A Novel Transfer Learning Ensemble based Deep Neural Network for Plant Disease Detection

Abstract

The intelligent detection and diagnosis of plant diseases are one of the primary goals in sustainable agriculture. Although most disease symptoms are visible on plant leaves, it is time consuming and expensive process by manual observations. Automated detection of diseases is a significant concern in monitoring the plants to make timely decisions. The advent of recent deep learning models has led to several applications for automatic plant disease diagnosis. However, the diagnostic performance of these applications is substantially reduced when employed on test data sets due to overfitting. In this study, we propose a novel ensemble deep convolution neural network to classify the plant leaf diseases, and its performance was assessed with other benchmark deep learning models, namely, VGG16, ResNet152, Inceptionv3, DenseNet121. Three crops with 18 distinct categories were considered from the plant village dataset. Empirical findings show that the proposed model achieves 98.96% accuracy, significantly higher than other benchmark state-of-the-art models.