The National Marine Fisheries Service (NMFS; informally known as “NOAA Fisheries”) of the National Oceanic and Atmospheric Administration (NOAA) is responsible for the stewardship of the nation’s living marine resources and their habitat. As part of this charge, NOAA Fisheries conducts stock assessments of the abundance and composition of fish stocks in several bodies of water. The use of images and videos, when accompanied by appropriate statistical analyses of the inferred data, is of increasing importance for estimating the abundance of species and their age distributions. NOAA Fisheries is actively seeking to improve the quality and reliability of data from still and stereo-video imagery, and, more generally, to automate more of the stock assessment process. In particular, NOAA Fisheries is interested in identifying promising directions for advancing its analytical capabilities, including opportunities to leverage capabilities from other fields (such as the use of machine learning in pattern recognition).
The accuracy and efficiency of fisheries stock assessments are limited in large part by data collection tools and techniques. At present, stock assessments rely heavily on human data-gathering and analysis. Automatic means of fish stock assessments are appealing because they offer the potential to improve efficiency and reduce human workload and perhaps develop higher-fidelity measurements. However, automatic counting or characterization remains a complex and difficult task because of numerous factors: many species move about during observations, individuals often look very similar, some species blend in with their background, lighting can be variable, and the correlation between measurable features and desired features (such as age or gender) may be weak. These complexities are com-
pounded by data collection techniques that may involve trawling or the collection of images via a camera on a moving platform.
A workshop was developed to enable experts from diverse communities to share perspectives about the most efficient path toward improved automation of visual information for fisheries stock assessments and to discuss both near-term (3 to 5 years) and long-term goals that can be achieved through modest research and development efforts. On May 16-17, 2014, the National Research Council’s (NRC’s) Committee on Applied and Theoretical Statistics convened a workshop to discuss analysis techniques for images and videos for fisheries stock assessment. To conduct the workshop, a planning committee was established to refine the workshop topics, identify speakers, and plan the workshop agenda. The workshop was held at the National Academy of Sciences building in Washington, D.C., and was sponsored by NOAA. Approximately 40 participants, including speakers, members of the parent committee, invited guests, and members of the public, participated in the 2-day workshop. The workshop was also webcast live, with approximately 10 people participating remotely via webcast. A complete statement of task is shown in Box 1.1.
This report has been prepared by the workshop rapporteur as a factual summary of what occurred at the workshop. The planning committee’s role was limited to planning and convening the workshop. The views contained in the report are those of individual workshop participants and do not necessarily represent the views of all workshop participants, the planning committee, or the NRC.
In addition to the workshop summary provided here, materials related to the workshop can be found online at the website of the Board on Mathematical Sciences and their Applications (http://www.nas.edu/bmsa), including the agenda, speaker presentations, archived webcasts of the presentations and discussions, and other background materials.
Subsequent chapters of this report summarize the workshop presentations and discussion, following the organization of the workshop. Chapter 2 sets the stage for current practice and future needs in fisheries stock assessment. Chapter 3 focuses on multi-modal sensing. Chapter 4 discusses methods of image processing and detection. Chapter 5 focuses on multi-object tracking. Chapter 6 discusses shape and motion analysis. Chapter 7 describes methods of identification and classification, and Chapter 8 summarizes lessons learned and strategies moving forward
Statement of Task
An ad hoc committee will plan and conduct a public workshop that will examine the frontiers in methodology for examining image, video, and possibly other sensor data related to the following tasks of importance to the National Marine Fisheries Service (NMFS):
- Automatic counting or characterization of fish as they pass through a trawl against a semi-static background.
- Interpreting video (e.g., identifying the species, counting individuals, characterizing their size distribution) from a stationary camera that views fish against the bottom of a body of water.
- Automatic interpretation (counting and characterizing) of individual snapshot images taken from a remotely operated moving camera.
- Automatic counting and characterization of fish in videos against a natural background.
NMFS will provide the committee with information about its current capabilities for collecting and analyzing images and video for these tasks. Based on that input, the committee will organize a workshop that will feature invited presentations and discussions involving participants from diverse fields to address the following topics:
- Identify promising directions for advancing NMFS’s analytical capabilities for the tasks listed above, including opportunities to leverage capabilities from other fields.
- Share perspectives about the most efficient path toward more automation of fisheries stock assessments, identifying goals that might be achieved through 3-5 years of modest R&D investment and goals that should be considered longer term.
One or more rapporteurs who are not members of the committee will be designated to prepare an individually authored or co-authored summary of the presentations and discussions at the event.