Classification of Multiple Panamanian Watermelon Varieties Using Convolutional Neural Networks with Transfer Learning

Abstract

Panama is well regarded as an international exporter of watermelon in the Central American region. Most of the watermelon selection process is made by hand with empirical techniques, based on firmness, color, sound and random sampling from other specimens of the same batch. The overall goal of the project is to have an automated system able to distinguish between watermelon varieties and that is able to classify watermelons that are ready for export or can be sold locally. For this matter, traditional and novel computer vision and spectral pattern recognition algorithms are used.