A Proficient Convolutional Neural Network for Detecting Watermelon Disease with Occlusion Sensitivity

Watermelon is a popular summer fruit around the world. Especially in Bangladesh there is a huge demand for watermelons and there is always a big harvest of watermelons. Recent studies have shown watermelon has multiple health benefits, especially with regard to intestinal and kidney safety. Just like any other fruit watermelon is prone to diseases. Failure to identify those disease in times results in bad harvest. The aim of this research is to create an automated system that will assist farmers in detecting diseases to get better harvest. The dataset used in this research has 4 classes: Anthracnose, Downey Mildew, Mosaic virus and healthy. The dataset has been augmented, and the augmented dataset has 5775 images. A model using deep convolution neural network architecture (CNN) has been deployed. The model has 25-layers in total. The has been optimized by proper hyper parameter tuning. The model achieved perfect accuracy of 99.68%(std:0.0004) (K=5). Occlusion Sensitivity method has been utilized to interpret the model performance and for validation K-fold validation was applied.