Artificial Intelligence-Based Robust Hybrid Algorithm Design and Implementation for Real-Time Detection of Plant...

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Plant diseases significantly impact global food security and agricultural productivity.
• Manual visual inspection for plant disease identification is labor-intensive and error-prone.
• Automated, accurate, and computationally efficient plant disease classification is crucial.

Purpose of the Study:

• To develop a novel hybrid model for high-accuracy, low-complexity plant leaf disease classification.
• To integrate metaheuristic optimization for efficient feature selection in plant disease identification.
• To enable real-time plant disease classification using unmanned aerial vehicles (UAVs).

Main Methods:

• A hybrid model combining wrapper-based feature selection (Flower Pollination Algorithm and Support Vector Machine) with a Convolutional Neural Network (CNN) classifier was developed.
• Features were extracted using 2D Discrete Wavelet Transform (2D-DWT) across various wavelet families.
• The model was optimized for minimal features while maintaining high classification performance and deployed on an NVIDIA Jetson Nano for UAV-based real-time testing.

Main Results:

• The proposed hybrid model achieved high accuracy in classifying apple, grape, and tomato plant leaf diseases.
• The wrapper approach effectively selected a minimal set of features, reducing computational complexity.
• Real-time classification tests on UAVs demonstrated the model's practical applicability and precision.

Conclusions:

• The developed hybrid classification model offers a robust and efficient solution for real-time plant disease detection.
• Metaheuristic optimization significantly enhances feature selection for improved classification accuracy and reduced complexity.
• The model's successful UAV deployment highlights its potential to improve agricultural monitoring and management.