Food Adulterant Detection Using Hyperspectral Imaging System in Okra

The adulteration of food products in relation to food safety and quality assurance for the project results in serious consequences with regards to consumer health and market integrity. This work attempts to achieve a reliable and rapid means of detecting adulteration in food by use of hyperspectral imaging (HSI) system. Given its potential in capturing and analyzing a wide spectrum of light outside the visible range, hyperspectral imaging becomes a very great tool in identifying and quantifying adulterants not easily detectable with visual or chemical methods. A collection of comprehensive set of hyperspectral images taken from both pure and adulterated okra (lady's finger) samples were processed in the data extraction process of spectral signatures and other important features which will differentiate the adulterated okra from the non-adulterated okra. Advanced image processing techniques such as Principal Component Analysis (PCA) was applied to extract the features, augment by applying noise injection and rotation techniques. In addition to this, superior machine learning algorithms like Conventional Neural Network (CNN) were used for hyperspectral data analysis and classification of the okra samples. The main focus is to develop computational models that could classify the level of okra adulteration with better accuracy and quantify the same. Three CNN models (1D CNN, 2D CNN and 3D CNN) were applied to the samples. It was inferred that 1D CNN model classified the okra samples more accurately with a test accuracy of 0.81 when compared to other models 2D and 3D.