The Student Poster Competition at OCEANS 2019 Marseille

The Student Poster Competition (SPC) is a flagship event of the MTS/OES OCEANS conferences in which students (both undergraduate and graduate) from colleges and universities around the world participate. Col. Norman Miller had  created and implemented the first SPC in 1989 during the OCEANS conference in Seattle, Washington, and it has been a feature of OCEANS conferences ever since. During each conference, typically about twenty students are selected from the pool of aspiring applicants, based on their submitted abstracts, to present a poster describing their work. The selected students’ travel, accommodation and conference registration are fully supported by the conference. In addition, monetary prizes are also awarded to the top three best presenters among the selected participants.

The first of this year’s SPC was held during OCEANS 2019 Marseille at the Marseille Chanot Exhibition and Convention Centre from the 17th to the 20th of June. The SPC program was funded by a grant from the Office of Naval Research Global (ONR-G), and we offer our sincere gratitude for their continued support. Preparations in the run-up to the event were ably managed by the SPC chair on the local organizing committee, Gaultier Real, with minimal assistance from me and Liesl Hotaling (MTS SPC Chair, who unfortunately could not attend). Out of the 59 applicants (from 16 different countries) that chose to submit their abstracts for consideration for the SPC, 18 students were chosen to present their work at the event.

Following a mandatory short oral presentation to the judges on Monday morning, the students put up their posters at the designated slots in the Exhibition Hall and were eager to pitch their research to conference attendees. They enthusiastically described their work to the judges and the other delegates that visited their posters. We had a total of 11 volunteers that offered to serve as judges, some of whom were contestants themselves in prior editions of the SPC. Given the high standards in the work presented by the participants and the breadth of topics covered, judging seemed to be a rather difficult task, spanning over two days and numerous hours of very intensive assessment. Our gratitude and appreciation goes out once again to the judges.

The award ceremony took place during the conference gala banquet at the Casa Delauze, Marseille on Wednesday June 19th. Each participating student was awarded a Certificate of Attendance, and the judges’ tireless efforts were acknowledged by presenting each of them a Certificate of Appreciation. The three top-ranked posters were announced and prizes were awarded. The awards were handed out to the winners by Dr. Vincent Rigaud, Director of Underwater Systems Unit, Ifremer, the organization which generously sponsored the monetary awards.

The full list of participants (including the prize winners) together with their affiliation, poster title and an abstract of their poster is given below.

 

First Prize and Norman Miller Award (Certificate and $3000 prize)
Léa Bouffaut, Naval Academy Research Institute, France

Abstract—The size of sound archives collected globally by the community to monitor cetaceans, including blue whales, is rapidly increasing. Analyzing these vast amounts of data efficiently requires reliable automated detection algorithms. Typically these algorithms focus on a specific call type produced by a single species. We developed an automatic transcription algorithm which can identify multiple concurrently calling species in sound recordings. The algorithm was tested on data containing series of calls (songs) of Madagascar pygmy blue whales and series of the 27Hz tonal unit named ”P-call”. The algorithm is based on pattern recognition of tonal calls in the time-frequency domain where (1) segmentation is realized by detection of tonal signals, (2) features are extracted from their time-frequency-amplitude information, and (3) classification is realized by clustering. The classified tonal signals are then used to reconstruct, separately, the underlying songs. We successfully trained and tested the algorithm on data (>4000 annotated calls) in the Western Indian Ocean and achieved an overall precision of 97.2% and a recall of 96.9%, respectively.

 

Second Prize (Certificate and $2000 prize)
Mengxue Hou, Georgia Institute of Technology, USA

Abstract—In this paper, we present a flow field partition method that extracts the key features, which are the spatial and temporal variation of the flow field. The partition method is developed based on K-means algorithm. In the case where the temporal pattern of the flow field contains only a periodic tidal  component, we propose an algorithm that partitions the flow field into static clusters of piece-wise constant flow, by performing K- means clustering over the time-averaged flow field. Then the method is extended to partitioning the flow field into clusters of uniform time-varying flow, by fitting the spatially averaged flow in each static partitioned region to a parametric flow model. Simulation results of partitioning both a simulated jet flow field, as well as the ocean surface flow data into time-invariant and time-varying uniform flow are presented to demonstrate that the proposed method can represent the true flow field with significantly less amount of data. Result of using Method of Evolving Junctions to plan the time-optimal path in the partitioned flow field is also presented to demonstrate that the proposed flow partitioning method can be applied to facilitate path planning, and can reduce the path planning computational cost.

