Figure - available via license: Creative Commons Attribution 4.0 International
Content may be subject to copyright.
Source publication
Marine organisms detection based on machine vision requires high real-time performance and accuracy. Network feature extraction is frequently made more difficult when underwater robots collect information on the seafloor because of the uneven distribution of light on the seafloor, the significant impact of water waves, and the complexity of the sea...
Context in source publication
Context 1
... targets are divided into small (small) (area <32²), medium (medium) (32²<area <96²), and large (large) (area >96²) categories to be evaluated separately. As shown in Table 2 its evaluation experiment 01 is YOLOv4-tiny without any improvement; in experiment 02 only the CA module is added and it can be seen in the evaluation metrics that compared to experiment 01, the AP metrics of small and medium targets have slightly improved and the mAP metrics have improved by 1.6%. In Experiment 03, only the HDC module is added and compared with Experiment 01, the AP metrics of medium and large targets are slightly improved, and the mAP metrics are improved by 1.7%. ...Citations
... Model Algorithm description Improvement MODA [43] YOLOv4 Cross-stage partial connections (CSP), SAM spatial attention module (SAM) and PAN Coordinated attention module in backbone for feature extraction and hybrid dilated convolutions into FPN structure to expand the feature map perception field MAD-YOLO [13] YOLOv5 Cross-stage convolution with a large window size Multiscale feature extraction and attention mechanism reinforced PAN for feature fusion DAS [12] Spatial pyramid pool structure based on the samesized fast mixed pool layers YOLOXT [44] YOLOX Decoupled head, anchor-free Deformable coordinate attention (DECA)module, feature pyramid swin transformer, improved path aggregation network (FPST-PAN) for marine benthic target diversity EFP-YOLO [45] CSP with interactive global attention (IGACSP) in backbone and asymmetric task focused head (ATFhead) for enhanced scale perception Underwater-YCC [23] YOLOv7 Extended efficient layer aggregation network (E-ELAN) CBAM and Conv2Former in neck for the underwater blurry image network U-YOLOv7 [11] Cross-Conv with an efficient squeeze-excitation module UUV [15] YOLOv8 Cross-stage partial bottleneck with two convolutions (C2f) ...
In this paper, we propose a underwater target detection method that optimizes YOLOv8s to make it more suitable for real-time and underwater environments. First, a lightweight FasterNet module replaces the original backbone of YOLOv8s to reduce the computation and improve the performance of the network. Second, we modify current bi-directional feature pyramid network into a fast one by reducing unnecessary feature layers and changing the fusion method. Finally, we propose a lightweight-C2f structure by replacing the last standard convolution, bottleneck module of C2f with a GSConv and a partial convolution, respectively, to obtain a lighter and faster block. Experiments on three underwater datasets, RUOD, UTDAC2020 and URPC2022 show that the proposed method has mAP50\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$_{50}$$\end{document} of 86.8%, 84.3% and 84.7% for the three datasets, respectively, with a speed of 156 FPS on NVIDIA A30 GPUs, which meets the requirement of real-time detection. Compared to the YOLOv8s model, the model volume is reduced on average by 24%, and the mAP accuracy is enhanced on all three datasets.
... However, it compresses global spatial information into one channel descriptor, making it challenging to retain positional information crucial for capturing spatial structure in vision tasks. To address this limitation, CA decomposes the global pooling specified in Equation (1) into two 1D feature encoding operations [43]. ...
Due to the issues of remote sensing object detection algorithms based on deep learning, such as a high number of network parameters, large model size, and high computational requirements, it is challenging to deploy them on small mobile devices. This paper proposes an extremely lightweight remote sensing aircraft object detection network based on the improved YOLOv5n. This network combines Shufflenet v2 and YOLOv5n, significantly reducing the network size while ensuring high detection accuracy. It substitutes the original CIoU and convolution with EIoU and deformable convolution, optimizing for the small-scale characteristics of aircraft objects and further accelerating convergence and improving regression accuracy. Additionally, a coordinate attention (CA) mechanism is introduced at the end of the backbone to focus on orientation perception and positional information. We conducted a series of experiments, comparing our method with networks like GhostNet, PP-LCNet, MobileNetV3, and MobileNetV3s, and performed detailed ablation studies. The experimental results on the Mar20 public dataset indicate that, compared to the original YOLOv5n network, our lightweight network has only about one-fifth of its parameter count, with only a slight decrease of 2.7% in mAP@0.5. At the same time, compared with other lightweight networks of the same magnitude, our network achieves an effective balance between detection accuracy and resource consumption such as memory and computing power, providing a novel solution for the implementation and hardware deployment of lightweight remote sensing object detection networks.
... Moreover, the performances for average accuracy and frames per second of YOLOv4 are increased compared to YOLOv3. YOLOv4-tiny [25,26] is a compressed version of YOLOv4. Based on YOLOv4, it is proposed to simplify the network structure, reduce parameters and enable development on embedded devices, and YOLOv4-tiny based model performs faster training and faster detection by comparison with YOLOv4. ...
