A Computer Vision System Based on Majority-Voting Ensemble Neural Network for the Automatic Classification of Three...

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate crop cultivar identification is vital to prevent fraudulent sales and ensure specific applications.
• Human expert classification can be subjective and inconsistent due to factors like fatigue.
• Chickpea (Cicer arietinum L.) is a globally important legume with several visually similar varieties.

Purpose of the Study:

• To develop and present a computer vision system for the automatic classification of three distinct chickpea varieties: Adel, Arman, and Azad.
• To address the challenge of visually similar chickpea cultivars through automated analysis.
• To enhance the reliability and objectivity of chickpea variety identification.

Main Methods:

• Image segmentation using Hue Saturation Intensity (HSI) color space thresholding.
• Extraction of color and textural features (Gray Level Co-occurrence Matrix - GLCM) from chickpea images.
• Feature selection using a hybrid Artificial Neural Network-Cultural Algorithm (ANN-CA) to identify the five most effective discriminant features.
• Classification using an ensemble methodology (ANN-PSO/ACO/HS majority voting - MV) combining three hybrid classifiers (ANN-PSO, ANN-ACO, ANN-HS).

Main Results:

• The hybrid ANN-CA effectively selected five key features for classification.
• The ensemble ANN-PSO/ACO/HS-MV classifier achieved a high average classification accuracy of 99.10 ± 0.75%.
• The system demonstrated robust performance over 1000 random iterations on the test set.

Conclusions:

• The developed computer vision system provides a highly accurate and reliable method for automatic chickpea variety classification.
• The ensemble machine learning approach significantly enhances classification performance compared to individual classifiers.
• This technology offers a potential solution to combat chickpea variety fraud and ensure agricultural product integrity.