Deep learning-based classification of coconut leaf diseases using optimized CNN models

Abstract

Leaf diseases of coconut severely affect the yield and quality of the crops. Early detection is imperative for any control measures to be effective. Here, an approach on the deep learning-based classification of coconut leaf diseases is presented with a dataset consisting of five categories. We fine-tuned five deep CNN models-ResNet50, VGG16, InceptionV3, MobileNetV2, and DenseNet121-by freezing their base layers and adding our custom classification layers on top. All the models were trained on the augmented dataset with rotation, zooming, sharing, and flipping for better generalization. The performances of the networks studied are as follows: Custom InceptionV3 yielded the highest validation accuracy, 96.22%, followed closely by MobileNetV2 and DenseNet121, which had an accuracy of 95.52%. Furthermore, his system could be extended for mobile deployment, enabling real-time disease diagnosis directly from field images captured by smartphones, which would further enhance accessibility for farmers and support precision agriculture practices. Thus, the model system provides live performance visualization on accuracy and loss metrics; therefore, model selection should also be enabled. Scalability and performance in this presented research on Coconut Leaf Disease Classification will highly contribute to precision agriculture and help farmers take proactive approaches toward crop protection.