Complex vs. Amplitude-Based Machine Learning Techniques for Non-Destructive Evaluation at mm-Waves: Application to Apple's Sorting
Résumé
This paper investigates different machine learning tools for sorting out damage from healthy apples based on mm-Wave measurements. Previous work showed that we can successfully detect damaged apples with a nonlinear SVM (Support Vector Machine) algorithm in case of frequency and spatial diversity. Here we try to get rid of the frequency dependence and achieve the same performance when measurements are performed at a single frequency in D-Band. We aim to compensate for this loss of information by processing complex data instead of amplitude when SVM processing is applied for classification.