Recent advances of machine vision technology in fish classification

ICES Journal of Marine Science, Journal Year: 2022, Volume and Issue: 79(2), P. 263 - 284

Published: Jan. 9, 2022

Abstract Automatic classification of different species fish is important for the comprehension marine ecology, behaviour analysis, aquaculture management, and health monitoring. In recent years, many automatic methods have been developed, among which machine vision-based are widely used with advantages being fast non-destructive. addition, successful application rapidly emerging deep learning techniques in vision has brought new opportunities classification. This paper provides an overview models applied field classification, followed by a detailed discussion specific applications various methods. Furthermore, challenges future research directions discussed. would help researchers practitioners to understand applicability encourage them develop advanced algorithms address complex problems that exist practice.

Language: Английский

---

Accelerating Species Recognition and Labelling of Fish From Underwater Video With Machine-Assisted Deep Learning

Frontiers in Marine Science, Journal Year: 2022, Volume and Issue: 9

Published: Aug. 2, 2022

Machine-assisted object detection and classification of fish species from Baited Remote Underwater Video Station (BRUVS) surveys using deep learning algorithms presents an opportunity for optimising analysis time rapid reporting marine ecosystem statuses. Training BRUVS significant challenges: the model requires training datasets with bounding boxes already applied identifying location all individuals in a scene, it labels. In both cases, substantial volumes data are required this is currently manual, labour-intensive process, resulting paucity labelled models detection. Here, we present “machine-assisted” approach i) generalised to automate application any underwater environment containing ii) identification level, up 12 target species. A catch-all “ ” that remain unidentified due lack available validation data. box annotation was shown detect label on out-of-sample recall between 0.70 0.89 automated labelling targeted F 1 score 0.79. On average, 12% were given labels 88% located identified manual labelling. Taking combined, machine-assisted advancement towards use workflows has potential future ecologist uptake if integrated into video software. Manual effort still required, community address limitation presented by severe would improve automation accuracy encourage increased uptake.

Language: Английский

---

‘Small Data’ for big insights in ecology

Trends in Ecology & Evolution, Journal Year: 2023, Volume and Issue: 38(7), P. 615 - 622

Published: Feb. 15, 2023

Big Data science has significantly furthered our understanding of complex systems by harnessing large volumes data, generated at high velocity and in great variety. However, there is a risk that collection prioritised to the detriment 'Small Data' (data with few observations). This poses particular ecology where Small abounds. Machine learning experts are increasingly looking drive next generation innovation, leading development methods for such as transfer learning, knowledge graphs, synthetic data. Meanwhile, meta-analysis causal reasoning approaches evolving provide new insights from Data. These advances should add value high-quality catalysing future ecology.

Language: Английский

---

Image dataset for benchmarking automated fish detection and classification algorithms

Scientific Data, Journal Year: 2023, Volume and Issue: 10(1)

Published: Jan. 3, 2023

Abstract Multiparametric video-cabled marine observatories are becoming strategic to monitor remotely and in real-time the ecosystem. Those platforms can achieve continuous, high-frequency long-lasting image data sets that require automation order extract biological time series. The OBSEA, located at 4 km from Vilanova i la Geltrú 20 m depth, was used produce coastal fish series continuously over 24-h during 2013–2014. content of photos extracted via tagging, resulting 69917 tags 30 taxa identified. We also provided a meteorological oceanographic dataset filtered by quality control procedure define real-world conditions affecting quality. tagged be great importance develop Artificial Intelligence routines for automated identification classification fishes extensive time-lapse sets.

Language: Английский

---

A multitask model for realtime fish detection and segmentation based on YOLOv5

PeerJ Computer Science, Journal Year: 2023, Volume and Issue: 9, P. e1262 - e1262

Published: March 10, 2023

The accuracy of fish farming and real-time monitoring are essential to the development "intelligent" farming. Although existing instance segmentation networks (such as Maskrcnn) can detect segment fish, most them not effective in monitoring. In order improve image promote accurate intelligent industry, this article uses YOLOv5 backbone network object detection branch, combined with semantic head for segmentation. experiments show that precision reach 95.4% 98.5% algorithm structure proposed article, based on golden crucian carp dataset, 116.6 FPS be achieved RTX3060. On publicly available dataset PASCAL VOC 2007, is 73.8%, 84.3%, speed up 120

