Automatic deep neural network hyper-parameter optimization for maize disease detection

Abstract

Deep Convolutional Neural Networks (DCNNs) have proved to be very useful for image classification. These need to be optimized for the dataset in hand, for its optimal use. DCNNs have many attached hyper parameters. These hyper-parameters are fine-tuned for optimizing the DCNN model. The present research uses the Bayesian Optimization technique for fine-tuning the hyper-parameters for AlexNet DCNN model. The model is trained and tested on Maize (corn) disease sub-dataset of the Plant Village dataset. The trained DCNN model with optimized parameters achieved the accuracy of 96.05%.