VirLeafNet: Automatic analysis and viral disease diagnosis using deep-learning in Vigna mungo plant

Various viral diseases affect the growth of the plants that causes a huge loss to farmers. If the viral infection could be noticed at earlier stages, then recovery procedures and respective action can be taken on time. Thus, there is a need for developing automatic viral infection detection methods for monitoring of crops analysing symptoms at different parts of plants. This paper proposes an automatic deep-learning-based viral infection detection method for a leguminous plant, Vigna mungo which is grown largely in the Indian subcontinent. Due to viral infection, some properties of the leaf image changes but the pattern is very random throughout the leaf structure. Hence, it is quite challenging to make an automatic disease detection method and perform the detection tasks in real-time. The collected image dataset of Vigna mungo leaves belonging to different categories are segmented and augmented to introduce more variety in the leaf image dataset. The convolutional neural network VirLeafNet is trained with different leaf images consisting of healthy, mild-infected and severely infected leaves for multiple epochs. The proposed methodology can be integrated with drones for wider crop area analysis. The proposed method is completely automatic, non-destructive and quickly classifies the leaf images of different categories in real-time. All the proposed models achieved high validation accuracy and yielded testing accuracy for VirLeafNet-1, VirLeafNet-2, and VirLeafNet-3 as 91.234%, 96.429%, and 97.403% respectively on different leaves images after extensive testing of the algorithm.