A YOLOv5 Baseline for Underwater Object Detection

Wang, Hao; Sun, Shixin; Wu, Xiaohui; Li, Li; Zhang, Hao; Li, Mingjie; Ren, Peng

doi:10.23919/oceans44145.2021.9705896

Cited by 24 publications

(7 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…You Only Look Once version 5 (YOLOv5) is a state-of-the-art object-detection model known for its efficiency and high performance [51][52][53]. At its core, YOLOv5 is designed to identify and localize multiple objects in images or video feeds, executing these tasks with high accuracy.…”

Section: Object Detection Using Yolov5 Architecturementioning

confidence: 99%

Improving SLAM Techniques with Integrated Multi-Sensor Fusion for 3D Reconstruction

Cai,

Ou,

Qin

2024

Sensors

View full text Add to dashboard Cite

Simultaneous Localization and Mapping (SLAM) poses distinct challenges, especially in settings with variable elements, which demand the integration of multiple sensors to ensure robustness. This study addresses these issues by integrating advanced technologies like LiDAR-inertial odometry (LIO), visual-inertial odometry (VIO), and sophisticated Inertial Measurement Unit (IMU) preintegration methods. These integrations enhance the robustness and reliability of the SLAM process for precise mapping of complex environments. Additionally, incorporating an object-detection network aids in identifying and excluding transient objects such as pedestrians and vehicles, essential for maintaining the integrity and accuracy of environmental mapping. The object-detection network features a lightweight design and swift performance, enabling real-time analysis without significant resource utilization. Our approach focuses on harmoniously blending these techniques to yield superior mapping outcomes in complex scenarios. The effectiveness of our proposed methods is substantiated through experimental evaluation, demonstrating their capability to produce more reliable and precise maps in environments with variable elements. The results indicate improvements in autonomous navigation and mapping, providing a practical solution for SLAM in challenging and dynamic settings.

show abstract

Section: Object Detection Using Yolov5 Architecturementioning

confidence: 99%

Improving SLAM Techniques with Integrated Multi-Sensor Fusion for 3D Reconstruction

Cai,

Ou,

Qin

2024

Sensors

View full text Add to dashboard Cite

show abstract

“…For object (class "fish") detection, we first compared the performance of Faster RCNN (Ren et al, 2015) and several configurations of the fifth version of the You Only Look Once (YOLO) algorithm (first described by Redmon et al, 2016), using the implementation from Ultralytics (https://github.com/ultralytics/yolov5 ). YOLOv5 has been shown to work particularly well in underwater environments (Wang et al, 2021). The medium pre-trained model from YOLOv5, YOLOv5m (pre-trained on COCO image database, http://cocodataset.org/ ) was selected after training on the E0 scenario with the l, m and x pre-trained models (Supplementary Table S1).…”

Section: Object Detectionmentioning

confidence: 99%

Automatic detection and classification of coastal Mediterranean fish from underwater images: Good practices for robust training

et al. 2023

View full text Add to dashboard Cite

Further investigation is needed to improve the identification and classification of fish in underwater images using artificial intelligence, specifically deep learning. Questions that need to be explored include the importance of using diverse backgrounds, the effect of (not) labeling small fish on precision, the number of images needed for successful classification, and whether they should be randomly selected. To address these questions, a new labeled dataset was created with over 18,400 recorded Mediterranean fish from 20 species from over 1,600 underwater images with different backgrounds. Two state-of-the-art object detectors/classifiers, YOLOv5m and Faster RCNN, were compared for the detection of the ‘fish’ category in different datasets. YOLOv5m performed better and was thus selected for classifying an increasing number of species in six combinations of labeled datasets varying in background types, balanced or unbalanced number of fishes per background, number of labeled fish, and quality of labeling. Results showed that i) it is cost-efficient to work with a reduced labeled set (a few hundred labeled objects per category) if images are carefully selected, ii) the usefulness of the trained model for classifying unseen datasets improves with the use of different backgrounds in the training dataset, and iii) avoiding training with low-quality labels (e.g., small relative size or incomplete silhouettes) yields better classification metrics. These results and dataset will help select and label images in the most effective way to improve the use of deep learning in studying underwater organisms.

show abstract

“…As mentioned before, YOLO is one of the most used detection algorithms. In [28], the fifth version of that algorithm was tested and compared versus other algorithms such as faster RCNN or a fully convolutional one-stage (FCOS) method, and the results showed that the small version of YOLOv5 reached an mAP of 62.7% and around 50 FPS showing the best-combined results of the YOLOv5 versions.…”

Section: Related Workmentioning

confidence: 99%

Inspection Operations and Hole Detection in Fish Net Cages through a Hybrid Underwater Intervention System Using Deep Learning Techniques

López-Barajas,

Sanz,

Marín-Prades

et al. 2023

JMSE

View full text Add to dashboard Cite

Net inspection in fish-farm cages is a daily task for divers. This task represents a high cost for fish farms and is a high-risk activity for human operators. The total inspection surface can be more than 1500 m2, which means that this activity is time-consuming. Taking into account the severe restrictions for human operators in such hostile underwater conditions, this activity represents a significant area for improvement. A platform for net inspection is proposed in this work. This platform includes a surface vehicle, a ground control station, and an underwater vehicle (BlueROV2 heavy) which incorporates artificial intelligence, trajectory control procedures, and the necessary communications. In this platform, computer vision was integrated, involving a convolutional neural network trained to predict the distance between the net and the robot. Additionally, an object detection algorithm was developed to recognize holes in the net. Furthermore, a simulation environment was established to evaluate the inspection trajectory algorithms. Tests were also conducted to evaluate how underwater wireless communications perform in this underwater scenario. Experimental results about the hole detection, net distance estimation, and the inspection trajectories demonstrated robustness, usability, and viability of the proposed methodology. The experimental validation took place in the CIRTESU tank, which has dimensions of 12 × 8 × 5 m, at Universitat Jaume I.

show abstract

A YOLOv5 Baseline for Underwater Object Detection

Cited by 24 publications

References 9 publications

Improving SLAM Techniques with Integrated Multi-Sensor Fusion for 3D Reconstruction

Improving SLAM Techniques with Integrated Multi-Sensor Fusion for 3D Reconstruction

Automatic detection and classification of coastal Mediterranean fish from underwater images: Good practices for robust training

Inspection Operations and Hole Detection in Fish Net Cages through a Hybrid Underwater Intervention System Using Deep Learning Techniques

Contact Info

Product

Resources

About