“…We used a database of 1,134 images from a high-frequency SAS system to generate the plots of Figure 2 and 3 and to evaluate our results. The images were dynamic range compressed using the method in [42]. We used splits of 405/135/594 images for training/validation/testing.…”
Synthetic aperture sonar (SAS) is an imaging modality which produces high and constant resolution images of the seafloor. These sonars are often mounted to a unmanned underwater vehicle (UUV) to autonomously collect imagery of a prescribed survey area. While a survey is underway, UUV communications back to the operator are often limited due to the use of a low-bandwidth acoustic communications (ACOMMS) channel. Because of this, high-quality SAS imagery is rarely sent over this link due to the lack of an efficient compression scheme to send such information. Creating an efficient SAS image compression scheme provides at least two operational benefits:(1) image chips beamformed and tagged by onboard processing algorithms can be quickly communicated to operators while a survey is ongoing, and (2) cooperative UUVs can exchange salient image chips among themselves to reconcile position ambiguity and obtain a shared reference frame. In this work we propose a learned image compression scheme for SAS imagery using deep neural networks (DNNs). DNNs have already been applied to the image compression problem but almost exclusively for optical imagery. We highlight some important differences between SAS imagery and optical imagery which prevents the simple application of off-the-shelf (OTS) methods like JPEG and WebP to SAS imagery. We propose an image compression scheme which specifically addresses the domain-specific properties of SAS imagery to obtain useful image compression performance on a real-world SAS dataset. We show that we can reduce the bitrate by up to thirty-five percent while still maintaining the same perceptual image quality as OTS codecs.Index Terms-Synthetic aperture sonar (SAS), image Compression, deep learning, unmanned underwater vehicles, acoustic communications (ACOMMS)
“…We used a database of 1,134 images from a high-frequency SAS system to generate the plots of Figure 2 and 3 and to evaluate our results. The images were dynamic range compressed using the method in [42]. We used splits of 405/135/594 images for training/validation/testing.…”
Synthetic aperture sonar (SAS) is an imaging modality which produces high and constant resolution images of the seafloor. These sonars are often mounted to a unmanned underwater vehicle (UUV) to autonomously collect imagery of a prescribed survey area. While a survey is underway, UUV communications back to the operator are often limited due to the use of a low-bandwidth acoustic communications (ACOMMS) channel. Because of this, high-quality SAS imagery is rarely sent over this link due to the lack of an efficient compression scheme to send such information. Creating an efficient SAS image compression scheme provides at least two operational benefits:(1) image chips beamformed and tagged by onboard processing algorithms can be quickly communicated to operators while a survey is ongoing, and (2) cooperative UUVs can exchange salient image chips among themselves to reconcile position ambiguity and obtain a shared reference frame. In this work we propose a learned image compression scheme for SAS imagery using deep neural networks (DNNs). DNNs have already been applied to the image compression problem but almost exclusively for optical imagery. We highlight some important differences between SAS imagery and optical imagery which prevents the simple application of off-the-shelf (OTS) methods like JPEG and WebP to SAS imagery. We propose an image compression scheme which specifically addresses the domain-specific properties of SAS imagery to obtain useful image compression performance on a real-world SAS dataset. We show that we can reduce the bitrate by up to thirty-five percent while still maintaining the same perceptual image quality as OTS codecs.Index Terms-Synthetic aperture sonar (SAS), image Compression, deep learning, unmanned underwater vehicles, acoustic communications (ACOMMS)
“…In a typical remote sensing survey, the raw acoustic data is processed to generate imagery data products used for post-mission analysis (PMA) or other scientific analyses. ASASIN (Advanced Synthetic Aperture Sonar Imaging eNgine) 6 is used to generate SAS imagery from measured and simulated data. ASASIN is a time-domain back-projection image reconstruction software that utilizes a GPU for highly parallelized signal processing.…”
Section: Signal Processing and Multiple Representationsmentioning
“…For IDUS, we dynamic range compress the magnitude image by using Schlick's rational mapping operator [59]. We set the target brightness [60] to 0.5. Additionally, we further use the OpenCV function "equalizeHist" [61] to equalize the gray-scale histogram of the tone-mapped images and normalize each to zero mean and unit standard deviation.…”
Section: A Dataset Description and Pre-processingmentioning
Synthetic aperture sonar (SAS) systems produce high-resolution images of the seabed environment. Moreover, deep learning has demonstrated superior ability in finding robust features for automating imagery analysis. However, the success of deep learning is conditioned on having lots of labeled training data, but obtaining generous pixel-level annotations of SAS imagery is often practically infeasible. This challenge has thus far limited the adoption of deep learning methods for SAS segmentation. Algorithms exist to segment SAS imagery in an unsupervised manner, but they lack the benefit of state-of-theart learning methods and the results present significant room for improvement. In view of the above, we propose a new iterative algorithm for unsupervised SAS image segmentation combining superpixel formation, deep learning, and traditional clustering methods. We call our method Iterative Deep Unsupervised Segmentation (IDUS). IDUS is an unsupervised learning framework that can be divided into four main steps: 1) A deep network estimates class assignments. 2) Low-level image features from the deep network are clustered into superpixels. 3) Superpixels are clustered into class assignments (which we call pseudo-labels) using k-means. 4) Resulting pseudo-labels are used for loss backpropagation of the deep network prediction. These four steps are performed iteratively until convergence. A comparison of IDUS to current state-of-the-art methods on a realistic benchmark dataset for SAS image segmentation demonstrates the benefits of our proposal even as the IDUS incurs a much lower computational burden during inference (actual labeling of a test image). Because our design combines merits of classical superpixel methods with deep learning, practically we demonstrate a very significant benefit in terms of reduced selection bias, i.e. IDUS shows markedly improved robustness against the choice of training images. Finally, we also develop a semi-supervised (SS) extension of IDUS called IDSS and demonstrate experimentally that it can further enhance performance while outperforming supervised alternatives that exploit the same labeled training imagery.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.