Apple Grading System based on Near Infrared Spectroscopy and Evidential Classification Forest

Apple grading plays an important role in increasing the commercial value of apple products. In this paper, an apple grading system for Red Fuji apples is proposed. Based on the non-destructive measurement of near infrared spectrum, the machine learning algorithm of evidential classification forest is applied to classify apples into four quality grades. To build the training set of classification forest, features are extracted by partial least square approach, meanwhile plausibilities of different grades are decided depending on corresponding soluble solids content of apple. Experiments with Red Fuji apple products shows a recognition rate around 80%.