Generalization of Machine Learning-Based Compression Method to Hyperspectral Images
- Resource Type
- Conference
- Authors
- Maksimov, Aleksey; Gashnikov, Mikhail
- Source
- 2022 VIII International Conference on Information Technology and Nanotechnology (ITNT) Information Technology and Nanotechnology (ITNT), 2022 VIII International Conference on. :1-4 May, 2022
- Subject
- Communication, Networking and Broadcast Technologies
Computing and Processing
Fields, Waves and Electromagnetics
Photonics and Electrooptics
Signal Processing and Analysis
Image coding
Data compression
Machine learning
Computer architecture
Computational efficiency
Task analysis
Information technology
hyperspectral data
machine learning
compression
generalization
deviation
- Language
The article deals with the problem of generalizing machine learning-based two-dimensional compression frameworks to hyperspectral (HS) data. The number of HS-floors in the HS-data is usually so large that we can treat the HS-data as three-dimensional arrays with a full third dimension. However, the architecture of machine learning-based compression frameworks usually relies heavily on the two-dimensionality of the source data. Therefore, the task of generalizing two-dimensional compression frameworks based on machine learning to HS-data is non-trivial and relevant. Most HS-floors in HS-data are usually very similar to each other. We use the strong similarity of HS-floors to generalize two-dimensional machine learning-based compression frameworks to HS-data. We perform computational experiments to investigate the effectiveness of the proposed approach for generalizing two-dimensional machine learning-based compression frameworks to HS-data. Experimental results prove that the use of the proposed approach can significantly increase the efficiency of HS-data compression.