Improved accuracy for automated counting of a fish in baited underwater videos for stock assessment

Abstract

The ongoing need to sustainably manage fishery resources necessitates fishery-independent monitoring of the status of fish stocks. Camera systems, particularly baited remote underwater video stations (BRUVS), are a widely-used and repeatable method for monitoring relative abundance, required for building stock assessment models. The potential for BRUVS-based monitoring is restricted, however, by the substantial costs of manual data extraction from videos. Computer vision, in particular deep learning models, are increasingly being used to automatically detect and count fish at low abundances in videos. One of the advantages of BRUVS is that bait attractants help to reliably detect species in relatively short deployments (e.g. 1 hr). The high abundances of fish attracted to BRUVS, however, make computer vision more difficult, because fish often occlude other fish. We build upon existing deep learning methods for identifying and counting a target fisheries species across a wide range of fish abundances. Using BRUVS imagery targeting a recovering fishery species, Australian snapper (Chrysophrys auratus), we tested combinations of three further mathematical steps likely to generate accurate, efficient automation: 1) varying confidence thresholds (CTs), 2) on/off use of sequential non-maximum suppression (Seq-NMS), and 3) statistical correction equations. Output from the deep learning model was accurate at very low abundances of snapper; at higher abundances, however, the model over-predicted counts by as much as 50%. The procedure providing the most accurate counts across all fish abundances, with counts either correct or within 1 to 2 of manual counts (R2 = 93.4%), used Seq-NMS, a 55% confidence threshold, and a cubic polynomial corrective equation. The optimised modelling provides an automated procedure offering an effective and efficient method for accurately identifying and counting snapper in BRUV footage. Further testing is required to ensure that automated counts of snapper remain accurate in the survey region over time, and to determine the applicability to other regions within the distributional range of this species. For monitoring stocks of fishery species more generally, the specific equations will differ but the procedure demonstrated here would help to increase the usefulness of BRUVS, while decreasing costs.

Automatic detection of fish and tracking of movement for ecology

“…fish in a single frame), previous research has shown that occlusion can influence the accuracy of both the object detection algorithms and Seq-NMS (Connolly et al, 2021). Moreover, Seq-NMS is not an object tracking algorithm and it requires a high-performing object TA B L E 1 Object detection mAP50 and the evaluation results of the Mask R-CNN yellowfin bream model.…”

“…The number of fish that were simultaneously detected and tracked by our framework ranged from 1 to 30 individuals per video. While the movement dataset did not contain videos with a very large number of individuals (e.g., >50 fish in a single frame), previous research has shown that occlusion can influence the accuracy of both the object detection algorithms and Seq‐NMS (Connolly et al., 2021). Moreover, Seq‐NMS is not an object tracking algorithm and it requires a high‐performing object detection model because it uses the object detection outputs of every frame to create the detection links and track the movement direction.…”

Fish use of restored mangroves matches that in natural mangroves regardless of forest age

“…We supplemented the training dataset (i.e., transfer learning) with a previously trained model of the target species from southeast Queensland (Ditria et al 2021). In addition to the target species, we annotated five less common and patchily distributed species to increase the accuracy of predictions for target species (Connolly et al 2021). These five species were: sand whiting ( Sillago ciliata ), crescent grunter ( Terapon jarbua ), moses snapper ( Lutjanus russellii ), banded toadfish ( Marilyna pleurosticta ), and weeping toadfish ( Torquigener pleurogramma ).…”

Computer Vision Reveals Fish Behaviour Through Structural Equation Modelling of Movement Patterns

“…We selected Moses perch (Lutjanus russellii), southern herring (Herklotsichthys castelnaui), and yellow n bream (Acanthopagrus australis) as the target species, as these were common across all videos (> 1,500 annotations per species). We used a software developed at Gri th University for data preparation and annotation tasks (FishID -https://globalwetlandsproject.org/tools/ shid/), which has proven successful at a range of sh detection and tracking tasks [11,[45][46][47].…”