Paddy Disease Recognition and Classification using Artificial Neural Networks: Introducing a Novel Hybrid Optimization Algorithm
No Thumbnail Available
Date
2024-03-20
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Paddy diseases have significantly deteriorated the production capacity of the paddy plant,
resulting in substantial economic losses. It becomes imperative to identify paddy diseases
early to effectively mitigate and minimize the detrimental impacts of these diseases. Timely
detection of infections enables effective management strategies and leads to enhanced yield
in terms of quantity and quality. Hence, it is imperious to develop a methodology that can
swiftly and accurately detect paddy diseases. Artificial Neural Networks (ANNs) have
emerged as a promising approach for paddy disease recognition. Nevertheless, training ANNs
pose significant challenges, particularly when dealing with intricate networks containing
numerous parameters. Prior research has explored single optimization algorithms for ANN
training; however, these algorithms are hindered by constraints such as convergence
problems, limited exploration capabilities, inadequate adaptability, and inconsistent
performance across diverse problem domains. Consequently, there is a pressing demand for a
hybrid optimization algorithm that can effectively address these challenges by offering an
optimal solution for optimizing ANN parameters. Therefore, this paper aims to devise a novel
hybrid optimization algorithm, namely Adaptive Particle Swarm Water Wave Optimization
(APSWWO), for finding the disease in the paddy plant, where the training of ANN is
completed using the APSWWO, which is formed by assimilating two nature-inspired
optimization algorithms Adaptive Particle Swarm Optimization (APSO) and Water Wave
Optimization (WWO). Initially, the images of paddy leaf disease are pre-processed and
balanced. Two customized ANN models: 26-layer Convolutional Neural Network model and
13-layer Lightweight Vision Transformer (LiteViT) are used to classify ten types of paddy
diseases and one healthy class. The optimization of parameters in the training process is
achieved by incorporating the APSWWO algorithm. This study also employs Explainable AI
techniques such as Gradient-weighted Class Activation Mapping (Grad-CAM) to provide
model interpretability and visualizations in each layer, enabling more transparent and
trustable disease detection. Accuracy, recall, precision, and F-measure metrics are utilized to
assess classification performance. The hybrid approach's effectiveness is evaluated by
comparing ANN techniques performance with and without optimization (Original, PSO,
APSO, WWO, and APSWWO), highlighting the importance of hybrid APSWWO in
enhancing prediction performance. The hybrid CNN_APSWWO outperforms other forms
(CNN_Original, CNN_PSO, CNN_APSO, CNN_WWO, LiteViT_Original, LiteViT_PSO,
LiteViT_APSO, LiteViT_WWO, and LiteViT_APSWWO), offering valuable insights for
optimizing paddy disease recognition and classification, ultimately enhancing agricultural
practices and contributing to improved crop health and food security.
Keywords: Paddy; Disease; Artificial Neural Network; Nature-inspired Optimization; Hybrid
Algorithm; Explainable AI