Damaged Apple Sorting with mmWave Imaging and Non-Linear Support Vector Machine

This paper is a proof of concept proposing and describing a complete workflow to differentiate healthy from damaged apples, starting with mmWave measurements and ending with a classification based on Support Vector Machine. The method has proven to be successful with only 6% error when scan angle and frequency diversity are used. In a first step, we build a database of more than 1800 images obtained by processing measurements with a two-dimensional fast Fourier transform. Images are then converted to binary and used as the input to a non-linear SVM. At this stage, 90% of the database is used for training, and coefficients C and γ are tuned to minimize the error. The remaining 10% of images are used for testing. In a second step, we assess and discuss the influence of the physical inputs of the database: the frequency, the sparsity of measurement points and the size of the apples. Finally we explore new scenarios considering other fruits.