Hybrid Optimization Approach

  • 期刊论文

  • 发表日期：

    2026年7月6日

查看全文
  • Reshmi S.,
  • Grace Mary Kanaga E

摘要

Food adulteration threatens the public health and requires strong detection methods using advanced techniques in machine learning and deep learning. This paper presents a new approach that combines quantum-inspired hybrid optimization with deep-learning models to achieve efficiency and accuracy in food adulteration detection. The data collection involves two publicly available datasets: the one called Food Adulteration Dataset and another from Kaggle titled Groundnut Oil Adulteration Dataset. Preprocessing techniques concerned dealing with missing values using Multiple Imputation by Chained Equations (MICE); then, outliers removal using the Interquartile Range (IQR) method; normalization using min–max scaling on the numerical features; encoding of text data using TF-IDF and Word2Vec embeddings; among others and for Generative Artificial Intelligent Data augmentation using Synthetic Minority Over-sampling Technique (SMOTE). The feature extraction computation will be assigned independently to numeric and textual data. The feature extraction of numeric data relies on statistical and ratio methods. The text features will be extracted based on N-grams, character N-grams processing, and part-of-speech tagging of parts of the sentence. Also, for further processing on text-based feature extraction, RoBERTa-based deep learning techniques will be used. For the optimization of feature selection, a Quantum-Inspired Hybrid Optimization Approach has been introduced that is coupled by Grey wolf Optimizer (GWO) and Quantum Walk Optimization Algorithm (QWOA) to gain efficiency in feature subset selection. The training of the model by Hybrid Deep Learning Quantum Multilayer Perceptron (MLP) and EfficientNet for numerical data and MOGNET (Mixture of Gaussians-based Neural Network) for textual classification is conducted. This experiment demonstrates the obtained results proving that this novel hybrid approach significantly increased the classification accuracy as it reduced the computational complexity. This proposed framework is also scalable in providing a high-performance solution for real-time food adulteration detection and has scope in cases of regulatory enforcement and quality assurance.