A Method for Detection of Corn Kernel Mildew Based on Co-Clustering Algorithm with Hyperspectral Image Technology

Area of Science:

• Agricultural Science
• Image Processing
• Machine Learning

Background:

• Hyperspectral imaging is crucial for non-destructive grain quality assessment, providing spectral and spatial data.
• Supervised learning methods dominate hyperspectral mildew detection in corn but require extensive training data.
• Existing methods struggle with complex mildew distribution patterns and computational complexity.

Purpose of the Study:

• To develop an unsupervised algorithm for detecting non-uniformly distributed mildew in corn kernels.
• To overcome limitations of traditional fuzzy c-means and spectral clustering algorithms.
• To enhance the accuracy and efficiency of mildew detection in grain quality analysis.

Main Methods:

• An unsupervised redundant co-clustering algorithm (FCM-SC) was developed, combining multi-center fuzzy c-means (FCM) and spectral clustering (SC).
• The algorithm performs FCM clustering, extracts redundant cluster centers, and merges them using SC.
• Sample features are classified by assigning them to the identified cluster centers.

Main Results:

• The FCM-SC algorithm effectively describes complex mildew distribution patterns in corn kernels.
• The proposed method demonstrated superior stability, anti-interference, generalization, and accuracy compared to supervised models.
• It addresses the limitations of traditional algorithms in handling complex data structures and computational demands.

Conclusions:

• The unsupervised FCM-SC algorithm provides a robust and accurate solution for mildew detection in corn kernels using hyperspectral imaging.
• This approach offers significant advantages over supervised methods, particularly when training data is limited or mildew distribution is complex.
• The study highlights the potential of advanced unsupervised learning techniques for non-destructive grain quality assessment.