Identification of genetically modified cotton seeds by terahertz spectroscopy with MPGA-SVM

The purpose of this paper was to propose a method that combines support vector machine (SVM) and multi-population genetic algorithm (MPGA) for identifying Genetically modified (GM) cotton seeds with THz spectroscopy. The parameters of SVM were tuned by using MPGA. Comparing previous reports of THz-TDS for GM crops detection, we used a larger sample size and more evaluation criterions (i.e. confusion matrix, average accuracy and 95th percentile of accuracy) in the experiment. Principal component analysis (PCA) was utilized to reduce dimensions of THz absorbance spectra, and then used the result of PCA as the input of different classifiers. When the input dimensionality of MPGA-SVM was 12, recall, precision and F-score were more than 97.9%, 96% and 0.9796, respectively, and accuracy was 99%. To further study the performance of MPGA-SVM with different input dimensionality, different number of principal components (ranged from 2 to 16 at intervals of 2) was selected. In addition, the proposed method was compared with traditional classifiers (decision trees (DT), k-nearest neighbor (KNN) and discriminant analysis (DA)). All the results showed that MPGA-SVM has better performance than other classifier. Thus, MPGA-SVM combined with THz spectroscopy is a potential identification tool for GM cotton seeds detection.