An Approach for the Customized High-Dimensional Segmentation of Remote Sensing Hyperspectral Images.

Area of Science:

• Computer Vision
• Image Processing
• Remote Sensing

Background:

• Hyperspectral image segmentation faces challenges including application-specific needs, lack of labeled data, and information loss from dimensionality reduction.
• Existing methods often project high-dimensional data to lower spaces, leading to information loss crucial for accurate segmentation.

Purpose of the Study:

• To propose a novel Multi-Gradient based Cellular Automaton (MGCA) structure for multidimensional image segmentation.
• To address the limitations of current hyperspectral image segmentation techniques, including data scarcity and information loss.

Main Methods:

• Developed the Multi-Gradient based Cellular Automaton (MGCA) structure for segmenting multidimensional images without projection.
• Coupled MGCA with an evolutionary algorithm (ECAS-II) to generate customized transition rule sets.
• Utilized low-dimensional training images to define user-specific segmentation requirements and generate high-dimensional segmenters.

Main Results:

• The proposed MGCA structure effectively segments hyperspectral images without information loss due to dimensionality reduction.
• Customized segmentation rules were successfully generated using evolutionary algorithms and limited training data.
• The strategy demonstrated robust performance on synthetic and real hyperspectral images, outperforming state-of-the-art methods.

Conclusions:

• The MGCA coupled with ECAS-II offers a flexible and effective solution for hyperspectral image segmentation.
• This approach alleviates the need for extensive labeled datasets by enabling online generation of training data.
• The method shows significant promise for applications in remote sensing and other fields requiring precise hyperspectral image analysis.