Spectral Detection Method for Chilling Damage of Sweet Potato Based on Support Vector Machine

The storage of sweet potato is susceptible to chilling damage, which not only affects its quality but also is susceptible to other diseases, which greatly reduces the commercial and economic value of sweet potato. For the biological detection technology and traditional detection methods will cause irreversible damage to sweet potato, so a non-destructive detection method of sweet potato chilling damage based on optical fiber spectroscopy technology was established. The feature spectrum bands were selected by the key feature ranking method based on the class separability criterion. And the support vector machine algorithm was used to train and evaluate the data set to detect the accuracy of the characteristic spectrum and the occurrence of sweet potatoes before the symptoms of chilling damage were visible to the human eyes. Totally five sweet potato varieties were used for experiments. The accuracy of detecting chilling damage in sweet potatoes was as high as 99.52% with the ratio of training data and test data of 5:5, and the incidence of the ratio of 7:3 was as high as 99.63%. The results proved the correctness of the characteristic spectrum, and the characteristic spectral bands were as follows: Jishu26 was 821.3~823.5 nm; Yanshu25 was 810.5~821.9 nm; Xiguahong was 818.8~821.9 nm; Xinong431 was 601.5~606.4 nm; and Longshu9 was 759.6~761.2 nm. The results showed that the spectroscopy technology can effectively detect and identify sweet potato chilling damage. This research provided technical method for subsequent work for sweet potato storage classification. © 2020, Chinese Society of Agricultural Machinery. All right reserved.