研究目的
To develop real-time global and local anomaly detectors for hyperspectral imagery using the Woodbury matrix identity for efficient pixel-by-pixel processing.
研究成果
The proposed real-time global and local anomaly detectors significantly reduce processing time using the Woodbury matrix identity, enabling efficient pixel-by-pixel processing. They demonstrate effective anomaly detection in both synthetic and real hyperspectral images, with computational improvements of up to 100-fold. The methods are causal and suitable for hardware implementation, offering benefits in data storage and real-time decision-making.
研究不足
The real-time detectors may be affected by background suppression issues where weak anomalies are overwhelmed by strong ones. The causal array window for local detection is linear and not a traditional square matrix, which might limit spatial context. Computational savings depend on the recursive update efficiency, and boundary effects in local windows can increase processing time.
1：Experimental Design and Method Selection:
The study designs real-time versions of global and local RX detectors by leveraging the Woodbury matrix identity to enable recursive updates of correlation matrices, avoiding repeated matrix inversions. Methods include causal processing and sliding window techniques.
2：Sample Selection and Data Sources:
Synthetic hyperspectral images are created using a real Cuprite AVIRIS scene with known mineral signatures, and a real AVIRIS Lunar Crater Volcanic Field (LCVF) image is used. Data sources include the USGS website for the Cuprite data.
3：List of Experimental Equipment and Materials:
A computer with a 64-bit operating system, Intel(R) Core(TM) i7-4770K CPU at 3.5 GHz, and 16 GB RAM is used for simulations. Hyperspectral imagery data from AVIRIS sensors are processed.
4：5 GHz, and 16 GB RAM is used for simulations. Hyperspectral imagery data from AVIRIS sensors are processed.
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: For global detectors, the causal RX detector is implemented with recursive updates. For local detectors, causal matrix and causal array windows are defined and processed recursively. Experiments involve applying detectors to synthetic and real images, comparing processing times and detection results.
5：Data Analysis Methods:
Detection results are analyzed visually through grayscale and 3D plots. Computational times are measured and averaged over multiple runs. Performance is evaluated based on anomaly detection accuracy and efficiency.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
