Apple Identity Recognition Based on SVM Model Parameter Optimization and Near Infrared Hyperspectral

Abstract

Xinjiang Aksu apple with national geographical indication protection products is often counterfeited. In order to solve this problem, this paper uses hyperspectral imaging technology combined with SVM algorithm to identify apple samples from different producing areas and varieties. A total of 258 apple samples from different regions were collected by hyperspectral imager. The region of interest (ROI) of apple hyperspectral image was selected by ENVI software. Nine ROI were selected from the positive and nine from negative sides of each apple sample, and the average spectral value within the ROI was calculated. Then, SPA was used to reduce the dimension of the original spectral signal. Then, four different models were constructed by SVM, GS-SVM, GA-SVM and PSO-SVM. Different kernel functions were used to establish the hyperspectral apple classification prediction model. The results showed that the indicators of PSO-SVM with sigmoid as the kernel function were the best values. The accuracy was 91.6016%, precision was 96.1574%, recall was 88.6111%, F1 was 92.2269%. The model had high stability and prediction accuracy, which could meet the actual prediction needs. The results show that the near infrared hyperspectral based on SVM model parameters optimization can quickly identify apple species, which provides the basis for large-scale apple classification in the future, and also provides reference for standardizing apple trading market.