 

Third Prize (Certificate and $1000 prize)
Wendy Snyder, University of Rhode Island, USA

Abstract—Seaglider is a buoyancy driven autonomous under- water vehicle (AUV) primarily used as a platform for oceanographic measurements of salinity, temperature, and oxygen, where precise localization is not crucial. A recent experiment tracked a Seaglider, instrumented with an attitude and heading reference system (AHRS), on an acoustic tracking range in an effort to quantify the accuracy of the Seaglider’s two localization solutions and determine the effects of incorporating additional inertial measurements into the solution. Preliminary analysis of results has shown the Seaglider’s glideslope model (GSM) is more accurate and reliable than the hydrodynamic model (HDM) during typical flight dynamics. Errors in the GSM solution did not exhibit a clear drift behavior but remained on average within 22 m of the ground truth acoustic track over dives with a maximum depth of 90 to 125 m. Errors in the HDM solution exhibited linear growth until the apogee point of the dive when errors began to linearly reduce. On average the error increased at a rate of about 5 m/min between the surface and maximum dive depth, with maximum errors in excess of 100 m during dives with a maximum depth of 90 m. The substitution of attitude estimates from the AHRS into the localization models provided similar but less consistent results with slightly higher errors.

 

Alfredo Bayuelo Sierra, National University of Colombia, Colombia

Abstract—Simultaneous Localization and Mapping (SLAM) is a desired capability for an autonomous vehicle. SLAM is the process used to construct a representation map of an unknown surrounding world, localize in it, and use this map to navigate through it. For aquatic environments, this is an even harder challenge since it is a non-structured and highly dynamical environment in which classical landmark-based approaches to  the SLAM problem are not suitable. In this paper, we use previously developed representations of aquatic environments based on Dynamical Systems and Augmented Bathymetry to construct a preliminary solution to the SLAM problem. A simple motion model is combined with physicochemical readings of the water to find candidate locations. Our ideas were preliminarily tested in a dataset acquired in the Lake Nighthorse, CO, obtaining promising results in the new strategy.

Filippo Campagnaro, University of Padova, Italy

Abstract—Grounding of ships is as of today one of the most common causes of maritime accidents. An effective anti- grounding service, where updated bathymetry data is sent in real time to a ship can improve maritime and human safety. The bathymetry data is collected by an autonomous surface vessel equipped with a multi-beam sonar while the ship is approaching the port. This data is then conveyed immediately to the ship, by employing a wireless communication link. In this paper, a study on the enabling technologies for an effective anti-grounding service is provided, with special focus on the wireless communication devices to employ for the real-time bathymetry data transmission.

Matias Carandell, Universitat Politecnica de Catalunya, Spain

Abstract—A Kinetic Energy Harvester (KEH) device under design for drifter applications has been tested at real sea conditions in a controlled area. The KEH consists on a pendulum system capable to transform the oscillations of the waves into rotation which will be converted into electrical energy thanks to a dc micro generator. The KEH has been placed inside a drifter. First, the motion of the drifter was obtained through an embedded Inertial Measurement Unit, showing natural pendulum frequency of 1.5 Hz. Simulations with OrcaFlex validate the experimental results. Then, the rotation speed of the micro generator was measured. Results show a prevalence of speed of 1400 rpm, which should lead to a potential power output of 2 mW.

Pedro Geraldes, ISEP, Portugal

Abstract—Zooplankton plays a key-role on Earth’s ecosystem, emerging in the oceans and rivers in great quantities and diversity, making it an important and rather common topic on scientific studies. Given the numbers of different species it is not only necessary to study their numbers but also their classification. In this paper a possible solution for the zooplankton in situ detection and classification problem in real- time is proposed using a portable deep learning approach based on Convolutional Neural Networks deployed on INESC TEC’s MarinEye system. For detection a Single Shot Detection model with MobileNet was used, and ZooplanktoNet for classification. System portability is guaranteed with the use of MovidiusTM Neural Compute Stick as the deep learning motor.

