Identification of Sigatoka Leaf Spot Disease in Banana Using Convolutional Neural Network (CNN)

Abstract

Aim: Rapid and Early identification of the cause of the disease enables prompt selection of the protection method and reduces the yield loss. Lag in disease diagnosis reduces the crop output and increases the cost of cultivation. To overcome this problem, deep learning techniques are used to identify Sigatoka leaf spot disease in bananas through image detection. CNN is a dependable method to identify the disease at the initial stages to help farmers.
 Place and Duration of the Study: Agricultural College and Research Institute, Tamil Nadu agricultural university, Coimbatore. Duration of the experiment was three months.
 Methodology: Image of diseased leaves and healthy leaves were collected from the different areas of Pollachi and Coimbatore, Tamil Nadu, India. Variations in colors were removed and the quality of the image was enhanced to increase the accuracy. The image dataset comprised of a total of 2008 images of banana Sigatoka leaf spot and healthy banana leaves for differentiation and evaluation. Initial stage of training includes loading of the image data for training, determining the learning rate, running the optimizer, and compiling the training convolution model. The model's accuracy is assessed in the last step, saving the accuracy and loss occurred during training.
 Results: The experiment result shows that disease detection accuracy of the intended model is 96.41%. It gives higher accuracy and improved performance.