Fast and simultaneous detection of wheat kernel adulteration using hyperspectral imaging technology and deep convolutional neural network

Abstract

In this study, hyperspectral imaging technology combined with a novel convolution neural network was utilized to detect wheat kernels adulteration. Two groups of wheat kernels were used as samples in this study. Sound wheat kernels from various varieties were in one group, while unsound wheat kernels of the same variety were in another. Hyperspectral images collected from these two groups of wheat kernels were preprocessed using a series of commonly used methods. Following the collection of hyperspectral data, a method of separating and recombining individual wheat kernels from entire hyperspectral images was applied to create training sets and validation sets. Subsequently, a series of tests were carried out to verify whether the proposed model Following that, a number of experiments were conducted to confirm if the suggested model was effective in simultaneously detecting adulterated wheat kernels, and the results gave a positive conclusion. Finally, accuracy, precision, recall and F1-scores were used as indicators to evaluate the performance of the proposed models on the test set. As the results demonstrated, satisfactory performance in detecting adulteration of the two groups of wheat kernels was obtained by the proposed model. According to the results, the proposed model combined with HSI technology has a good prospect of being used as an efficient method for detecting wheat kernels adulteration.