Hyperspectral Image Classification Using Deep Genome Graph-Based Approach.

Area of Science:

• Computer Science
• Machine Learning
• Remote Sensing

Background:

• Hybrid models combining 3D and 2D Convolutional Neural Networks (CNNs) are popular for hyperspectral image classification.
• Biological genome graphs enhance genomic analysis scalability and accuracy.

Purpose of the Study:

• To propose an innovative deep genome graph-based network (GGBN) for hyperspectral image classification.
• To leverage the strengths of hybrid CNN models and genome graphs.

Main Methods:

• The GGBN model employs a hybrid architecture with 3D-CNNs in lower layers and 2D-CNNs in upper layers.
• It processes spectral-spatial features to improve classification scalability and accuracy.
• Experiments were conducted on the Indian Pines (IP), University of Pavia (UP), and Salinas Scene (SA) datasets.

Main Results:

• GGBN achieved high classification accuracies: 99.97% on SA, 96.85% on IP, and 99.74% on UP.
• These results were obtained using only 5% of labeled training data.
• Performance surpassed existing state-of-the-art methods.

Conclusions:

• The proposed GGBN model effectively integrates genome graph structures with hybrid CNNs for hyperspectral image classification.
• GGBN demonstrates superior performance and data efficiency compared to current methods.