Enhanced Coconut Tree Disease Classification Using ResNet50, EfficientNetB0, and DenseNet-201 with Comparative Analysis of Activation Functions

Accurate and timely prognosis of coconut tree diseases is essential for good crop health, high yields, and economic endurance in agricultural coconut trees, which are dominant in tropical economies. The paramount susceptibility of coconut trees to a variety of diseases can significantly reduce yields and impose a significant economic burden on farmers. Early pinpointing and intercession to protect the overall health of the trees is also important. To overcome symptomatic challenges and address the often unpredictable manifestations of coconut tree diseases, this study uses a novel approach employing advanced deep learning techniques. We used state-of-the-art image analysis techniques to extract key features from leaf images of coconut trees affected by various diseases like BudRoot Dropping, BudRot, Gray Leaf Spot, Leafrot, StemBleeding. Specifically, we used DenseNet-201, ResNet-50, and EfficientNet B0 algorithms to encapsulate complex visual structures associated with different disease stages. To increase the discriminative power of the model, we trained models using four activation functions: ReLU, LeakyReLU, Swish, and ELU individually. Through a pervasive evaluation process, our model achieved a remarkable accuracy of 99.85% in classifying coconut tree diseases, demonstrating its exceptional performance in differentiating between disease types.