A Study On Nondestructive Testing For Wheat Quality Based On THz Technology

Food Security concerns economic development and social stability. At present, in food quality safety detection, the main detection method is still manual method. Although there have been many biochemical detection methods, it is difficult to popularize in food domain due to complex operation programs. Spectrum methods greatly attract people’s attention because of its speediness and non-destructive. But traditional spectrum have still many shortcomings such as expensive device, narrow detection range. Considering terahertz spectrum is a new spectrum technology, and has many applications in biochemistry, and can make up much insufficient for traditional spectrum, we tried to apply terahertz spectrum in food quality safety detection and find new rules.In this paper, we took four different-quality wheat and eight sort of wheat as detection objects, and studied their terahertz spectrums respectively. At the same time, we also studied AFT B1’s terahertz spectrum in wheat, and do a quantitative analysis for their concentrations.For normal wheat, worm-eaten wheat, mildew wheat and sprout wheat, the four types of wheat, we collected their absorption spectrum, and adopted SVM combining PCV model to identify their classifications.At first, we used PCA to abstract four main components for absorption spectrum and three main components from refractive index spectrum as input variables for SVM, and use SVM to realize identification. The experimental results indicated that comparing with only using SVM, PCA-SVM has better performance for identification. In addition, compared with other methods, PCA-SVM has also better performance.For eight sorts of wheat, we used iPLS method to construct regression model to identify them. Through doing experiments, we determined the best area in terahertz spectrum, and constructed a iPLS model, and by comparing with PLS model, we proved that the iPLS model has better performance than PLS.For AFT in wheat, we developed a set of attraction method from wheat, and construct a regression model for forecasting their concentration. In order to improve forecasting performance, we improved D-S evidence inference algorithm. We attracted more frequency dots for fusion as evidence, and construct double evidence inference algorithm. The experimental results indicated that the model proposed in the paper can greatly improve the reliability for forecasting AFT’s concentration in wheat.