Performance and Applicability of Transfer Learners for Cocoa Swollen Shoot Detection

Abstract

An accurate and reliable cocoa swollen shoot disease diagnosis is the desire of traditional farmers with low-resolution smart devices. In this study, an efficient cocoa swollen shoot disease identification method base on transfer learners using pre-trained VGG16 and ResNet was proposed. These pre-trained models were trained using 456 samples and validated with 114 samples. The dataset constitutes low-resolution images, VGG16 and ResNet, and achieved an accuracy of 98.25 and 94.73%, respectively. With the objective of proposing a more reliable and accurate model, VGG16 is noted to scale better in terms of performance for implementation.