Citrus Greening Infection Detection (CiGID) by Computer Vision and Deep Learning

Abstract

The citrus greening infection detection algorithm is performed via computer vision techniques and deep learning for the purpose of extracting sub-images of fruit from a tree image and using a trained machine learning function to determine if the fruit shows signs of a citrus greening infection disease called Huanglongbing. We trained our deep learning inception model with 4000 iterations and achieved validation accuracy 93.3%. The computer vision fruit sub-image extraction resulted in at worst around 80% accuracy in tree images and was manually calibrated to detect a specific range of orange color values.