Taking advantage of Capella’s ability to dwell on a target for an extended period of time (nominally 30s) in its spotlight (SP) mode, an unsupervised methodology for detecting moving targets in this data is presented in this paper. By colourizing short segments (sub-apertures) of the total imaging time, a colourised sub-aperture image (CSI) can be formed. This can be used in conjunction with well-established computer vision techniques to detect moving targets and track them in the SP image. In essence, the moving target detection problem is transformed from temporal image stack identification to colour segmentation in a single image. The presented detection and tracking are wholly unsupervised. Additionally, computer-vision-based tracking algorithms are demonstrated on detected movers and qualitatively assessed for accuracy of tracking.

This paper introduces TACOSS a text-image alignment approach that allows explainable land cover semantic segmentation by directly integrating semantic concepts encoded from texts. TACOSS combines convolutional neural networks for visual feature extraction with semantic embeddings provided by a language model. By leveraging contrastive learning approaches, we learn an alignment between the visual and the (fixed) textual representations. In addition to producing standard semantic segmentation outputs, our model enables interactive queries with RS images using natural language prompts. The experimental results obtained on 50cm resolution aerial data from Switzerland show that TACOSS performs similarly to a standard semantic segmentation model while allowing the flexible usage of in- and out-of-vocabulary terms for the interactions with the image.

The rapid delineation of water extent in a flood-type event can be very beneficial to disaster relief efforts, and Synthetic Aperture Radar (SAR) is a modality ideally suited for such mapping. However, in a rapid-response scenario, it is desirable to produce such maps independent of historical or external data. To this end, we have propose a scheme to produce flood event maps from a single high-resolution StripMap (SM) imagery acquired from the Capella Space X-band VHR SAR constellation. The learning algorithm is solely trained on publicly available ancillary data, without the use of any human generated labels. The flood-maps are validated quantitatively on non-event scenes against water-occurrence data and qualitatively over the course of a flood-event caused by Hurricane Ida's landfall.

Floods are among the most common and destructive extreme weather events in the world. Spaceborne synthetic aperture radar (SAR) systems are well-suited for monitoring flood events given their ability to operate in near all-weather and all-time conditions. We present a case study classifying imagery from both the commercial very high resolution (VHR) Capella Space X-band constellation and the public C-band Sentinel-1 mission. We use inferred flood extents from these disparate sources to investigate the progression of unprecedented inundation over Abbotsford, British Columbia in November - December 2021.

Owing to the unique all-weather, day-night imaging capabilities of Synthetic Aperture Radar (SAR) imaging, the modality is advantageous in the detection of anthropogenic activity. In this paper we present a fully unsupervised change detection framework that operates on Very High Resolution (VHR) SAR image pairs to produce a binary change map, without a need for per-image parameter setting. The framework is demonstrated on a pair of Capella-2 VHR X-band imagery acquired over San Diego, USA.

Many applications utilizing SAR data, such as change detection, segmentation and classification, are impaired by the multiplicative speckle interference inherent in the imagery. Thus despeckling of SAR imagery is a often the key to developing robust algorithms for scene understanding. In recent years numerous deep learning-based approaches to speckle reduction have been proposed. However, the performance of these methods has largely failed to meet the expectations of researchers and industry alike. A key reason for this is due to the lack of accurate SAR-based ground truth training data. In this paper we propose the use of very high-resolution (VHR), low speckle aerial imagery and an accurate speckle model, as a ground truth signal for training a despeckling network based on the DnCNN architecture. Furthermore, we investigate modifications to the training formulation and finally demostrate the approach on Capella-2 VHR X-band imagery.