A Lightweight Underwater Object Detection Model: FL-YOLOV3-TINY

Tan, Cong; CHEN, DanDan; Huang, Haijie; Yang, Qiuling; Huang, XiangDang

doi:10.1109/iemcon53756.2021.9623066

Cited by 6 publications

(8 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To demonstrate the efficiency of the proposed method, we select a set of related state‐of‐the‐art object detection methods as the baselines, including several typical two‐stage methods and one‐stage methods, as well as some underwater detection methods. They are listed as follows: Two‐Stage Methods: We choose the classic Faster R‐CNN [21] as a representative two‐stage algorithm, which first uses RPN to generate region proposals and then uses convolution to classify and regress these boxes. One‐Stage Methods: We select some classic anchor‐based algorithms, including SSD [24], RefineDet [28], RFBNet [66], LRF‐Net [67] and EFGRNet [68], and some anchor‐free algorithms, including Fully Convolutional One‐Stage Object Detection [FCOS] [69] and YOLOX [70] for comparison. Underwater Methods: As a new topic of object detection, there are few underwater object detection algorithms, we choose FERNet [15], Fl‐yolo3‐tiny [16], S‐fpn [47] as a baseline. …”

Section: Methodsmentioning

confidence: 99%

“…Underwater Methods: As a new topic of object detection, there are few underwater object detection algorithms, we choose FERNet [15], Fl‐yolo3‐tiny [16], S‐fpn [47] as a baseline.…”

Section: Methodsmentioning

confidence: 99%

“…Tan et al. [16] utilised the depthwise separable convolution to improve the performance of yolo3‐tiny in underwater environment. Liu et al.…”

Section: Introductionmentioning

confidence: 99%

“…For example, Fan et al [15] proposed an underwater object detection algorithm named FERNet, which utilises a Composite Connection Backbone (CCB) and a Receptive Field Augmentation Module to boost the feature representation. Tan et al [16] utilised the depthwise separable convolution to improve the performance of yolo3-tiny in underwater environment. Liu et al [17] proposed the Domain Generalization-YOLO method, which aims to improve the domain generalisation in underwater object detection.…”

mentioning

confidence: 99%

See 3 more Smart Citations

A multi‐scale feature representation and interaction network for underwater object detection

Yuan

Sun

et al. 2022

IET Computer Vision

View full text Add to dashboard Cite

Compared with natural images, underwater images are usually degraded with blur, scale variation, colour shift and texture distortion, which bring much challenge for computer vision tasks like object detection. In this case, generic object detection methods usually fail to achieve satisfactory performance. The main reason is considered that the current methods lack sufficient discriminativeness of feature representation for the degraded underwater images. A a novel multi-scale feature representation and interaction network for underwater object detection is proposed, in which two core modules are elaborately designed to enhance the discriminativeness of feature representation for underwater images. The first is the Context Integration Module, which extracts rich context information from high-level features and is integrated with the feature pyramid network to enhance the feature representation in a multi-scale way. The second is the Dual-refined Attention Interaction Module, which further enhances the feature representation by sufficient interactions between different levels of features both in channel and spatial domains based on attention mechanism. The proposed model is evaluated on four public underwater datasets. The experimental results compared with state-of-the-art object detection methods show that the proposed model has leading performance, which verifies that it is effective for underwater object detection. In addition, object detection experiments on a foggy dataset of Real-world Task-driven Testing Set (RTTS) and the natural image dataset of pattern analysis statistical modelling and computational learning, visual object classes (PASCAL VOC) are conducted. The results show that the proposed model can be applied on the degraded dataset of RTTS but fails on PASCAL VOC. K E Y W O R D S computer vision, convolutional neural nets, object detection | INTRODUCTIONAs a fundamental and important problem in computer vision, object detection has been continuously investigated in the past decades. With the development of deep learning in recent years, object detection has made great progression and been widely applied in various real-world applications, such as vehicle detection [1-4], pedestrian detection [5-9], intelligent monitoring [10-13], etc.Although the generic object detection algorithms have promising performance in most cases, they often fail to achieve convincing results under some specific scenes, such as underwater, foggy and dark light, etc. Compared with the common scenarios, the object detection in these specific scenes faces much challenge and difficulty to overcome, including image blur, texture distortion, imbalanced illumination, etc. For this purpose, many researchers began and focussed on the object detection in specific scenes. For example, underwater object detection [14] has raised much attention recently, as it is an important and common problem in the fields of seafood industry, underwater engineering and ocean development. Specifically, for the aquaculture industry, underwater robots ...

