Disease Diagnosis in Grapevines–A Hybrid Resnet-Jaya Approach

Piyush Mishra, Puspanjali Mohapatra, Tapas Kumar Patra, P Subham

Abstract: Different diseases in grapevines have different kinds of effects on the various parts of the plant, the most drastic of such abnormalities easily seen in the leaves of the grapevines. Detection, diagnosis and prevention of diseases that could hinder the production and utilisation of the grapevine are of prime importance in viticulture. In view of the multiple hazards associated with viticulture, it is worthwhile to build on the optimisation and automation in the field. This study presents a methodology to marry the different aspects of computational mechanisms, i.e. neural network approaches along with different optimisation approaches, and apply them to diagnose the diseases that a said grapevine might suffer from. The objectives were, thus: (1) to be able to identify which disease corresponds to the effects shown in the leaves of the particular grapevine, (2) to make a robust identification mechanism by using methodologies of deep learning and soft computing, and (3) to make use of different such methodologies and infer which of those methodologies would work the best in this context. As a result, we mostly compared a 10 layer-feed forward neural network with a 34 layer-Resnet. We also employed other CNN methodologies like Densenet, VGG and Alexnet but found that they did not provide the best results in this context. Further, we hybridised these approaches with several optimisation algorithms, like the Jaya Algorithm, the Genetic Algorithm, the Particle Swarm Optimisation Algorithm, etc. We found that a hybrid Resnet-Jaya model gave the best output without overfitting, at 99.71%.