Hyperspectral Leaf Image-Based Cucumber Disease Recognition Using the Extended Collaborative Representation Model.

Area of Science:

• Agricultural Science
• Computer Vision
• Spectroscopy

Background:

• Collaborative representation (CR)-based classification is effective for plant disease recognition with ample training data.
• Acquiring sufficient training samples for plant diseases is challenging due to time and labor constraints.
• Environmental factors like poor illumination and leaf occlusion can corrupt sample data, hindering accurate classification.

Purpose of the Study:

• To develop an extended collaborative representation (ECR)-based classification model for cucumber leaf disease recognition.
• To address challenges posed by limited training samples and corrupted data in plant disease diagnosis.
• To enable fast and accurate identification of cucumber diseases using spectral analysis.

Main Methods:

• Constructed a pure spectral library with representative samples for each disease.
• Designed a universal variation spectral library to handle linear variables in samples.
• Encoded query samples using a linear combination of atoms from both libraries.
• Determined disease identity based on minimal reconstruction residuals.

Main Results:

• Achieved diagnostic accuracy exceeding 94.7% for cucumber anthracnose and brown spot.
• Demonstrated a short average online diagnosis time of approximately 1 to 1.3 milliseconds.
• Validated the feasibility of the ECR model for spectral curve analysis of leaf diseases.

Conclusions:

• The extended collaborative representation (ECR) model is effective for cucumber leaf disease recognition.
• ECR classification offers a feasible solution for fast and accurate plant disease diagnosis, even with imperfect data.
• This approach enhances the reliability of automated plant disease identification systems.