Limits...
A Weighted Spatial-Spectral Kernel RX Algorithm and Efficient Implementation on GPUs

View Article: PubMed Central - PubMed

ABSTRACT

The kernel RX (KRX) detector proposed by Kwon and Nasrabadi exploits a kernel function to obtain a better detection performance. However, it still has two limits that can be improved. On the one hand, reasonable integration of spatial-spectral information can be used to further improve its detection accuracy. On the other hand, parallel computing can be used to reduce the processing time in available KRX detectors. Accordingly, this paper presents a novel weighted spatial-spectral kernel RX (WSSKRX) detector and its parallel implementation on graphics processing units (GPUs). The WSSKRX utilizes the spatial neighborhood resources to reconstruct the testing pixels by introducing a spectral factor and a spatial window, thereby effectively reducing the interference of background noise. Then, the kernel function is redesigned as a mapping trick in a KRX detector to implement the anomaly detection. In addition, a powerful architecture based on the GPU technique is designed to accelerate WSSKRX. To substantiate the performance of the proposed algorithm, both synthetic and real data are conducted for experiments.

No MeSH data available.


ROC curves of LRX, LKRX and WSSKRX on the two real datasets. (a) Synthetic dataset; (b) San Diego Airport dataset; (c) SpecTIR dataset.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC5375727&req=5

sensors-17-00441-f015: ROC curves of LRX, LKRX and WSSKRX on the two real datasets. (a) Synthetic dataset; (b) San Diego Airport dataset; (c) SpecTIR dataset.

Mentions: In order to give a more objective evaluation, Figure 15 presents the ROC curves of GRX, LRX, LKRX, HKRX and WSSKRX on three datasets. Since the feature distribution of the synthetic dataset is more complex and the GRX and LRX are operated in low-dimensional space, they perform worse as shown in Figure 15a. The detection accuracy of HKRX is better than KRX because the global and local information is taken into account together. The WSSKRX obtains better detection performance than other detectors, because it more rationally uses the data information. In Figure 15b, GRX and LRX perform worse than other algorithms. The ROC curves of LKRX, HKRX and WSSKRX are the similar when the false alarm rate is lower than 0.06. After that, WSSKRX begins to obtain a higher detection probability than LKRX and HKRX, and it achieves 1 probability of detection with a lower false alarm rate than LKRX and HKRX. In Figure 15c, Although WSSKRX needs a slightly higher false alarm rate than LKRX and HKRX when achieving 1 probability of detection, the overall detection performance of WSSKRX is still clearly better than other algorithms. It is noted that the feature distribution of the SpecTIR dataset is more separable than other two datasets, so GRX obtains better detection performance.


A Weighted Spatial-Spectral Kernel RX Algorithm and Efficient Implementation on GPUs
ROC curves of LRX, LKRX and WSSKRX on the two real datasets. (a) Synthetic dataset; (b) San Diego Airport dataset; (c) SpecTIR dataset.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC5375727&req=5

sensors-17-00441-f015: ROC curves of LRX, LKRX and WSSKRX on the two real datasets. (a) Synthetic dataset; (b) San Diego Airport dataset; (c) SpecTIR dataset.
Mentions: In order to give a more objective evaluation, Figure 15 presents the ROC curves of GRX, LRX, LKRX, HKRX and WSSKRX on three datasets. Since the feature distribution of the synthetic dataset is more complex and the GRX and LRX are operated in low-dimensional space, they perform worse as shown in Figure 15a. The detection accuracy of HKRX is better than KRX because the global and local information is taken into account together. The WSSKRX obtains better detection performance than other detectors, because it more rationally uses the data information. In Figure 15b, GRX and LRX perform worse than other algorithms. The ROC curves of LKRX, HKRX and WSSKRX are the similar when the false alarm rate is lower than 0.06. After that, WSSKRX begins to obtain a higher detection probability than LKRX and HKRX, and it achieves 1 probability of detection with a lower false alarm rate than LKRX and HKRX. In Figure 15c, Although WSSKRX needs a slightly higher false alarm rate than LKRX and HKRX when achieving 1 probability of detection, the overall detection performance of WSSKRX is still clearly better than other algorithms. It is noted that the feature distribution of the SpecTIR dataset is more separable than other two datasets, so GRX obtains better detection performance.

View Article: PubMed Central - PubMed

ABSTRACT

The kernel RX (KRX) detector proposed by Kwon and Nasrabadi exploits a kernel function to obtain a better detection performance. However, it still has two limits that can be improved. On the one hand, reasonable integration of spatial-spectral information can be used to further improve its detection accuracy. On the other hand, parallel computing can be used to reduce the processing time in available KRX detectors. Accordingly, this paper presents a novel weighted spatial-spectral kernel RX (WSSKRX) detector and its parallel implementation on graphics processing units (GPUs). The WSSKRX utilizes the spatial neighborhood resources to reconstruct the testing pixels by introducing a spectral factor and a spatial window, thereby effectively reducing the interference of background noise. Then, the kernel function is redesigned as a mapping trick in a KRX detector to implement the anomaly detection. In addition, a powerful architecture based on the GPU technique is designed to accelerate WSSKRX. To substantiate the performance of the proposed algorithm, both synthetic and real data are conducted for experiments.

No MeSH data available.