Webcam based visible light spectrometer for accurate honey adulteration detection using variational autoencoder and Bayesian neural network

• Low-cost visible light spectrometer built using a webcam and DVD disc as optical components. • Differential absorbance features processed using Variational Autoencoder (VAE) for dimensionality reduction. • SVM classifier achieved 100% accuracy in detecting honey adulteration using VAE-extracted features. • Bayesian Neural Network predicted sweetener concentration with high accuracy. The authenticity of honey is often compromised due to adulteration, posing health risks and disadvantaging genuine producers. This study presents a compact, cost-effective visible light spectrophotometer, employing a webcam sensor and a DVD disc as the dispersive element, for honey purity classification. The device, measuring 17.8 × 8 × 7 cm, captured spectral intensity data from four types of honey samples: pure and those adulterated with orange syrup, cane syrup, and palm syrup at varying concentrations. Extracted features included absorbance and first-order differential absorbance, which were analysed using Principal Component Analysis (PCA) and Variational Autoencoder (VAE). The optimal VAE latent space features were used for classification via Support Vector Machine (SVM), achieving 100% accuracy with 5-fold cross-validation for differential absorbance. Additionally, a Bayesian Monte Carlo Dropout Neural Network (BNN) was trained on the VAE latent space to predict sweetener concentration, achieving high performance on testing data (MAE: 0.012, MSE: 0.007, R²: 0.9978). These results demonstrate the potential of the proposed webcam-based spectrophotometer as a reliable and affordable tool for honey adulteration detection.