Multi-threshold splitting tree algorithm to reduce the number of filters in programmable hyperspectral imaging for...

Area of Science:

• Optics and Photonics
• Computer Vision
• Spectroscopy

Background:

• Programmable hyperspectral imaging (PHI) enables rapid target classification by integrating algorithms into optical elements.
• PHI offers advantages like fast acquisition, no post-processing, and reduced data load compared to conventional methods.
• A key limitation of PHI is reduced speed in multi-target classification due to numerous required filters.

Purpose of the Study:

• To develop a novel splitting strategy for programmable hyperspectral imaging.
• To reduce the number of filters required for multi-target classification tasks.
• To maintain classification performance while enhancing speed in PHI.

Main Methods:

• A new splitting strategy was designed for programmable hyperspectral imaging.
• Numerical simulations were conducted using six public hyperspectral datasets.
• The proposed strategy was compared against conventional splitting methods.

Main Results:

• The novel splitting strategy reduced the filter count by 25% to 80%.
• Similar classification performance was achieved compared to conventional strategies.
• Significant improvements in the speed of multi-target classification were observed.

Conclusions:

• The proposed splitting strategy effectively reduces filter requirements in PHI for multi-target classification.
• This method enhances the speed of PHI for complex classification tasks.
• The technique holds significant potential for applications requiring rapid identification of multiple targets.