TATVA: Turmeric Adulteration Detection using Thermal Video Analysis
Rupinder Kaur, Shahbaz Ahmad Khanday, Simrandeep Singh, Uday Thakur, A.K. Verma, Mukesh Kumar Saini · 2025
In this study, a novel, non-invasive, and automated approach for turmeric (Curcuma Longa L.) adulteration detection using thermal video analysis (TATVANet) is presented. A custom dataset TATVA, comprising 1,478 thermal videos of pure and adulterated turmeric, is developed under controlled laboratory conditions. Adulterated samples are prepared by introducing varying concentrations of common adulterants like starch, chickpea flour, lead chromate, and metanil yellow into pure turmeric. The primary objective of this study is to perform binary classification of turmeric samples as pure or adulterated. Thermal videos are recorded using a Fluke TiX580 thermal camera and subsequently preprocessed to extract meaningful spatio-temporal features required to train transformer-based video models. The proposed method uses the thermal signature of turmeric in conjunction with an attention mechanism of transformer architecture. TATVANet detects the pixel intensity transition between thermal video frames over time, capturing the heat conduction dynamics. TATVANet incorporates patch embedding, substantial downsampling, VIT encoder layer, and CLS recalibration. The proposed method achieves an accuracy of 92.90% with an average precision of 0.94 and outperforms state-of-the-art approaches that mostly analyze RGB and thermal images.