Automatic Identification of Abaca Bunchy Top Disease using Deep Learning Models

aUsage of computer vision and artificial intelligence in the detection and identification of plant diseases has been explored and utilized in agricultural crops and had proven to perform efficiently. However, this disease detection and identification technology has not yet being explored and examined for some economically valuable crops like abaca and banana. This study intended to develop an automatic identification system for Abaca Bunchy Top Disease (ABTD) using different deep learning models. The study utilized a total of 3,840 petioles and petioles with leaves images taken using DSLR and mobile camera. Selected and preprocessed images were then subjected to augmentation techniques, normalization techniques, and morphometric and geometric analyses. Images were then trained using AlexNet, ZFNet, VGG16, and VGG19 architectures and the results were evaluated using Confusion Matrix in terms of accuracy, error rate, and precision. DSLR captured images on leaves and petioles with leaves showed an accuracy greater than 90% in all architectures except VGG16 with only 83% accuracy, while on mobile captured images, leaves showed above 90% accuracy compared to other groups. As to precision, DSLR captured images on petioles showed that out of four architectures, two models showed above 90% precision except for AlexNet and VGG16. However, for mobile captured images, three models showed above 90% precision using petioles image except VGG16. Furthermore, the models can be used for development of software application for detection, monitoring, and evaluation of ABTD. (C) 2021 The Authors. Published by Elsevier B.V.