Computer Vision Based Watermelon Plant Disease Prediction Using Multi-Scale Convolutional Neural Networks In Deep Learning Models

Watermelon is a global staple crop, yet its cultivation faces significant threats from diseases suchas Anthracnose, Downy Mildew, and Mosaic Virus. Timely and accurate detection of these diseases iscritical to prevent yield loss and ensure food security. Traditional visual inspection methods are labor-intensive, subjective, and prone to error. This research proposes a novel deep learning framework based on aMulti-Scale Convolutional Neural Network (MS-CNN) for the automated classification of watermelondiseases. The proposed architecture integrates parallel convolutional layers with varying kernel sizes tocapture multi-granular features, enabling robust identification of both fine texture details and large structuralpatterns. A dataset of 4,000 images, encompassing four classes (Healthy, Anthracnose, Downy Mildew, andMosaic Virus), was curated and augmented to train the model. Experimental results demonstrate that theMS-CNN achieves a classification accuracy of 98.2%, significantly outperforming state-of-the-art transferlearning models including VGG16, ResNet50, and MobileNetV2. The model exhibits high precision(98.0%) and recall (97.9%), validating its efficacy for real-time agricultural applications. This studycontributes a reliable, non-invasive, and automated tool for disease management, potentially aiding farmersin early diagnosis and targeted treatment.