Durian Trunk Diseases Identification Using Deep Learning

The demand for durian fruits has increased significantly since Malaysia and China adopted an export protocol for frozen durian pulp. However, Thailand, where durian is not a native fruit, remains the largest exporter of durian fruit globally, followed by Malaysia. The harvest of durian is affected by several diseases that attack the trees throughout the year. The most serious ones can kill trees and affect both immature and mature fruits. In large-scale farming in Malaysia, high-tech planting techniques are rarely adopted, and farmers often lack sufficient knowledge to identify diseases and pests that can reduce agricultural yields. Additionally, there is no publicly available dataset for diseases affecting durian tree trunks. The goal of this project is to identify durian trunk diseases. The objectives have been achieved, including gathering image data from several durian farms and categorizing the images to create a dataset of durian trunk diseases. The ideal deep learning approach for detecting durian trunk disease was identified following a thorough literature review. In this study, the architectures of MobileNetV2 and EfficientNetV2-B0 were compared using the transfer learning feature extraction technique. The best-performing model, achieving 90.4% accuracy for durian trunk disease identification, was determined to be MobileNetV2. The dataset generated by this project can serve multiple purposes, such as supporting reliable analyses or being used to train other machine learning models. This work helps farmers identify and prevent diseases early, thereby increasing durian production.