Coffee Fruit Recognition Using Artificial Vision and neural NETWORKS
Mauricio Serrano Fuentes, Nelson Alberto Lizardo Zelaya, José Luis Ordóñez-Ávila
Abstract
Coffee is main agricultural product of export in Honduras, during the 2016-2017 harvest 9.5 million (46 kg) bags of coffee were exported, this represents around 5 % of the GDP (Gross Domestic Product) and approximately 30 % of agricultural GDP in Honduras generating profits for about 1 billion dollars per year. In recent years' coffee production has decreased significantly due to migration issues, leaving farms with no workers. Manual methods of coffee harvest take long times and are not cost effective. A single crop could take several days and its highly labor intensive. This research work aims to present a system that detects and classify a coffee fruit, which allows coffee producers to reduce costs, time and increase the quality of final product. Using a methodology with a qualitative approach and an experimental design, an algorithm was designed to classify the coffee fruit as “ripe” or “not ripe”. The deep learning algorithm was trained with 196 images, where 108 were positives and 88 negatives. As results the system can classify correctly 41/42 tests, having an efficiency of 97.6 %. Therefore, it is concluded that the recognition of coffee fruit with artificial vision and neural networks allows coffee producers to accelerate the process and increase quality.