A low-cost, long-term underwater camera trap network coupled with deep residual learning image analysis

Abstract

Understanding long-term trends in marine ecosystems requires accurate and repeatable counts of fishes and other aquatic organisms on spatial and temporal scales that are difficult or impossible to achieve with diver-based surveys. Long-term, spatially distributed cameras, like those used in terrestrial camera trapping, have not been successfully applied in marine systems due to limitations of the aquatic environment. Here, we develop methodology for a system of low-cost, long-term camera traps (Dispersed Environment Aquatic Cameras), deployable over large spatial scales in remote marine environments. We use machine learning to classify the large volume of images collected by the cameras. We present a case study of these combined techniques’ use by addressing fish movement and feeding behavior related to halos, a well-documented benthic pattern in shallow tropical reefscapes. Cameras proved able to function continuously underwater at deployed depths (up to 7 m, with later versions deployed to 40 m) with no maintenance or monitoring for over five months and collected a total of over 100,000 images in time-lapse mode (by 15 minutes) during daylight hours. Our ResNet-50-based deep learning model achieved 92.5% overall accuracy in sorting images with and without fishes, and diver surveys revealed that the camera images accurately represented local fish communities. The cameras and machine learning classification represent the first successful method for broad-scale underwater camera trap deployment, and our case study demonstrates the cameras’ potential for addressing questions of marine animal behavior, distributions, and large-scale spatial patterns.

Artificial intelligence for fish behavior recognition may unlock fishing gear selectivity

“…The phytoplankton world is benefitting from citizen science as online portals are used by volunteers to do simple classification tasks that has led to millions of plankton ID's to be verified (Robinson et al, 2017). Moreover, scientists are adapting FAIR (Findability, Accessibility, Interoperability, and Reuse) data principles to realize the full value of fish behavior data and to carefully curate a unifying database (Guidi et al, 2020;Bilodeau et al, 2022).…”

“…As developments of non-invasive and autonomous underwater video cameras continue to advance (Graham et al, 2004;Moustahfid et al, 2020), behavioral observations can now be derived from a plethora of high-resolution marine imagery and videos (Logares et al, 2021). The reach of human vision continues to extend as cameras can be used in most conditions (Shafait et al, 2016;Christensen et al, 2018;Jalal et al, 2020), such as light, dark and muddy underwater conditions, and can go to greater depth and longer periods (Torres et al, 2020;Bilodeau et al, 2022;Xia et al, 2022). Cameras can now provide vision in 2D or 3D into how fishes interact with fishing gears used to capture marine species (e.g., pots, lines, trawls and nets) where behavior can be recorded by an observation system.…”

Image analysis and benthic ecology: Proceedings to analyze in situ long‐term image series

“…To our knowledge, this question has not specifically been studied before. However, within the development of underwater camera traps, automated classification algorithms are used to sort out images containing fish or not (Bilodeau et al 2022), and this part of the work can be compared to the detection of usable or nonusable images in this study.…”

Low-Cost, Deep-Sea Imaging and Analysis Tools for Deep-Sea Exploration: A Collaborative Design Study

“…Increasingly, emerging technologies for ocean exploration and research cost ~$10k-100k USD and are more portable, easier to operate, and offer a variety of capabilities, accuracy levels, and robustness (Sheehan et al, 2016;Dominguez-Carrió et al, 2021;Giddens et al, 2021). For the past few years, "do-ityourself " and open-sourced shallower tools (<300 m) have been developed using microcontrollers, single-board computers, and commercially available components to create camera and/or sensor systems within ~$100-$1000 USD (Simoncelli et al, 2019;Greene et al, 2020;Lertvilai, 2020;Mouy et al, 2020;Bilodeau et al, 2022;Butler and Pagniello, 2022). Two low-cost camera systems are designed for depths of 5,500-6,000 m (Phillips et al, 2019;Purser et al, 2020), and commercially available cameras such as GoPros can be after-market housed to ~3,000 m; none of these options, however, include sensors such as depth or temperature, which are critical for scientific understanding of the environment.…”