Phytopathogenic Classification from Durian Leaves Through the Convolution Neural Network

The quantity and quality of durian production rely on good agricultural practices, particularly for disease control. This research aims to develop an artificial intelligence prototype to classify durian pathogens based on leaf symptoms. Durian leaves expressing Leaf spots, Anthracnose leaf blight, Phytophthora leaf blight, Rhizoctonia leaf fall, and Algal leaf spots were collected from 40 durian orchards in Chanthaburi province, Thailand. Leaves exhibiting disease symptoms were photographed in the laboratory. The pathogens were isolated and identified from at least 30% of leaves in their genus. The correlation between the symptoms and pathogens was observed. The data from 7,691 leaves were used to train and validate the artificial intelligence system. A program was developed to compile a database of plant leaf symptoms and plant pathogens. An algorithm employing a convolutional neural network was constructed to identify and categorize durian leaves exhibiting the five disease symptoms, achieving an accuracy rate of 68.90%. This work could be further applied to detect durian symptoms.