Investigation on methods of detection and classification of diseases in plant leaves

Abstract

Developing countries mainly rely on agricultural products for their economy. But plant yield is greatly affected due to diseases, which produce a huge financial loss as well as a threat to food security. The traditional approach involves naked eye observation of plant diseases by experts as well as by farmers. Rapid identification of disease is a challenging task as many farmers lack expertise and the disease vary from plant to plant as well as varies at each stage of growth, which requires continuous monitoring and is expensive for large farms. Further many farmers are unaware of non-native diseases. Moreover, rural areas lack experts which force them to go long distances to get a suitable solution. Recent technological advancements paved a way for a computer-aided approach to automatically detect and classify diseases. newlineVarious image processing methods are suggested in this study to identify and classify leaf diseases. The key goal of this study is to define and classify the unhealthy leaf portion, as well as to determine the best features that contribute to disease classification. A novel training function for neural network classifiers is also proposed to improve the classification accuracy of disease detection. In addition, a novel deep learning model is proposed to increase classification accuracy. The research study used a traditional PlantVillage dataset and a soursop dataset to accomplish these goals. newlineGray level co-occurrence matrix (GLCM) and statistical parameters were used in the study. A novel crest factor-based segmentation algorithm to detect leaf diseases in apple, cherry, strawberry, peach and corn, even in the presence of specular noise is proposed. newline