An Efficacious Deep Fusion based Tomato Leaf and Fruit Classifier using Convolutional Gated Recurrent Unit and Bonobo Optimizer

Abstract

Agriculture is suffering a drastic blow because of sudden changes in environmental conditions and novel pathogenic attacks. These attacks have brought in heavy disruptionto the integrity of the food supply chain. In order to meet the growing global food demand,an automated system is needed for early identification of diseases and to enable smart farming with the advent of upcoming technologies. This paper proposes such an expert system that augments farming returns both economically and productively. Tomato leaf images are obtained from Kaggle repository and fruit images are self- captured, both making a total of 3676 images. Data is augmented using techniques like rotation, zooming, width shift, height shift, horizontal flip etc.Under preprocessing block, Wiener filter and Contrast Limited Adaptive Histogram Equalization model are used for noise smoothing and contrast enhancement.The U2Net architecture of convolutional neural networks is used for image segmentation after whichfeatures are extracted from a fusion layer based on InceptionV3 and EfficientNetB2 models. The final stage involves aConvolutional Gated Recurrent Unit classifier along with Bonobo optimizationfor obtaining optimal results.Evaluation metrics like accuracy, precision, recall, F1 score,Matthew’s correlation coefficient are calculated to find the effectiveness of the proposed system. The proposed algorithm is compared with currently prevailing algorithms like InceptionV3, MobileNet, VGG16, CNNandKernel Extreme Learning Machine and is found to produce promising results with an astounding accuracy of 96.90%.