Deep Learning With Convolutional Neural Networks For Cassava Leaf Diseases Via Line Bot: A Case Study Of Buriram Provincial Protection Service Group, Buriram Provincial Agriculture Office

Abstract

This research aims to study and compare the effectiveness of deep convolutional neural networks (DCNN) suitable for detecting cassava leaf diseases. The objective is to develop a system for detecting cassava leaf diseases using a convolutional neural network through Line Bot, evaluate its performance with the assistance of experts, and assess the satisfaction of farmers who grow cassava regarding the usability of the system. The research tests the performance of four CNN architectures: MobileNetV2, NAS Net Mobile, EfficientNetV2B0, and EfficientNetV2B1, by adjusting the learning rates to 0.01 and 0.001 and comparing the results with and without data augmentation techniques using the iCassava 2019 dataset. The evaluation of the system was conducted with a sample group consisting of three expert evaluators and 30 farmers who grow cassava under the supervision of the Plant Protection Service Group, Buriram Provincial Agriculture Office. The assessment of the system's usability was obtained through purposive sampling. The research findings revealed statistically significant differences in the accuracy of cassava leaf disease classification among the four tested architectures at a significance level of 0.01. The highest accuracy of 88.43% was achieved using the MobileNetV2 model combined with data augmentation. This system was further developed to detect cassava leaf diseases using the camera on a smartphone to capture images of abnormalities on cassava leaves. It also provided information on disease treatment to prevent the spread of diseases to nearby areas through Line Bot. Overall, the evaluation of the system's effectiveness by experts found it to be at a good level, and the overall satisfaction with its usability was found to be at a very good level.