Integrating SOMs and a Bayesian classifier for segmenting diseased plants in uncontrolled environments

Area of Science:

• Agricultural technology
• Computer vision
• Machine learning

Background:

• Plant disease detection is crucial for crop management.
• Uncontrolled environments like greenhouses present challenges in image segmentation due to variable lighting and background clutter.

Purpose of the Study:

• To develop an robust image segmentation methodology for identifying diseased plants in uncontrolled environments.
• To improve the accuracy of plant disease detection compared to existing color index methods.

Main Methods:

• A hybrid approach integrating unsupervised (Self-Organizing Map - SOM) and supervised (Bayesian classifier) learning.
• Utilized two SOMs for color grouping and error correction, generating color histograms for Bayesian classifier training.
• Developed a post-classification analysis to identify and recover misclassified diseased plant areas.

Main Results:

• The proposed methodology demonstrated superior performance over two common color index methods for plant disease segmentation.
• Successfully segmented diseased plant areas, even those initially misclassified as non-vegetation.

Conclusions:

• The integrated SOM and Bayesian classifier approach offers a more accurate and reliable solution for plant disease segmentation in challenging agricultural settings.
• This method enhances the potential for automated disease monitoring in greenhouses.