Multimodal Dataset for Turmeric Adulteration Detection Using Imaging, ATR-FTIR Spectroscopy, and Color Measurements

This dataset provides a multimodal representation for the detection and quantification of turmeric adulteration using imaging, attenuated total reflectance–Fourier transform infrared (ATR-FTIR) spectroscopy, and color measurements. It is designed to support research in food authentication, chemometrics, and machine learning-based analysis.

Turmeric samples were prepared under controlled laboratory conditions by mixing pure turmeric powder with wheat flour at predefined adulteration levels of 0%, 1%, 2%, 3%, 5%, 7%, and 9% (w/w). Each class contains 50 samples, resulting in a total of 350 samples. Each sample is labeled using a structured naming convention (e.g., T0S01), where the number indicates the adulteration level and sample index.

The dataset includes three data modalities: (i) high-resolution image data captured under uniform conditions, with deep features extracted using EfficientNet-B0 (1280-dimensional feature vectors); (ii) ATR-FTIR spectral data representing chemical composition; and (iii) color measurements obtained using a HunterLab colorimeter, including CIE L*a*b* values and spectral reflectance data.

This dataset enables the development of classification and regression models for both detection and estimation of adulteration levels. It also supports multimodal data fusion by combining visual, chemical, and color information. Due to its controlled preparation and balanced class distribution, the dataset can serve as a benchmark for food quality assessment and adulteration detection studies.