Non-destructive classification of defective potatoes based on hyperspectral imaging and support vector machine

Abstract

For the flaws of the traditonal manual selecting methods of potatoes classification, such as low efficiency, high labor intensity, poor objectivity and easy to be mischecked, this paper proposes a classification method of potatoes selection based on the combination of hyperspectral imaging (HSI) technology and multi-class support vector machine (MSVM), and this method can increase the accuracy and speed of potato classification. Here, potatoes are divided into six types: intact ones, green skin, germination, dry rot, wormhole and damage. The information of HSI potatoes is collected by using visible-near infrared system in the range of 400–1000 nm. Further, to improve the recognizability of the data, the background interference obtained from hyperspectral image is removed by masking. Then, the average spectra within the potato region is obtained. The lesion region and the normal region spectra are extracted as desired data sets. Next, the linear discriminant analysis (LDA) is used to reduce the dimension of the data, and the support vector machine (SVM) model is established to classify these types. For the potato samples not involved for modeling, K-means clustering method is used to segment the image, and the spectral input model is extracted for each region after segmentation for recognition. The test accuracy is up to 90%. It can be seen from the results that the potatoes with different types of internal and external defects can be identified more accurately and quickly using HSI technology, and they can be classified based on recognition, which provides a theoretical basis for the application of HSI technology in the actual potatoes automatic sorting system.