Optimized YOLOv10 Model for River Floating Object Detection.

Saved in:

Bibliographic Details
Title:	Optimized YOLOv10 Model for River Floating Object Detection.
Authors:	Luo, Weihua¹ 345850498@qq.com, Chen, Xiao² 631415345@qq.com, Huang, Yuxin² 2562273739@qq.com, Shuai, Dongsheng³ 28250083@qq.com, Xia, Juming³ 360717061@qq.com
Source:	IAENG International Journal of Computer Science. Jun2026, Vol. 53 Issue 6, p2060-2071. 12p.
Subjects:	Wavelet transforms, Object recognition (Computer vision), Feature extraction, Loss functions (Statistics), Environmental management
Abstract:	Intelligent detection of river floating objects is essential for maintaining river ecological balance and ensuring effective environmental management. To overcome the degradation of detection accuracy under complex riverine conditions with varying illumination, surface ripples, and scale variations, this paper proposes YOLOv10 FLOW, an optimized model based on YOLOv10 for floating object detection. First, to mitigate the interference of water ripples and illumination on the extraction of low-frequency features from floating objects, the C2f module in YOLOv10 is enhanced by integrating Waveletbased Convolution (WTConv) to strengthen low-frequency feature extraction. Second, to address the scale inhomogeneity of floating objects, a Content-Guided Attention module with Adaptive Weight Fusion (CGAWFusion) is proposed to replace the original Concat module, thereby enhancing multi-scale feature extraction. Third, Inner-MPDIoU is adopted to replace CIoU as the loss function, which further improves the detection accuracy. Comprehensive experiments were conducted on the dataset, including comparative analysis with different models and ablation studies to evaluate the individual contribution of each proposed component. Compared to the baseline YOLOv10, YOLOv10 FLOW achieves significant improvements of 10.8%, 8.7%, 10.4%, and 7.4% in Precision, Recall, mAP50, and mAP50-95 metrics, respectively. [ABSTRACT FROM AUTHOR]
	Copyright of IAENG International Journal of Computer Science is the property of International Association of Engineers (IAENG) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	Engineering Source

Description
Abstract:	Intelligent detection of river floating objects is essential for maintaining river ecological balance and ensuring effective environmental management. To overcome the degradation of detection accuracy under complex riverine conditions with varying illumination, surface ripples, and scale variations, this paper proposes YOLOv10 FLOW, an optimized model based on YOLOv10 for floating object detection. First, to mitigate the interference of water ripples and illumination on the extraction of low-frequency features from floating objects, the C2f module in YOLOv10 is enhanced by integrating Waveletbased Convolution (WTConv) to strengthen low-frequency feature extraction. Second, to address the scale inhomogeneity of floating objects, a Content-Guided Attention module with Adaptive Weight Fusion (CGAWFusion) is proposed to replace the original Concat module, thereby enhancing multi-scale feature extraction. Third, Inner-MPDIoU is adopted to replace CIoU as the loss function, which further improves the detection accuracy. Comprehensive experiments were conducted on the dataset, including comparative analysis with different models and ablation studies to evaluate the individual contribution of each proposed component. Compared to the baseline YOLOv10, YOLOv10 FLOW achieves significant improvements of 10.8%, 8.7%, 10.4%, and 7.4% in Precision, Recall, mAP50, and mAP50-95 metrics, respectively. [ABSTRACT FROM AUTHOR]
ISSN:	1819656X