Out of the shadows: automatic fish detection from acoustic cameras

Abstract

Efficacious monitoring of fish stocks is critical for efficient management. Multibeam acoustic cameras, that use sound-reflectance to generate moving pictures, provide an important alternative to traditional video-based methods that are inoperable in turbid waters. However, acoustic cameras, like standard video monitoring methods, produce large volumes of imagery from which it is time consuming and costly to extract data manually. Deep learning, a form of machine learning, can be used to automate the processing and analysis of acoustic data. We used convolutional neural networks (CNNs) to detect and count fish in a publicly available dual-frequency identification sonar (DIDSON) dataset. We compared three types of detections, direct acoustic, acoustic shadows, and a combination of direct and shadows. The deep learning model was highly reliable at detecting fish to obtain abundance data using acoustic data. Model accuracy for counts-per-image was improved by the inclusion of shadows (F1 scores, a measure of the model accuracy: direct 0.79, shadow 0.88, combined 0.90). Model accuracy for MaxN per video was high for all three types of detections (F1 scores: direct 0.90, shadow 0.90, combined 0.91). Our results demonstrate that CNNs are a powerful tool for automating underwater acoustic data analysis. Given this promise, we suggest broadening the scope of testing to include a wider range of fish shapes, sizes, and abundances, with a view to automating species (or ‘morphospecies’) identification and counts.

Identification and counting of fish targets using adaptive resolution imaging sonar

“…The DS tracking algorithm is mainly used to predict the target. Compared with the method of tracking and counting using shadows (Connolly et al ., 2022), the necessary statistical data for making the final judgement are lacking, but the programme is more convenient. Furthermore, the overlapping of information, due to the projected beam volume, hinders detecting edges and segmenting the fish, which the algorithm recognizes as one fish body.…”

Anadromous trout from opposite sides of the globe: biology, ocean ecology, and management of anadromous brown and cutthroat trout

“…In some areas, traditional tools like redd counts, riverscapes, tagging (Floy, elastomer and PIT tag technology) and genetics, have been applied and tailored specifically for anadromous trout given their diverse life histories providing great promise for these two species and the management around them. Additionally, new tools being developed for other focal species will be available to apply to anadromous trout in the near future including SONAR (Gaudet 1990;Connolly et al 2022), seismic monitoring (Dietze et al 2020) and drones (Groves et al 2016) to improve estimates of abundance.…”

Automated fish detection and classification on sonar images using detection transformer and YOLOv7

“…Connolly et al 9 used an approach based on convolutional neural networks (CNNs) to detect and count fish in a publicly available DIDSON data captured near the Ocqueoc River in Michigan (USA). They evaluated three types of detection, direct acoustic, acoustic shadows, and a combination of direct and shadows.…”

“…• Using all fish species available in the DIDSON visual acoustic video dataset for fish detection and classification, which is not available in previous related work. 4,9 By including a wider range of fish species, the analysis will be more extensive and the accuracy of fish detection and classification will be enhanced.…”