In this study, we present an unsupervised methodology for detecting moving targets using Capella Space’s latest generation SAR sensor. The sensor has the capability to dwell on a target for an extended period of time in its spot-light (SP) mode, which we take advantage of to track moving objects in an acquisition. An approach that combines a signal-processing-based workflow with an image-domain-based one is presented. From one side, the long-dwell SAR image is exploit by doing an interfeometric processing of different azimuth sub-apertures from the long-dwell. From another side, a kernelized cross correlation technique is used. By combining intermediate results from these complementary workflows, a smooth and robust track is obtained on the targets. The algorithm is demonstrated on a long-dwell spotlight obtained over a busy shipping channel with watercraft both small and large successfully tracked.

The ability to automatically detect hard targets is highly beneficial for an imagery analyst. Their initial task in the information gathering process is to identify and interpret these targets in a scene. By automating the detection process, the workflow can be expedited. This paper presents an algorithm for detecting hard targets in long dwell spotlight (SP) Capella Space SAR images. The spectrum of the SP image is divided into non-overlaping sub-bands in order to compute the interferometric coherence between them. The proposed algorithm is tested on real spaceborne Capella SAR data showing the potential to clearly identify hard targets such as buildings, vehicles, and other artificial man-made objects.

Capella Space is the first US commercial company to build, launch, and operate a constellation of synthetic aperture radar satellites capable of collecting very high resolution SAR imagery. All satellites in the constellation carry an X-band radar capable of acquiring imagery in spotlight, sliding spotlight, and stripmap modes. In 2023, Capella will launch the first of a new generation of satellites names Acadia. These satellites will provide high quality imagery and lay the platform for advanced SAR data products, such as interferometric SAR and bistatic imagery.

Inductive transfer learning aims to learn from a small amount of training data for the target task by utilizing a pre-trained model from the source task. Most strategies that involve large-scale deep learning models adopt initialization with the pre-trained model and fine-tuning for the target task. However, when using over-parameterized models, we can often prune the model without sacrificing the accuracy of the source task. This motivates us to adopt model pruning for transfer learning with deep learning models. In this paper, we propose PAC-Net, a simple yet effective approach for transfer learning based on pruning. PAC-Net consists of three steps: Prune, Allocate, and Calibrate (PAC). The main idea behind these steps is to identify essential weights for the source task, fine-tune on the source task by updating the essential weights, and then calibrate on the target task by updating the remaining redundant weights. Under the various and extensive set of inductive transfer learning experiments, we show that our method achieves state-of-the-art performance by a large margin.