Automating the Process of Distinguishing Marketable Apples

The effective transfer of information accumulated from years of experience to following generations is a frequent challenge for farmers across the world. This report describes a work in progress which aims to preserve and apply such knowledge using machine learning techniques to help local apple farmers who face the same problem in fruit sorting processes, known as "senka" in Japanese. The process of identifying scratches, bruising, or other signs of illness in apples is typically carried out manually by only a few experienced farmers who reached their levels of expertise after many years of training. By allowing a deep learning software model to study a sufficient sample of images of the fruit sorted by veteran farmers, we aim to develop an automatic process for distinguishing marketable and non-marketable apples automatically and report the results of preliminary experiments which reached approximately 80% classification accuracy.