Dimensionality reduction in hyperspectral imaging using standard deviation-based band selection for efficient...

Area of Science:

• Biomedical optics
• Machine learning applications
• Data science in healthcare

Background:

• Hyperspectral imaging (HSI) generates large datasets with rich spatial and spectral information.
• Dimensionality reduction is crucial for managing HSI data size while preserving key spectral features.
• Band selection and feature extraction are primary dimensionality reduction strategies.

Purpose of the Study:

• To evaluate the standard deviation as a band selection method for HSI data.
• To assess the efficiency of standard deviation combined with a convolutional neural network (CNN) for organ tissue classification.
• To compare the standard deviation method with other band selection techniques like mutual information and Shannon entropy.

Main Methods:

• Applied standard deviation for band selection in hyperspectral imaging datasets.
• Utilized a convolutional neural network (CNN) for classifying organ tissues.
• Tested the method on eleven groups of diverse organ samples (100 datasets per group).
• Compared classification accuracy and data reduction efficiency against unprocessed data and other band selection methods.

Main Results:

• Standard deviation reduced data size by up to 97.3% while retaining essential spectral features.
• Achieved a classification accuracy of 97.21% with the standard deviation method.
• This accuracy is comparable to the 99.30% achieved without any dimensionality reduction.
• Demonstrated superior stability and efficiency compared to mutual information and Shannon entropy band selection methods.

Conclusions:

• Standard deviation is an effective and efficient band selection approach for hyperspectral imaging.
• This method significantly reduces data size without compromising spectral feature integrity for classification.
• The approach shows great potential for hyperspectral imaging classification tasks demanding large datasets and high processing speeds.
• Highlights the value of dimensionality reduction in hyperspectral imaging for medical applications.