| 摘要: |
| 海天线检测在海洋工程安防活动中具有重要的意义，真实海洋环境中的海天线检测易受云朵、海浪、光照变化、目标遮挡物、边界模糊等外界干扰。为了实现对真实海洋环境中海天线的检测，本研究提出一种基于L0梯度平滑和图像分块聚类的海天线检测算法。首先，对图像进行L0梯度平滑滤波，以增强海天线边缘，弱化非海天线因素干扰；接着，将图像沿着竖直方向分割成若干等宽图像块，以降低整体环境干扰，加强局部海天线检测效果；然后，通过Canny算子和霍夫变换提取每个分割图像块中的直线段；最后，采取K-means聚类算法提取每个图像块中的海天线段，拟合生成完整海天线。实验结果表明，在真实的海天线数据集中，本研究方法获取的矩形框重叠率平均精度为93.22%，角度差平均精度为7.66%，均高于文中选取的近年典型对比算法。满足实际海天线检测抗干扰强、准确率高、适应性广等要求。 |
| 关键词: 海洋水文学 海天线检测 L0梯度平滑滤波 图像分块 K-means线段聚类 |
| DOI：10.3969/J.ISSN.2095-4972.20220726001 |
|
| 基金项目:自然资源部海洋环境探测技术与应用重点实验室自主设立课题(MESTA-2021-C004)；海洋科学技术局长基金(180214)；南方海洋科学与工程广东省实验室（珠海）项目(SML2021SP205） |
|
| Sea-sky-line detection base on L0 gradient smoothing and image segmentation clusters |
| ZHENG Bing,DONG Chao,HU Haiju,CHEN Yankun,LIU Wei |
| (South China Sea Marine Survey and Technology Center, SOA, Guangzhou 510300, China;Key Laboratory of Marine Environmental Survey Technology and Application, MNR, Guangzhou 510300, China;Southern Marine Science Engineering Guangdong Laboratory, Zhuhai 440402, China;Institute of Surveying and Mapping, Department of Natural Resources of Guangdong Province, Guangzhou 510500, China) |
| Abstract: |
| Sea-sky-line detection is of great significance in the security of marine engineering activities. Sea-sky-line detection is susceptible to external interference in real marine environment such as clouds, waves, illumination variation, target occlusions and boundary blur, etc. A sea-sky-line detection algorithm based on L0 gradient smoothing and image segmentation & clusters is proposed. Firstly, the image is filtered by L0 gradient smoothing to enhance sea-sky-lines edge and weaken the interference of non-sea-sky-lines. Then, the image is segmented into several equal-width image blocks along vertical direction to reduce environmental interference and strengthen the detection effect of local sea-sky-lines. The straight line segments in each segmented image block are extracted by Canny operator and Hough transform. Finally, K-means clustering algorithm is adopted to extract the sea-sky-line in each image block, thus fitting to generate the final sea-sky-line. Experimental results show that average accuracy of bounding box overlap rate is 93.22% and the average accuracy of angle difference ration is 7.66% in the real sea-sky-line dataset, both of which are higher in comparison with typical algorithms selected in recent years. Result meets the requirements of real sea-sky-line detection with characters of strong anti-interference, high accuracy and wide adaptability. |
| Key words: marine hydrology sea-sky-line detection L0 gradient smoothing image segmentation K-means linear cluster |