Language: Английский

---

Fish Detection under Occlusion Using Modified You Only Look Once v8 Integrating Real-Time Detection Transformer Features

Applied Sciences, Journal Year: 2023, Volume and Issue: 13(23), P. 12645 - 12645

Published: Nov. 24, 2023

Fish object detection has attracted significant attention because of the considerable role that fish play in human society and ecosystems necessity to gather more comprehensive data through underwater videos or images. However, always faced difficulties with occlusion problem dense populations plants obscure them, no perfect solution been found until now. To address issue detection, following effort was made: creating a dataset occluded fishes, integrating innovative modules Real-time Detection Transformer (RT-DETR) into You Only Look Once v8 (YOLOv8), applying repulsion loss. The results show dataset, mAP original YOLOv8 is 0.912, while our modified 0.971. In addition, also better performance than terms loss curves, F1–Confidence P–R curve actual effects. All these indicate suitable for scenes.

Language: Английский

---

High-resolution density assessment assisted by deep learning of Dendrophyllia cornigera (Lamarck, 1816) and Phakellia ventilabrum (Linnaeus, 1767) in rocky circalittoral shelf of Bay of Biscay

PeerJ, Journal Year: 2024, Volume and Issue: 12, P. e17080 - e17080

Published: March 7, 2024

This study presents a novel approach to high-resolution density distribution mapping of two key species the 1170 “Reefs” habitat, Dendrophyllia cornigera and Phakellia ventilabrum , in Bay Biscay using deep learning models. The main objective this was establish pipeline based on models extract data from raw images obtained by remotely operated towed vehicle (ROTV). Different object detection were evaluated compared various shelf zones at head submarine canyon systems metrics such as precision, recall, F1 score. best-performing model, YOLOv8, selected for generating maps high spatial resolution. also generated synthetic augment training assess generalization capacity proposed provides cost-effective non-invasive method monitoring assessing status these important reef-building their habitats. results have implications management protection habitat Spain other marine ecosystems worldwide. These highlight potential improve efficiency accuracy vulnerable ecosystems, allowing informed decisions be made that can positive impact conservation.

Language: Английский

---

Multi-classification deep neural networks for identification of fish species using camera captured images

PLoS ONE, Journal Year: 2023, Volume and Issue: 18(4), P. e0284992 - e0284992

Published: April 26, 2023

Regular monitoring of the number various fish species in a variety habitats is essential for marine conservation efforts and biology research. To address shortcomings existing manual underwater video sampling methods, plethora computer-based techniques are proposed. However, there no perfect approach automated identification categorizing species. This primarily due to difficulties inherent capturing videos, such as ambient changes luminance, camouflage, dynamic environments, watercolor, poor resolution, shape variation moving fish, tiny differences between certain study has proposed novel Fish Detection Network (FD_Net) detection nine different types using camera-captured image that based on improved YOLOv7 algorithm by exchanging Darknet53 MobileNetv3 depthwise separable convolution 3 x filter size augmented feature extraction network bottleneck attention module (BNAM). The mean average precision (mAP) 14.29% higher than it was initial version YOLOv7. utilized method features an DenseNet-169, loss function Arcface Loss. Widening receptive field improving capability achieved incorporating dilated into dense block, removing max-pooling layer from trunk, BNAM block DenseNet-169 neural network. results several experiments comparisons ablation demonstrate our FD_Net mAP YOLOv3, YOLOv3-TL, YOLOv3-BL, YOLOv4, YOLOv5, Faster-RCNN, most recent model, more accurate target tasks complex environments.

Language: Английский

---

Enhancing fish freshness prediction using NasNet-LSTM

Journal of Food Composition and Analysis, Journal Year: 2023, Volume and Issue: 127, P. 105945 - 105945

Published: Dec. 23, 2023

Language: Английский

---

A critical review of machine-learning for “multi-omics” marine metabolite datasets

Computers in Biology and Medicine, Journal Year: 2023, Volume and Issue: 165, P. 107425 - 107425

Published: Aug. 29, 2023

Language: Английский

---