Binary partition trees-based robust adaptive hyperspectral RX anomaly detection

Abstract

The Reed-Xiaoli (RX) is considered as the benchmark algorithm in multidimensional anomaly detection (AD). However, the RX detector performance decreases when the statistical parameters estimation is poor. This could happen when the background is non-homogeneous or the noise independence assumption is not fulfilled. For a better performance, the statistical parameters are estimated locally using a sliding window approach. In this approach, called adaptive RX, a window is centered over the pixel under the test (PUT), so the background mean and covariance statistics are estimated using the data samples lying inside the window's spatial support, named the secondary data. Sometimes, a smaller guard window prevents those pixels close to the PUT to be used, in order to avoid the presence of outliers in the statistical estimation. The size of the window is chosen large enough to ensure the invertibility of the covariance matrix and small enough to justify both spatial and spectral homogeneity. We present here an alternative methodology to select the secondary data for a PUT by means of a binary partition tree (BPT) representation of the image. We test the proposed BPT-based adaptive hyperspectral RX AD algorithm using a real dataset provided by the Target Detection Blind Test project.

Self-Supervised Learning Based Anomaly Detection in Synthetic Aperture Radar Imaging

“…9 presents a comparison of anomaly detection results obtained with different methods. All the results are here clipped at the highest 1% values, see (9).…”

“…This research field is essential in data mining for quickly isolating irregular or suspicious segments in large amounts of the database. Many anomaly detection schemes have been proposed in the literature [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17]. Among them, the unsupervised methods are the most interesting since they are widely applicable and do not require labeling the data.…”

Global and local anomaly detectors for tumor segmentation in dynamic pet acquisitions

“…These assumptions are often inaccurate for real images [22,23], as they might be in the case of PET medical images. In fact, dealing with images of the human body, the trouble of heterogeneous background arises when passing from a tissue type to another one; in this case the performance of RXD may impair because it strongly depends on the correct estimation of the statistical parameters (namely, mean and covariance).…”

“…Troubles may arise in particular when the parameters are estimated globally, as the assumption for all the different tissues in the body to have homogeneous statistics might not be accurate. An improvement to the parameters estimation may be achieved by limiting the sampling locally to a subset of voxels using a sliding window, chosen small enough to make the uniform background assumption verified [22]. For all the voxels in the image, the local approach centers two concentric windows on the voxel under test (VUT): an inner and smaller one, named guard window, and an external one, named outer window.…”

“…For all the voxels in the image, the local approach centers two concentric windows on the voxel under test (VUT): an inner and smaller one, named guard window, and an external one, named outer window. The size of the guard window should approximately be the same as that of the expected anomaly; the size of the outer window has to be large enough to make the covariance matrix always invertible, but small enough to justify both spatial and spectral homogeneity [22]. These windows have the shape of boxes described by three dimensions (namely height, width and depth); when all three dimensions are equal the shape reduces to a cube.…”