Median interacted pigeon optimization-based hyperparameter tuning of CNN for paddy leaf disease prediction

Abstract

Image processing is used for identifying and diagnosing rice leaf diseases in the field of agricultural information. However, in the paddy leaf, identifying fungal infections like powdery mildew, and viral infections are complex. Hence, a novel, “Median Interacted Pigeon Optimization-based Hyperparameter Tuning of CNN for Paddy Leaf Disease Prediction”, has been proposed, in which the existing works focus on size, shape, and texture for leaf disease identification, overlooking fungal disease (powdery mildew) branching patterns and making segmentation more challenging. Thus, a novel Coherent Point Graph Recurrent Network (CPGRN) is introduced, which captures structural branching patterns and recurrent neural networks for temporal coherence, enabling precise segmentation of fungal hyphae. Furthermore, to extract relevant features from images of rice leaf diseases, Convolutional Neural Networks (CNNs) require efficient hyperparameter tuning. Thus, a novel Median Interacted Pigeon-Inspired Optimization (MIPIO) is proposed, which optimizes CNN hyperparameters to enhance the accuracy of characterizing fungal infections and enable the recognition of antagonist interactions among virus species. Moreover, the existing virus identification techniques struggle with antagonistic interactions. To address the unpredictable synergistic effects of multiple viruses co-infecting rice plants and detect co-infections of various viruses, a novel Dynamic Bayesian Adaptive Aesthetic Learning (DBAAL) is proposed, which highly assists in improving the prediction of viral infections in paddy leaves. The experimental results confirm that the proposed approach enhances prediction accuracy, also helps in efficient identification of co-infections of different viruses in rice plants.

Graphical Abstract