Hossein Ghannadrezaii, Dalhousie University, Canada

Abstract—In this paper, a network relying on software-defined underwater acoustic nodes is proposed: by receiving a periodic beacon signal from the sink node, each node in a cluster obtains a prediction about the communication quality of the potential links to its one-hop neighbors as well as to the available relay nodes within the cluster. A hidden Markov process is used to predict the next state of the channels, using the probability distribution of  the channel amplitude and delay spread from previous observations. Each transmitting node evaluates the quality of links to its next hop neighbor relay nodes. Accordingly, a normalized weight representing the channel quality is assigned to each link. The channel gain and its delay spread are the two metrics that are used to define the channel quality. To evaluate the network performance in realistic conditions, the output of a statistical model combined with the Bellhop ray tracing software is compared to experimental data. Using the channel quality, each transmitter node in the network selects its next hop optimum relay node. This scheme minimizes the number of transmitted control packets and also reduces the re-transmission of data packets, by predicting the channel status rather than exchanging an excessive number of control packets which typically convey expired channel state information. Minimizing the overhead and selecting the optimum channel saves on the energy per bit consumption while maintaining high packet delivery ratio and low latency.

Yen-Hsiang Huang, National Taiwan University, Taiwan

Abstract—Due to the potential of the offshore wind energy in the Taiwan Strait, the development of the offshore wind farm is cost- effective. However, the expected develop area for potential offshore wind farms has overlapped with the habitat of the Sousa chinensis (also called Indo-Pacific Humpback Dolphin or Chinese White Dolphin). Therefore, to know if there are dolphins in the construction area is the most important. This paper provides a process for checking if there are dolphins in the specific area and using an unmanned surface vehicle to track the source and find the area they are. This process has four main topics, including real-time dolphin whistle detection, underwater acoustic source direction calculate, source tracking behavior (STB) and real-time shoreside monitoring system of the surface vehicle. The whole system was developed using the open source middleware MOOS-IvP and testing in the embedded system (NVIDIA Jetson TX1). The whole system was implemented and successfully tested in the field experiment using artificial dolphin sound.

Keunhwan Kim, KAIST, Republic of Korea

Abstract—Marine radar is essential for ship navigation and control. Several studies exist that are focused on eliminating unwanted noise signals to properly detect targets in radar images. However, in general, radar image processing techniques are sensitive to radar sensor specifications and operating conditions, and ad-hoc image preprocessing steps are required to adjust many parameters to achieve an overall satisfactory performance. This paper addresses the semantic segmentation of marine radar images using a convolutional neural network (CNN). SegNet is utilized as the CNN algorithm for the semantic segmentation. In order to improve the performance of the CNN considering the measurement characteristics of the radar, Cartesian coordinate system based images are transformed into polar coordinate systems. The Cartesian and polar images are used and the original SegNet model is modified to train the network in a variety of methods. The results show the feasibility of using a deep learning network to achieve effective and robust semantic segmentation in several different sets of marine radar images.

Ivan Masmitja, Universitat Politècnica de Catalunya, Spain

Abstract—The use of autonomous underwater vehicles for ocean research has increased as they have a better cost-to-performance ratio than crewed oceanographic vessels. For example, autonomous surface vehicles (e.g. a Wave Glider) can be used to localise and track targets even in the deep-sea. While other researchers have focused on target tracking using acoustic modems, none explored up to date the area-only target tracking methodology. Here we present such a novel method with commercially available acoustic tags, thereby reducing the costs and complexity over other tracking systems. Moreover, this method can be used to track small targets (e.g., jellyfish) due to the tag’s small size. The methodology behind the area-only technique is shown, and results from field tests conducted in Monterey Bay are also presented.

Tiago Matos, University of Minho, Portugal

Abstract—A cost-effective (less than 20€) and low-power device is present for in situ continuous monitoring of suspended sediments (SPM) concentration in estuarine and coastal areas. The sensor uses nephelometric technique for SPM values less than 20g/L and backscatter technique for higher ones. A transmitted infrared (IR) and ultraviolet (UV) channels are used to perform the distinguish of inorganic from organic matter in the suspended particles. It is explained the design and built of the sensor as its calibration and preparation for in field tests. The sensor was deployed for one week in a small dock in the estuarine zone of Cavado river (Esposende, Portugal) where is analyzed the SPM and organic/inorganic matter change with the tidal cycles.

