基于轻量化YOLOv8网络的多图像内河航标检测  

Multi-image inland navigation mark detection based on lightweight YOLOv8 network

作　　者：王晓健 黄细霞 WANG Xiaojian;HUANG Xixia(Institute of Logistics Science and Engineering,Shanghai Maritime University,Shanghai 201306,China)

机构地区：[1]上海海事大学物流科学与工程研究院,上海201306

出　　处：《上海海事大学学报》2025年第1期8-16,共9页Journal of Shanghai Maritime University

基　　金：国家自然科学基金(52001197)。

摘　　要：针对传统单图像的检测方法在内河航标检测中准确性与时效性差的问题,提出一种基于多图像特征融合的内河航标检测方法。在多图像快速鲁棒特征(speeded up robust feature,SURF)匹配搜索算法中加入一个确保特征样本饱和化的循环判断机制,防止关键特征缺失。采用曼哈顿距离完成图像特征的相似性度量,充分利用多图像间的融合信息提高航标清晰度。采用硬件感知网络架构与NetAdapt算法结合的MobileNetV3网络对YOLOv8的主干网络进行轻量化改进,降低模型空间复杂度。实验结果表明,对内河航标检测的平均精度均值(mean average precision,mAP)达到了97.02%,模型大小仅4.75 MB,推理耗时仅10.9 ms。改进方法能够快速、准确地检测定位到图像中的航标,能更好地适应特殊场景,为解决实际问题提供支持。In order to solve the problem of poor accuracy and timeliness of traditional single-image detection methods in the inland navigation mark detection,an inland navigation mark detection method based on multi-image feature fusion is proposed.In the multi-image speeded up robust feature(SURF)matching search algorithm,a cyclic judgment mechanism is introduced to ensure the saturation of feature samples,preventing the loss of key features.The Manhattan distance is employed for measuring the similarity of image features,and the fusion information among multiple images is used to enhance the clarity of the navigation mark.The MobileNetV3 network combined with the hardware-aware network architecture and NetAdapt algorithm is used to achieve lightweight improvements in the backbone network of YOLOv8,and reduce the spatial complexity of the model.The experimental results show that the mean average precision(mAP)of inland navigation mark detection reaches 97.02%,the model size is only 4.75 MB,and the inference time is only 10.9 ms.The improved method can quickly and accurately detect the navigation marks in an image,which can better adapt to specific scenes and provide support for solving practical problems.

关 键 词：多图像 轻量化 快速鲁棒特征(SURF) YOLOv8 内河航标 

分 类 号：U644.4[交通运输工程—船舶及航道工程] TP751[交通运输工程—船舶与海洋工程]