CRIMAC - Centre for research-based innovation in marine acoustic abundance estimation and backscatter classification

The primary objective of the SFI is to advance the frontiers in fisheries acoustic methodology and associated optical methods, and to apply such methods to 1) surveys for marine organisms, 2) fisheries, 3) aquaculture and 4) the energy sector. What are and how do the various parts of marine organisms contribute to broadband backscatter? Complex broadband frequency responses from marine organisms are explored through various numerical models. We have tested and established working models to predict acoustic backscatter from a range of different frequencies and target properties. Numerical stability and run time are key benchmark metrics, and different implementations have been tested. CRIMAC has developed an efficient infrastructure to model acoustic backscatter. The standard signal processing stages of the Simrad EK80 broadband echosounders have been published. This includes example code, an accompanying paper and an efficient implementation within the commercially available LSSS system. The approach was presented at the “CRIMAC Workshop on data processing methods for large volumes of plankton and fisheries-acoustics data”, with attendance from 18 countries and 49 different institutions. What are the broadband frequency responses of marine organisms and other scatterers? IMR has acquired broad banded echosounder data on zooplankton, mackerel, herring, and demersal fish, from net pens, probes, research vessels and fishing vessels . A large test data set covering the range of different use cases has been published through the Norwegian Marine Data Centre. This is a useful resource for the community when developing and testing methods. IMR has built a new marine research facility that includes 16 12x12 m2 net pens at the Austevoll research station. CRIMAC has contributed to the planning of the facility, and an echosounder rig has been developed to fit the net pen allowing us to efficiently perform echosounder measurements on a range of species. Measuring the size of fish is important for the fishing industry, the fish farming industry and for scientific surveys, and we are working with various strategies to estimate length from the broadband signal. A second net-pen experiment has been carried out on salmon while growing to assess the change in BB backscatter as the fish grows. Datasets covering a range of sizes have been secured and will be used to test different algorithms for fish sizing. What are the organisms and targets that generate broadband backscatter? The taxonomic resolution of acoustic data is limited, and ground truthing methods are required to reliably classify acoustic backscatter to species or species groups. Scantrol Deep Vision has developed an optical system to mount inside trawls, and CRIMAC has implemented machine learning methods to interpret the image stream coming from the unit as well as integrated this into the IMR workflow. Clear images from in-trawl camera systems are necessary for using optical methods near the seabed, both for commercial and scientific purposes. When sampling close to the bottom the sediment cloud will affect image quality. Scantrol has developed a new version of the system called "CamSounder”. This is a lighter version and has the potential to be deployable on any vessel, more like a conventional trawl sensor, and the plan is to compare the result from the new sensor with the previous sensor. Can machine learning techniques reliably and accurately categorize acoustic backscatter? Modern machine learning algorithms can be used on large volumes on historical acoustic data, and datasets from long time series of acoustic surveys have been curated. Algorithms to improve the performance of machine learning algorithms for acoustic target classification have been developed, including semi-supervised algorithms. We have developed sampling strategies to address imbalance in the classes when training the algorithms. Self-supervised methods have been trained on the historical data archive using the DINO architecture. The method has been tested on the acoustic target classification as a downstream task. How to utilize acoustic sensors on autonomous platforms, assess uncertainty and utilize the effect of behaviour on acoustic backscatter? The introduction of autonomous or remotely controlled platforms provides an efficient way to deploy acoustic sensors. The Sounder platform developed by KD carries broad band echosounders, and the platform has been tested on two different CRIMAC cruises, including data quality and operational capabilities. We arranged a CRIMAC Workshop on Fisheries acoustics and Uncrewed Surface Vehicles, with attendance from several national and international institutions. The workshop report has been published in ICES Journal of Marine Science.

Fisheries acoustics is used to monitor the largest fish and krill stocks in the world’s oceans and to study marine ecosystems. A modern fishery without acoustic tools for detection, inspection and monitoring of seabed, schools, and the catching process is unthinkable. New wideband echo sounders offer a new opportunity in this arena for Norwegian science and industry. Science and fishing vessels can not only observe the echo amplitude and density of fish under the vessel, but also utilize the backscattered echo spectrum from the organisms. For simplicity, we prefer to define this as the echo dialect of the objects, as for example, an echo from an individual herring is affected by body shape, swim bladder, body constituent and behavior, and is different from the mackerel “echo dialect". We propose that systematic experimental and in situ research can be used to understand and interpret the different echo dialects from fish and marine organisms. We will further expand on existing multifrequency methods for classification and target sizing by utilizing modern machine learning techniques. This will improve the accuracy of existing monitoring methods and help the fishing skipper to make good catch decisions. Further, direct optical observations from the trawl and use of active selection devises will reduce bycatch. For accurate verification of acoustic recordings, we need continuous optical information from the trawl cod end. This will be achieved with the Scantrol DeepVision system, here tested with active selection devices, and open/closing nets. Discrete samples may then be taken sequentially in deep water, such as in mesopelagic communities. Wideband technology has been miniaturized and can be installed in probes, bottom landers, and surface and underwater unmanned vehicles (drones). We will assess how these can improve scientific monitoring by increased adaptive sampling, and how drones can be used in fishing for forward-mapping and inspection prior to catching.