Plant Disease Detection and Classification using CNN Model with Optimized Activation Function

This paper deals with the optimized real time detection of diseases that affect the plant and the area affected using Convolutional Neural Networks (CNN) algorithms so that appropriate fertilizers can be used to prevent further damage to plants from pathogenic viruses. The activation function is the core of the CNN model as it incorporates the non - linearity to have an authentic artificial intelligence system for classification. ReLu is one among the best activation functions, but has a disadvantage that the derivative of the function is zero for negative values and leads to neuronal necrosis. New mathematical activation function is developed and compared with existing activation functions to improve the accuracy and performance of the system on a TensorFlow framework. The experimental results on trained databases show that the developed activation function has improved the CNN model accuracy and performance i.e. 95%. The training speed of the CNN model is improved by 83% when implemented in ARM processor using the proposed optimizer. Further area affected by disease is calculated by using K - means clustering algorithm for optimization of fertilizer usage.