Investigation on Data Fusion of Multisource Spectral Data for Rice Leaf Diseases Identification Using Machine...

Area of Science:

• Agricultural Science
• Spectroscopy
• Plant Pathology

Background:

• Rice diseases significantly impact global crop yield and quality.
• Early and precise detection of rice diseases is crucial for effective management and prevention.
• Spectroscopic techniques offer promising non-destructive methods for plant disease identification.

Purpose of the Study:

• To evaluate the efficacy of visible/near-infrared hyperspectral imaging (HSI), mid-infrared spectroscopy (MIR), and laser-induced breakdown spectroscopy (LIBS) for detecting three major rice diseases.
• To investigate the impact of different data fusion strategies (raw, feature, and decision fusion) on disease identification accuracy.
• To develop and compare machine learning models for rice disease classification using spectral data.

Main Methods:

• Three spectroscopic techniques (HSI, MIR, LIBS) were employed to collect spectral data from rice plants infected with leaf blight, rice blast, and rice sheath blight.
• Principal Component Analysis (PCA) and Autoencoder (AE) were utilized for feature extraction from the spectral data.
• Support Vector Machine (SVM), Logistic Regression (LR), and Convolutional Neural Network (CNN) models were trained and validated for disease classification at various data fusion levels.

Main Results:

• Hyperspectral Imaging (HSI) based models demonstrated superior performance compared to MIR and LIBS, achieving over 93% accuracy on the test set using PCA features.
• Feature fusion and decision fusion strategies significantly improved the identification performance of rice diseases.
• The study confirmed the potential of combining different spectroscopic techniques and data fusion for enhanced rice disease detection.

Conclusions:

• Visible/near-infrared hyperspectral imaging (HSI) is a highly effective technique for rapid and accurate identification of key rice diseases.
• Data fusion approaches, particularly feature and decision fusion, offer substantial benefits for improving the robustness and accuracy of spectral-based disease detection systems.
• The integrated application of advanced spectroscopic methods and data fusion holds significant promise for developing next-generation tools for precision agriculture and crop health management.