A Satellite Incipient Fault Detection Method Based on Decomposed Kullback–Leibler Divergence

Abstract

Detection of faults at the incipient stage is critical to improving the availability and continuity of satellite services. The application of a local optimum projection vector and the Kullback–Leibler (KL) divergence can improve the detection rate of incipient faults. However, this suffers from the problem of high time complexity. We propose decomposing the KL divergence in the original optimization model and applying the property of the generalized Rayleigh quotient to reduce time complexity. Additionally, we establish two distribution models for subfunctions F1(w) and F3(w) to detect the slight anomalous behavior of the mean and covariance. The effectiveness of the proposed method was verified through a numerical simulation case and a real satellite fault case. The results demonstrate the advantages of low computational complexity and high sensitivity to incipient faults.

Influence of Si intermixing on optical properties of Ge on Si and monolithic integration of Ge on Si nanocavity (Conference Presentation)

“…The second category [8][9][10] focuses on developing layers or structures that provide equivalence and invariance of representation for rotated object detection, which is crucial for effective angle regression and classification. The third category [7,[10][11][12][13][14][15][16][17] concentrates on proposing appropriate angle or oriented bounding box representations and loss functions to address the challenges arising from the circularity of angles.…”

“…There are other alternatives to replace the GIoU term for rotated DETRs. We compare the models with differentiable rotated IoU (RIoU), the GIoU version of RIoU (G-RIoU), GWD [58] and KLD [60] in Table 8b. We observed the KLD term, which is used in our model, shows the best performance in terms of AP 50 .…”

“…Replacement of GIoU Loss. The standard costs and losses used in DETRs are Focal Loss, L1 Loss, and GIoU Loss However, calculating overlaps between two rotated objects is well-known to be indifferentiable [58,60,61]. Therefore, we need to consider alternative options to replace GIoU loss for rotated objects.…”

“…Therefore, we need to consider alternative options to replace GIoU loss for rotated objects. Two representative substitutes for representing the error between predictions and ground-truth for rotated objects are Gaussian Wasserstein Distance (GWD) [58] and Kullback-Leibler Divergence (KLD) [60]. The center distance term in GWD and KLD also works as the penalty term for non-overlapping bounding boxes, similar to GIoU loss.…”

“…In aerial and satellite images, oriented bounding box representation is required to detect densely packed arbitrary-oriented objects without overlap [1,2]. A range of rotated object detectors are proposed by developing conventional convolution-based object detectors to handle oriented objects and achieve promising results [3][4][5][6][7][8][9][10][11][12][13][14][15][16][17]. In most cases, extending conventional detectors to oriented object detectors involves more complicated hand-designed components like rotated anchor priors or complex box representations.…”