Inconsistency-based active learning with adaptive pseudo-labeling for fish species identification DOI
M M Nabi, Chiranjibi Shah, Simegnew Yihunie Alaba

et al.

Published: June 6, 2024

The deep neural network has found widespread application in object detection due to its high accuracy. However, performance typically depends on the availability of a substantial volume accurately labeled data. Several active learning approaches have been proposed reduce labeling dependency based confidence detector. Nevertheless, these tend exhibit biases toward high-performing classes, resulting datasets that do not adequately represent testing In this study, we introduce comprehensive framework for considers both uncertainty and robustness detector, ensuring superior across all classes. robustness-based score is calculated using consistency between an image augmented version. Additionally, leverage pseudo-labeling mitigate potential distribution drift enhance model performance. To address challenge setting threshold, adaptive threshold mechanism. This adaptability crucial, as fixed can negatively impact performance, particularly low-performing classes or during initial stages training. For our experiment, employ Southeast Area Monitoring Assessment Program Dataset 2021 (SEAMAPD21), comprising 130 fish species with 28,328 samples. results show outperforms state-of-the-art method significantly reduces annotation cost. Furthermore, benchmark model's against public dataset (PASCAL VOC07), showcasing effectiveness comparison existing methods.

Language: Английский

YOLOv8-TF: Transformer-Enhanced YOLOv8 for Underwater Fish Species Recognition with Class Imbalance Handling DOI Creative Commons
Chiranjibi Shah, M M Nabi, Simegnew Yihunie Alaba

et al.

Sensors, Journal Year: 2025, Volume and Issue: 25(6), P. 1846 - 1846

Published: March 16, 2025

In video-based fish surveys, species recognition plays a vital role in stock assessments, ecosystem analysis, production management, and protection of endangered species. However, implementing detection algorithms underwater environments presents significant challenges due to factors such as varying lighting conditions, water turbidity, the diverse appearances this work, transformer-enhanced YOLOv8 (YOLOv8-TF) is proposed for recognition. The YOLOv8-TF enhances performance by adjusting depth scales, incorporating transformer block into backbone neck, introducing class-aware loss function address class imbalance dataset. considers count instances within each assigns higher weight with fewer instances. This approach enables through object detection, encompassing classification localization estimate their position size an image. Experiments were conducted using 2021 Southeast Area Monitoring Assessment Program (SEAMAPD21) dataset, detailed extensive reef dataset from Gulf Mexico. experimental results on SEAMAPD21 demonstrate that model, mean Average Precision (mAP)0.5 87.9% mAP0.5–0.95 61.2%, achieves better compared state-of-the-art YOLO models. Additionally, publicly available datasets, Pascal VOC MS COCO datasets model outperforms existing approaches.

Language: Английский

Citations

0
Active detection for fish species recognition in underwater environments DOI
Chiranjibi Shah, M M Nabi, Simegnew Yihunie Alaba

et al.

Published: June 6, 2024

Fish species must be identified for stock assessments, ecosystem monitoring, production management, and the conservation of endangered species. Implementing algorithms fish detection in underwater settings like Gulf Mexico poses a formidable challenge. Active learning, method that efficiently identifies informative samples annotation while staying within budget, has demonstrated its effectiveness context object recent times. In this study, we present an active model designed recognition environments. This can employed as system to effectively lower expense associated with manual annotation. It uses epistemic uncertainty Evidential Deep Learning (EDL) proposes novel module denoted Model Evidence Head (MEH) employs Hierarchical Uncertainty Aggregation (HUA) obtain informativeness image. We conducted experiments using fine-grained extensive dataset reef collected from Mexico, specifically Southeast Area Monitoring Assessment Program Dataset 2021 (SEAMAPD21). The experimental results demonstrate framework achieves better performance on SEAMAPD21 demonstrating favorable balance between data efficiency recognition.

Language: Английский

Citations

1
Multi-fish tracking for marine biodiversity monitoring DOI
Simegnew Yihunie Alaba, Jack Prior, Chiranjibi Shah

et al.

Published: June 6, 2024

Accurate recognition of multiple fish species is essential in marine ecology and fisheries. Precisely classifying tracking these enriches our comprehension their movement patterns empowers us to create precise maps species-specific territories. Such profound insights are pivotal conserving endangered species, promoting sustainable fishing practices, preserving ecosystems' overall health equilibrium. To partially address needs, we present a proposed model that combines YOLOv8 for object detection with ByteTrack tracking. YOLOv8's oriented bounding boxes help improve across angles, while ByteTrack's robustness various scenarios makes it ideal real-time Experimental results using the SEAMAPD21 dataset show model's effectiveness, YOLOv8n being lightweight yet modestly accurate option, suitable constrained environments. The study also identifies challenges tracking, such as lighting variations appearance changes, proposes solutions future research. Overall, shows promising counting results, which monitoring life.

Language: Английский

Citations

0
Inconsistency-based active learning with adaptive pseudo-labeling for fish species identification DOI
M M Nabi, Chiranjibi Shah, Simegnew Yihunie Alaba

et al.

Published: June 6, 2024

The deep neural network has found widespread application in object detection due to its high accuracy. However, performance typically depends on the availability of a substantial volume accurately labeled data. Several active learning approaches have been proposed reduce labeling dependency based confidence detector. Nevertheless, these tend exhibit biases toward high-performing classes, resulting datasets that do not adequately represent testing In this study, we introduce comprehensive framework for considers both uncertainty and robustness detector, ensuring superior across all classes. robustness-based score is calculated using consistency between an image augmented version. Additionally, leverage pseudo-labeling mitigate potential distribution drift enhance model performance. To address challenge setting threshold, adaptive threshold mechanism. This adaptability crucial, as fixed can negatively impact performance, particularly low-performing classes or during initial stages training. For our experiment, employ Southeast Area Monitoring Assessment Program Dataset 2021 (SEAMAPD21), comprising 130 fish species with 28,328 samples. results show outperforms state-of-the-art method significantly reduces annotation cost. Furthermore, benchmark model's against public dataset (PASCAL VOC07), showcasing effectiveness comparison existing methods.

Language: Английский

Citations

0