Elsevier

Measurement

Volume 126, October 2018, Pages 72-75
Measurement

Automated estimate of fish abundance through the autonomous imaging device GUARD1

https://doi.org/10.1016/j.measurement.2018.05.035Get rights and content

Highlights

  • A standalone imaging device for underwater image acquisition and processing is presented.

  • An automated image recognition approach for the macro- and mega- fauna is presented.

  • The automated recognition was assessed on a relevant image dataset.

  • The potentiality of such a device within the Argo programme is discussed.

Abstract

Many technologies have been developed for monitoring the ocean interior. Among them the monitoring approaches based on imaging devices are capable to disclose important data on species behaviour and spatio-temporal variations of richness and evenness. In this context, the Argo programme (http://doi.org/10.17882/42182) is a valuable instrument for monitoring the deep sea at global scale in space and time. Argo floats equipped with imaging devices are candidate to become a new monitoring tool for studying macro- and mega-fauna in large areas and for extended time periods, potentially providing monitoring results never attained before. This work summarises the results obtained on the automated fish recognition task performed on the images acquired by the GUARD1 imaging device1.

Introduction

The deep sea (200 m depth) encompasses the 95% of the ocean volume, it is largely unexplored and presents important challenges such as new species discovery, biodiversity richness and evenness assessment, impact evaluation of global changes and the management of ecosystem services [1].

Many technologies have been developed so far for monitoring the ocean interior [2], [1]. Among them, the monitoring approaches based on imaging devices are capable to disclose important data on species behaviour and spatio-temporal variation of richness and evenness [3], [4], especially as computer vision and artificial intelligence methodologies are becoming more reliable and capable to compete with the traditional visual inspection of data [5]. In this context, most of the currently available imaging devices have been designed to be towed by support vessels [6] or to be installed on cabled observatories [7]. In general the monitoring action performed by towed instruments is expensive, due to the need of supporting vessels, and limited to specific transects. Moreover, the temporal continuity of the data acquisition is often not guaranteed. In contrast, cabled observatories provide temporal continuity and high frequency data acquisition. Nevertheless, their monitoring action is strictly limited to the seabed and in particular to the surrounding of the observatory itself, thus limiting the acquisition of information to small areas of interest.

Also the Autonomous Underwater Vehicles (AUV) are widely used for monitoring the ocean [8]. Although expensive technologies allow for measurements at large temporal and spatial scales [9], [10], in general the AUVs are used for monitoring limited areas for short periods of time. Among the existing different AUVs, the Argo floats profilers are low-cost vehicles and are capable to monitor the ocean down to 2000 m depth, for a considerable long period (4 years) [11]. The Argo Float Programme in the last 15 years collected almost 1.5 millions of water column profiles [12] and at present more than 3800 floats are monitoring the deep sea, as shown in Fig. 1.

Even if in recent time new devices have been designed and tested for the acquisition of ecological data, as for example chlorophyll and dissolved oxygen, traditionally the Argo Float profilers are equipped with instruments for the measurement of physical quantities like temperature, water conductibility and salinity [12]. Despite these new technological advances, very few attempts have been made for equipping the Argo floats profilers with imaging devices capable of monitoring high complexity pelagic ecosystem components such as the deep meso- and bathypelagic fauna. Such imaging devices would transform the Argo float profilers into autonomous sensors for quantifying biotic components at a global scale and for extended periods of time, being this kind of monitoring activity at the core of important ecosystem services such as food provisioning (e.g. fisheries), maintaining of nursery population and recreation (e.g. nature viewing, recreational fishing) [13].

Following the work discussed in [14], [15], [16], [17], this paper presents the research progress so far obtained in the study, development and experimentation of the GUARD1 imaging device [18], designed to be installed on board on Argo floats profilers for the automated image acquisition and recognition of the macro- and mega- fauna populating the water column. The GUARD1 imaging device has been designed for acquiring images, recognising and extracting the relevant image content and for transmitting the extracted information. The image acquisition capabilities have been widely experimented in different application contexts [14], [15], [16] and the algorithms for the automated recognition of gelatinous zooplankton (e.g. Ctenophora) have been also studied and validated [16]. This work discusses the approach for the automated recognition and counting of fishes specifically conceived to be executed on board the GUARD1 imaging device. Experiments were performed in laboratory based on images acquired by the GUARD1 in a real application context. The automated recognition results were compared to the manual counting of the fishes occurring in the images, showing a strong correlation between recognised and observed fish abundance.

Section snippets

The GUARD1 imaging device

The GUARD1 imaging device was specifically designed as a stand-alone system capable of acquiring and automatically process the content of underwater images. It is equipped with its own lighting system and it is capable to run image processing algorithms for the automated recognition of the relevant image content. The information extracted from the acquired images can be either stored on board the system or transmitted through a dedicated communication device [18]. The currently developed

The image dataset

The images used for learning and validating the automated recognition algorithm were acquired by the GUARD1 in the period February-May 2107 at the Acqua Alta Oceanographic Tower. These images were acquired every 10 min, continuously during the daylight and the night, for a total of 12331 images. The most critical acquisition conditions affecting the automated recognition performance are shown in Fig. 2. Fig. 2a shows an image acquired during the daylight with transparent water. In contrast,

Automated Fish Recognition

The image recognition methodology proposed in this work combines an image segmentation process with a Supervised Machine Learning approach based on Genetic Programming, coupled with a K-fold Cross-Validation framework as discussed in [20], [16], [21].

