Agriculture industry is one of major industry in Thailand. Nowadays, agriculture industry uses more innovation for development, especially image processing technology and computer vision. One of important problem in agriculture industry is overdose on chemicals to handle plant diseases that cause many problems such as pollutions, health hazards and expensive cost. So, if we can initial plant disease diagnosis, we can reduce productivity damage and avoid overdose on chemicals. This research presents algorithm for grape leaf disease diagnosis from complex background imagery. The system consists genetic algorithm (GA) and structure-adaptive self-organizing maps (SASOM) called GA-SASOM algorithm, which we developed basic GA structure to be main structure of classification system with new chromosome type that developed from basic SASOM node map called chromosome map. This research used each chromosome map to represent each color characteristics and pattern class of leaf disease, which was feature extraction model of grape leaf disease image. This work was test grape leaf disease diagnosis system with 4 diseases include Scab, Rust, Downy mildew and Powdery mildew, which each image had various size, position, rotation and light condition. These system achieved a performance up to 94.35% of accuracy.
Keywords: Pattern classification, Genetic Algorithm, Disease diagnosis, Structure-adaptive self-organizing maps