Transfer Learning Based Model for Classification of Cocoa Pods

Abstract

Over time, more sophisticated open source convolutional neural networks are becoming accessible for use. In conjunction with transfer learning, this allows projects with more focused and industry-specific requirements to generate positive results and make innovations within their field. This paper further describes the process of refining a deep learning model trained to identify cocoa pods from their surroundings, to additionally differentiate the ripe cocoa pods that are ready to be harvested. This is an ongoing project aiming to eventually automate the harvesting of cocoa pods on cocoa plantations, using automated robotic workers which can gather and transport products. This refinement is a vital step towards eventually circumnavigating the pitfalls which have, until now, left the cocoa industry working at pre-industrial standards. The obtained results show the satisfactory ability of the model proposed to classify correctly cocoa pods within static images.