Automatic Identification of Diseases and Pests for Thai Rice Leaf from IoT Camera Using ResNet50

Abstract

This paper presents the combination of IoT cameras and the ResNet50 classification technique to remotely detect and identify rice diseases and pests. The training image data was collected from the locals of Ron Thong, Satuek District, Buriram Province, Thailand. The focused rice anomalies for detection are five common diseases and three types of pests in Thailand, including rice blast disease, bacterial leaf blight disease, rice tungro disease, sheath blight disease, brown spot disease, brown planthopper, green rice leafhopper, and rice gall midge. The annotated local images are trained for the best compatibility with the local environment. For detecting anomalies, installed IoT cameras are set to capture images of rice leaves within the field three times a day and upload the image as an input to the cloud API, which contains the classification model to detect the symptoms of disease and pest. If a disease or pest is detected, the system automatically alerts with the identified anomaly to responsible field workers. The experiment results show that the performance of the model from the ResNet50 technique achieves a satisfied result of 0.956 F1 score.