Jackfruit Fruit Damage Classification using Convolutional Neural Network

Abstract

Insufficient understanding on the incidence of plant pests and diseases as well as on the appropriate cultural practices against them may worsen the damage and caused a tremendous loss on fruit production. The use of mobile-based solution will significantly contribute on the availability and accessibility of human expert’s knowledge on this domain. In this study, a convolutional neural network was used and deployed on an android-based mobile application to perform detection and diagnosis of jackfruit fruit damages caused by pests (fruit borer and fruit fly) and diseases (rhizopus fruit rot and sclerotium fruit rot). The sequential type model was implemented which is mainly composed of 3 convolutional layers, each activated by a Rectified Linear Unit function and followed by a max pooling layer, and finally 2 dense layers. The model was trained using a total of 2409 images, and when evaluated on a validation dataset with 516 images, a loss rate of 4.58% and an accuracy rate of 97.93% were attained. On the other hand, when it was tested to predict on another set of 516 images, a remarkable success rate of 97.87% was obtained. The result indicates that the application can carry out a reliable and real time assessment on pest infestation and disease infection. Likewise, it provides recommendations on fruit protection measures.