CNN and SVM-based Model for Effective Watermelon Disease Classification
Deepak Banerjee, Vinay Kukreja, Amit Gupta, Vijay Singh, Tejinder Pal Singh Brar
Abstract
Watermelon farming is hampered by a variety of illnesses that harm its leaves, resulting in lower quality and yield of crops. We present a unique strategy for the precise identification of watermelon disease of the leaves using a model of deep learning in this study. Our model is made up of four convolutional layers, four max pooling levels, and one fully connected layer. Bacterial Fruits Blotch, Anthracnose, Gummy Rope Blight, Downy Mildew, Alternatives Leaves Spot, Cercospora Leaves Spot, Myrothecium Leaves Spot, and powdery Mildew are the eight watermelon leaf diseases to be predicted. For each disease, we analyse recall, precision, and F1-score to determine the model's performance. The obtained results show potential accuracy and efficacy in illness classification. Precision rates vary with the disease, with Anthracnose getting 76.00%, while values for recall also vary, with Gummy Trunk Blight reaching 75.56%. Overall, the model performs well with F1 scores, especially for Anthracnose (73.17%) and Gummy Trunk Blight (73.12%). Furthermore, the micro-average precision of 70.29% and weighting average precision of 70.26% show the model's capacity to manage imbalanced class distribution. These findings highlight the importance of our deep learning-driven technique in assisting in the early detection and discrimination of watermelon leaves diseases, hence contributing to more effective crop health monitoring and assuring long-term watermelon production.