Compressive hyperspectral image classification using a 3D coded convolutional neural network

Area of Science:

• Remote Sensing
• Computer Vision
• Optics

Background:

• Hyperspectral image classification (HIC) faces challenges due to large data volumes.
• Existing methods struggle with data acquisition, storage, transmission, and processing.

Purpose of the Study:

• To develop a novel deep learning approach for HIC using compressive measurements.
• To overcome the limitations of large hyperspectral data cubes.
• To improve classification accuracy by integrating hardware and deep learning.

Main Methods:

• A 3D coded convolutional neural network (3D-CCNN) is proposed for HIC.
• The coded aperture in CASSI is treated as a network layer.
• An end-to-end training method jointly optimizes network parameters and coded apertures.

Main Results:

• The 3D-CCNN approach achieves effective hyperspectral image classification.
• Classification accuracy is significantly improved by leveraging the synergy between deep learning and coded apertures.
• The proposed method demonstrates superiority over state-of-the-art HIC techniques.

Conclusions:

• The novel deep learning strategy based on CASSI compressive measurements offers an efficient HIC solution.
• Joint optimization of network parameters and coded apertures enhances classification performance.
• This method effectively addresses the challenges posed by large hyperspectral data cubes.