show abstract

Section: Methodsmentioning

confidence: 99%

“…Underwater Methods: As a new topic of object detection, there are few underwater object detection algorithms, we choose FERNet [15], Fl‐yolo3‐tiny [16], S‐fpn [47] as a baseline.…”

Section: Methodsmentioning

confidence: 99%

“…Tan et al. [16] utilised the depthwise separable convolution to improve the performance of yolo3‐tiny in underwater environment. Liu et al.…”

Section: Introductionmentioning

confidence: 99%

mentioning

confidence: 99%

See 2 more Smart Citations

A multi‐scale feature representation and interaction network for underwater object detection

Yuan

Sun

et al. 2022

IET Computer Vision

View full text Add to dashboard Cite

show abstract

“…Yingqiong Peng et al [4] proposed an improved image classification model for improved CNN, which replaced Softmax classifier with SVM and had a good accuracy in the application of fruit fly classification. Cong Tan et al [5] proposed a lightweight FL model using YOLOV3-Tiny, which has good data convergence and accuracy while ensuring data privacy. ISSN 2616-5775 Vol.…”

Section: Related Workmentioning

confidence: 99%

UAV Land Classification Method Based on Federated Learning

2022

AJCIS

View full text Add to dashboard Cite

In the application of unmanned aerial vehicle (UAV), land classification is an important field. Image recognition technology based on deep learning is a new method to solve the problem of land classification in recent years. In this paper, we propose a land classification method based on federated learning (FL) which uses Fedavg-Adam algorithm. This method enables UAVs to learn land models using training data distributed in their local database. By aggregating local computational updates of the land classification model, a shared global model is constructed. UAVs can collectively benefit from global models without sharing datasets and protect sensitive information collected by UAVs. To obtain good classification performance, we further propose an improved CNN network. The experimental results show that the improved CNN network suitable for federated learning has the good performance considering the influence of time and accuracy. In the RSSCN7 Dataset, the FedAvg-Adam algorithm converges in the 43rd rounds with an accuracy rate of 82.71%. Compared with FedAvg and FedAdam, the final accuracy of the three is similar, but the latter two converge at 56 rounds and 114 rounds respectively, and Fedavg-Adam has the fastest speed, which proves the superiority of our method.

show abstract

A simplified network topology for fruit detection, counting and mobile-phone deployment

Lawal,

Zhu,

Cheng

et al. 2023

PLoS ONE

View full text Add to dashboard Cite

The complex network topology, deployment unfriendliness, computation cost, and large parameters, including the natural changeable environment are challenges faced by fruit detection. Thus, a Simplified network topology for fruit detection, tracking and counting was designed to solve these problems. The network used common networks of Conv, Maxpool, feature concatenation and SPPF as new backbone and a modified decoupled head of YOLOv8 as head network. At the same time, it was validated on a dataset of images encompassing strawberry, jujube, and cherry fruits. Having compared to YOLO-mainstream variants, the params of Simplified network is 32.6%, 127%, and 50.0% lower than YOLOv5n, YOLOv7-tiny, and YOLOv8n, respectively. The results of mAP@50% tested using test-set show that the 82.4% of Simplified network is 0.4%, -0.2%, and 0.2% respectively more accurate than 82.0% of YOLOv5n, 82.6% of YOLOv7-tiny, and 82.2% of YOLOv8n. Furthermore, the Simplified network is 12.8%, 17.8%, and 11.8% respectively faster than YOLOv5n, YOLOv7-tiny, and YOLOv8n, including outperforming in tracking, counting, and mobile-phone deployment process. Hence, the Simplified network is robust, fast, accurate, easy-to-understand, fewer in parameters and deployable friendly.

show abstract

A Lightweight Underwater Object Detection Model: FL-YOLOV3-TINY

Cited by 6 publications

References 9 publications

A multi‐scale feature representation and interaction network for underwater object detection

A multi‐scale feature representation and interaction network for underwater object detection

UAV Land Classification Method Based on Federated Learning

A simplified network topology for fruit detection, counting and mobile-phone deployment

Contact Info

Product

Resources

About