Deep learning based detector for identification of woodiness disease in passion fruits

Abstract

In this project, Artificial Intelligence based algorithms were developed based on simple leaves to detect woodiness disease, a passion fruit disease in presence of the other classes of diseases through deep learning methodologies. Training of the algorithms was performed with the use of a data-set of 10,510 images that contained several distinct classes of plant disease combinations including the healthy plants that were obtained after visiting several farms in Hoima and Luwero using a 24MP NIKON D5300 with an APS-C sensor camera. Training and Testing were done with frameworks(Faster Regional Convolutional Neural Network V2 COCO with ResNet and Single Shot Multibox detector with ResNet), using a single graphical processor Unit (GPU), the NVIDIA-1080TI. The best performance reached a 99% success rate in identifying woodiness disease with Faster RCNN with ResNet framework. The significantly high success rate makes the model a very useful advisory or early warning tool, and an approach that could be further expanded to support an integrated plant disease identification.