Thumbnail Tensor-A Method for Multidimensional Data Streams Clustering with an Efficient Tensor Subspace Model in the...

Area of Science:

• Data Science
• Signal Processing
• Machine Learning

Background:

• Multidimensional data streams present challenges for accurate signal change detection.
• Existing methods often lack efficiency and speed for real-time applications.

Purpose of the Study:

• To propose an efficient and accurate method for signal change detection in multidimensional data streams.
• To introduce a novel tensor-based approach for signal representation and analysis.

Main Methods:

• A novel tensor model is constructed using orthogonal tensor subspaces computed via higher-order singular value decomposition (HOSVD).
• Successive time windows of the data stream are compared against the model, with adaptive updating or rebuilding based on statistical inference.
• The method processes the signal tensor in scale-space, generating a thumbnail-like output.

Main Results:

• Experimental validation on annotated video databases and real underwater sequences demonstrated significant performance improvements.
• The thumbnail tensor method outperformed existing approaches in both accuracy and operational speed.
• The method effectively detects signal changes in complex multidimensional data.

Conclusions:

• The thumbnail tensor offers a computationally efficient and accurate solution for signal change detection in multidimensional data streams.
• This novel tensor-based approach enhances real-time data analysis capabilities.
• The method shows promise for applications in video analysis and underwater surveillance.