Eight Convolutional Layered Deep Convolutional Neural Network based Banana Leaf Disease Prediction
M. Shyamala Devi, Baddela Vamshikrishna, J. Arun Pandian, Paladugu Venkata Rao, Shaik Rusum Yaseen
Abstract
One of the most significant fruit crops in the world is the banana, however due to the variety of banana leaf diseases, it can be challenging for gardening experts to detect them. The incorrect diagnosis, however, lowers production effectiveness. The application of deep learning and computer vision technology for plant disease diagnosis has risen to prominence with the advancement of computer science. With this influence, this paper propose a novel Eight Convolutional Layered Deep Convolutional Neural Network to forecast the banana leaf disease with maximum accuracy. The model exploits the Banana leaf dataset retrieved from the KAGGLE machine-learning repository. The Banana leaf dataset contains four disease classes as Cordana, Healthy, Pestalotiopsis and Sigatoka with 937 banana leaf images and preprocessed for improving the image quality and resolution. The proposed Eight Convolutional Layered Deep Convolutional Neural Network had designed with single input, dense, average pooling and output layer. The model built with 8 convolution layers accompanied with max pooling layer for each convolution. The banana leaf dataset splitted with 777 training images, 80 validation images and 80 testing images. The banana leaf training dataset have applied to Eight Convolutional Layered Deep Convolutional Neural Network, same dataset was applied to VGGl9Net, Resnet152, MobileNetV2, DenseNet201, InceptionV3Net and NASNet Large models for analyzing the performance metrics. For Implementation, 30 training epochs with batch size of 64 was used on NVidia Tesla V100 GPU server. Implementation shows that the proposed method gives the accuracy of 98.75%, Recall of 98.75%, Precision of 96.42% and FScore of 97.57% in comparison with other convolutional neural network models.