Frames Learned by Prime Convolution Layers in a Deep Learning Framework.

This brief addresses understandability of modern machine learning networks with respect to the statistical properties of their convolution layers. It proposes a set of tools for categorizing a convolution layer in terms of kernel property (meanlet, differencelet, or distrotlet) or kernel sequence property (frame spectra and intralayer correlation matrix). These tools are expected to be relevant for determining the generalization capabilities of a convolutional neural network.

General adaptive-neighborhood technique for improving synthetic aperture radar interferometric coherence estimation.

A new method for filtering the coherence map issued from synthetic aperture radar (SAR) interferometric data is presented. For each pixel of the interferogram, an adaptive neighborhood is determined by a region-growing technique driven by the information provided by the amplitude images. Then pixels in the derived adaptive neighborhood are complex averaged to yield the filtered value of the coherence, after a phase-compensation step is performed.

Detection of linear features in synthetic-aperture radar images by use of the localized Radon transform and prior information.

A new linear-features detection method is proposed for extracting straight edges and lines in synthetic-aperture radar images. This method is based on the localized Radon transform, which produces geometrical integrals along straight lines. In the transformed domain, linear features have a specific signature: They appear as strongly contrasted structures, which are easier to extract with the conventional ratio edge detector.