10
J. Eur. Opt. Society-Rapid Publ. 21, 31( 2025)
Fig
. 5. Detection results on the Texas Coast dataset.
Fig. 6. Detection results on the San Diego-I dataset.
0.8573, and the average AUC value after dimensionality reduction was 0.8537, a decrease of 0.42 %. Figure 11 shows the ROC curves of detection results for different data and detection methods before and after dimensionality reduction( the detection performance of 2S-GLRT is too poor regardless of dimensionality reduction, so it is not shown). Per-Pixel method is used as the noise estimation method. The ROC curve and score plot of the detection results both indicate that dimensionality reduction increases the detection score of the target while enhancing the generation of false alarms.
Another factor worth considering is the impact of noise estimation methods on the overall dimensionality reduction time. Table 2 shows the running time of the dimensionality reduction process under different noise estimation methods across various datasets, representing the average time after ten independent runs. The Median filtering method takes the shortest time due to its simplicity and efficient computational approach. In contrast, the Per-Pixel method has a running time that is an order of magnitude higher than the other three methods, resulting in a significant increase in computational cost. The size of the dataset also significantly impacts the running time of dimensionality reduction methods, with larger datasets generally requiring longer processing times. Figure 12 demonstrates that the running time scales linearly with the data volume.