Novel segmentation and classification algorithm for detection of tomato leaf disease

Summary The prevalence of tomato leaf diseases should be diagnosed in early‐stage to prevent spoilage of the entire field. Manually checking tomato diseases consumes more time and is labor‐intensive. In modern agriculture, machine and deep learning‐based disease identification techniques have been developed to effectively classify diseases. Most of the existing methods are inappropriate for horticulture due to their incompetence in handling the complex backgrounds of the image. In this article, a novel segmentation and classification algorithm is proposed for detecting tomato leaf diseases with complex background interference based on leaf segmentation fuzzy CNN (LSFCNN) and ant colony‐based mask RCNN (AC‐MRCNN). Foremostly the collected images are annotated and enhanced for further processing. Then the novel LSFCNN is implemented to separate the tomato leaf in a complex background. For classification, AC‐MRCNN is developed, which masks the disease spot and recognizes the diseases. Herein ant colony optimization algorithm is utilized to optimize the mask RCNN to increase the flow of information and gradients of the network. Over 14,817 uniform and complex background images are collected to train the model. The proposed method is profoundly effective for quite challenging background leaf disease classification, with an accuracy of 97.66% of eight diseases and one healthy class.