Brian Page, Purdue University, USA

Abstract—This paper presents an early stage work on docking and recovery of Autonomous Surface Vehicles (AUVs) with Unmanned Surface Vehicles (USVs). The docking system is collapsible and adaptable to a wide range of AUVs. Control of the docking maneuver is completed based on a one camera, one light approach using a frontseat-backseat control setup. In this scenario, the vehicle navigates under frontseat control to within visual range of the dock. The backseat controller then takes over and performs terminal homing. Pool and open water testing was completed with docking to a small dinghy in both a manned and unmanned configuration. Two experimental docking attempts are presented, one docking to a slow moving target and another docking to a quickly moving target. The empirical results show that docking and recovery of an AUV from a USV while underway is feasible.

Jonatan Scharff Willners, Heriot-Watt University, United Kingdom

Abstract—This work looks at when to transmit localisation messages in the scenario where an Autonomous Surface Vehicle (ASV) has a planned path and can estimate a submerged AUV’s positions based on acoustic updates. Acoustic localisation messages are in general transmitted periodically, but by estimating the movement of the involved vehicles, messages can be transmitted at a time when the geometrical relationship between vehicles should reduce the receiving vehicles error and uncertainty to the greatest extent. This paper looks at how selection of Time of Launch (ToL) within a Time-Division Multiple Access slot can reduce this based on history of transmission and the estimated geometrical relationship between vehicles over time. The method to select ToL is dependent on the localisation method, we look at this from the perspective of Extended Kalman Filter and to solve the trilateration problem using Non-linear Least Squares. The benefit of proposed approach is in scenarios where the operating vehicles have their own objectives, and cannot adapt their path to achieve a more beneficial transmission position. The proposed approach shows a reduction of error and uncertainty, while using a navigational dataset collected by an Autonomous Underwater Vehicle when compared to other methods of which transmission times are selected.

Maëlle Torterotot, Institut Universitaire Européen de la Mer, France

Abstract—The most common approach to monitor mysticete acoustic presence is to detect and count their calls in audio records. To implement this method on large datasets, polyvalent and robust automated call detectors are required. Evaluating their performance is essential, to design a detection strategy adapted to study the available datasets. This assessment then enables accurate post-analyses and comparisons of multiple independent surveys. In this paper, we present the performance of a detector based on dictionaries and sparse representation of the signal to detect blue whale stereotyped and non-stereotyped vocalizations (D-calls) in a large acoustic database with multiple sites and years of recordings in the southern Indian Ocean. Results show that recall increases with the SNR (Sound to Noise Ratio) and reaches 90% for positive SNR stereotyped calls and between 80% and 90% for high SNR D-calls. A detailed analysis of the influence of dictionary composition, SNR of the calls, manual ground truth as well as interference types and abundance, on the performance variability is presented. Eventually, a detection strategy for long term acoustic monitoring is defined.

Franz Weidmann, Fulda University of Applied Sciences, Germany

Abstract—Underwater imaging enables marine researchers to collect large datasets of seagrass images. These images can be used to monitor the health state of underwater meadows by estimating the area that is covered by seagrass and how this area changes over time. Since the manual analysis of such images is slow and error-prone, we follow the path of deep learning for automatic image analysis. Our contribution is the investigation of deep semantic segmentation for the specific task of seagrass coverage estimation. We evaluated multiple Deep Neural Network Architectures including the DeepLabv3Plus Network which performs best, with a mean intersection over union of 87.78%. The qualitative results in our experiments indicate that the Deep Learning approach is not only more accurate than a human but also multiple times faster in annotating underwater meadows. Our code is available on GitHub: https://enviewfulda.github.io/LookingForSeagrassSemanticSegmentation/

 

Qingyun Yan, Memorial University of Newfoundland, Canada

Abstract—In this paper, an effective model is developed for retrieving sea ice thickness (SIT) from scattering coefficient (σ0) produced with TechDemoSat-1 (TDS-1) data. Here, σ0 is formulated as the product of the propagation loss due to SIT and the reflection coefficient of underlying seawater. In application, σ0 at specular point was firstly generated based on radar equation using TDS-1 data. Next, SIT was calculated from TDS-1 σ0 using the proposed model, and verified with reference SIT data obtained by the Soil Moisture Ocean Salinity (SMOS) satellite. The data used here were from measurements over the year of 2015. Comparison results showed a good consistency between the derived and reference SIT, with a correlation coefficient of 0.90 and a root mean square difference of 8.68 cm, which demonstrates the potential of developed model and the utility of TDS-1 data for SIT retrieval.