Enhancing Watermelon Diseases Detection using Dense-EfficientNet and Explainable AI

Abstract

Watermelon, Citrullus lanatus, is an edible fruit of the flowering plant species of the Cucurbitaceae family. This South African fruit is highly cultivated worldwide with more than 1000 varieties. Like all fruits, watermelon also has some major diseases such as Downy Mildew, Anthracnose, and Mosaic Virus which are caused by Pseudoperonospora cubensis, Colletotrichum orbiculare, and Watermelon mosaic virus respectively. While watermelon exportation and cultivation helps with huge economic sectors, these diseases are continuing to hold the production back, making it difficult for the farmers as well. Traditional methods for detecting these diseases are expensive, time-consuming, and require years of experience, whereas Deep learning offers promising solutions for early and accurate detection, which are cheap, less time consuming, and standard maintained procedure. Proposed deep learning model is DenseEfficient which combines the strengths of DenseNet121 and EfficientnetV2B0. DenseNet121 achieves high accuracy and efficiency by connecting each layer to every other layer, with up to 121 or 201 layers, enhancing feature reuse. Additionally, it incorporates the stability and accuracy of EfficientnetV2B0’s progressive learning rate and advanced neural architecture search to enhance the image recognition tasks. Image Resize, Color inversion Image Augmentation and Outlier handling has been used as pre-processing techniques. In this study, the dataset had 1,115 images of five different classes of watermelon. Using the DenseEfficient model, the generated accuracy was 98.45% where the precision, recall, and F1-score were 0.9993, 0.9981, and 0.9991 respectively. These results inform that DenseEfficient can correctly categorize watermelon diseases and due to its quick performance in classification, it can be very useful tool for increasing watermelon cultivation.