Discussion

One of the most urgent tasks for the coming decades is the development of technologies for continuously tracking and accurately predicting the biological responses to human impacts and global climate changes. According to these needs, the estimate of the deep pelagic biodiversity and dynamics, at different scales in time and space, is of strategic importance for the scientific community. In this context, many methodologies for the automated recognition, classification and counting of fishes

Acknowledgements

This work was partially supported by the TISANA project within the Argo Italy programme and by the project FixO3 (FP7/2013-2017) under the grant agreement n 312463, through the TNA project FISHAUT.

References (34)

  • N. MacLeod et al.

    Time to automate identification

    Nature

    (2010)
  • M.C. Benfield et al.

    Rapid: research on automated plankton identification

    Oceanography

    (2007)
  • J. Aguzzi et al.

    Coastal observatories for monitoring of fish behaviour and their responses to environmental changes

    Rev. Fish Biol. Fish.

    (2015)
  • K.J. Morris et al.

    A new method for ecological surveying of the abyss using autonomous underwater vehicle photography

    Limnology Oceanography: Methods

    (2014)
  • C.C. Eriksen et al.

    Seaglider: a long-range autonomous underwater vehicle for oceanographic research

    IEEE J. Oceanic Eng.

    (2001)
  • D.T. Roper, A.B. Phillips, C.A. Harris, G. Salavasidis, M. Pebody, R. Templeton, S.V.S. Amma, M. Smart, S. McPhail,...
  • Argo, part of the global observation system, http://www.argo.ucsd.edu/, (accessed:...
  • Cited by (33)

    • Fast and accurate mapping of fine scale abundance of a VME in the deep sea with computer vision

      2022, Ecological Informatics
      Citation Excerpt :

      These innovations are influencing how remote sensing is used in ecology to circumvent the bottleneck of manual labour and CV is now commonly employed in vegetation monitoring (Kattenborn et al., 2021) and animals detection on camera trap footage (Norouzzadeh et al., 2018; Tabak et al., 2019; Whytock et al., 2021). In the marine environment, CV has been applied to the study of coral reefs (Beijbom et al., 2015; González-Rivero et al., 2020; Pavoni et al., 2020; Williams et al., 2019) and fish community diversity and abundance (Ditria et al., 2021; Marini et al., 2018). It has also successfully been used to study the diversity, abundance and distribution of benthic communities (Abad-Uribarren et al., 2022; Durden et al., 2021; Marburg and Bigham, 2016; Marini et al., 2022; Möller and Nattkemper, 2021; Piechaud et al., 2019; Schoening et al., 2012) and these authors report satisfying performances.

    • Assessing plume impacts caused by polymetallic nodule mining vehicles

      2022, Marine Policy
      Citation Excerpt :

      Although large areas of seafloor may need to be surveyed it is feasible to obtain sufficient spatial coverage and sample replication using image platforms such as AUVs [73] and crawlers [9,16], which can monitor changes of benthic community structure along a gradient away from the disturbance sites. The feasibility of this approach would be limited by the time needed to analyse large numbers for seafloor images, although improvements in automated image analysis via Artificial Intelligence can help to address this in the future [46–48]. The investigative methodology is also critical since small changes in the operational altitude of AUVs or ROVs, illumination and lens type can alter estimations of community structure [68] especially if comparisons are to be made between the impacts of different mining vehicles under similar operational conditions.

    • AI Naturalists Might Hold the Key to Unlocking Biodiversity Data in Social Media Imagery

      2020, Patterns
      Citation Excerpt :

      This increases the difficulty, but recent progress in automated plant species identification35 reinforces our belief that this type of study will become easier in the years to come. Because of similar results obtained on other phyla,3,36,37 we are confident this approach could soon be adapted for use well beyond plants, for example, to corals, fish, or birds. Biases in our data arise in part from differences between the aims of the original data collectors (i.e., the photographers) and our aims as biodiversity researchers and ecologists.

    • Ecological video monitoring of Marine Protected Areas by underwater cabled surveillance cameras

      2020, Marine Policy
      Citation Excerpt :

      Time lapse photography can be used to observe species and their behaviour in situ, deriving high-value ecological knowledge about local abundance, biomass, trophic features, and biodiversity [27]. In situ video monitoring can provide a record of temporal changes in the number of individuals and assemblage structure providing insights on the daily (tidal or 24 h) cycles, seasonal and long-term temporal changes [23–25,28]. Since field sampling is typically scattered over time, behavioural rhythms can influence our biodiversity assessment [29–33].

    View all citing articles on Scopus
    1

    This paper is an extended version of the conference paper, Marini et al., An Autonomous Imaging System for Argo Floats, IMEKO International Conference on Metrology for The Sea, Naples-Italy, October 11–13.

    View full text