Application of microscopic image processing and artificial intelligence detecting and classifying the spores of three novel species of Trichoderma

Abstract

Abstract Trichoderma is a type of fungus genus usually used in agriculture to prevent and control various plant diseases. One of the ways Trichoderma helps control plant diseases is by producing secondary metabolites that prevent the growth of pathogens. The ability to identify Trichoderma spores is important to detect the presence of this fungus in plants and soil. However, traditional microscopic techniques used for spore detection are time-consuming and tedious, and require the intervention of highly skilled technicians, leading to delayed diagnosis and ineffective control measures. Microscopic image processing techniques have been developed to make the spore detection process intelligent, resulting in a faster and more accurate output. In this study, a microscopic image dataset has been developed, followed by identifying and classifying the spores of three Trichoderma species including T. harzianum, T. atroviride, and T. virens using microscopic image processing techniques. The genetic algorithm was used to identify the most effective visual features for spore classification, including color, texture, and shape. Then, using the identified effective features, the random forest was capable of classifying the spores with an accuracy of 95.38%. Also, the classification precision results for T. harzianum, T. atroviride, and T. virens were obtained as 100, 90.48, and 95.24% respectively. This study showed that texture is the most important visual feature for spore classification. Besides, the findings reveal that instead of deep learning-based methods which require big data for training, traditional feature extraction methods still provide promising results with low computational complexities. The use of such new techniques can help control plant diseases more effectively by providing faster and more accurate detection of Trichoderma spores in crop plants.

Keywords

• Biological control
• Genetic algorithm
• Feature selection
• Microscopic image dataset
• Machine learning