基于全景海天线提取的小目标检测方法研究

发布时间：2018-06-28 07:32

本文选题：全景图像 + 全景设备区提取　；参考：《哈尔滨工程大学》2016年硕士论文

【摘要】：我国具有绵长的海岸线和广阔的海洋领土,海洋维权执法及海上搜救任务繁重,大力开发先进的海域监控设备,研究海上目标检测方法具有重要的意义。在海域监控领域采用全景视觉系统,可以满足海洋环境下大视场、全范围、远距离监控的需求,有效减少监控设备数量、降低硬件成本,但该系统目前缺乏成熟的目标检测技术支持。由远及近驶来的远景目标会最先出现在海天线上,在海天线区域检测小目标可为拍摄取证等工作提供更多的反应时间,因此本论文开展基于全景海天线提取的小目标检测算法研究。折反射全景视觉系统成像原理的特殊性,不仅使得常规视觉中的直线海天线在全景图像中呈椭圆形,而且使得全景图像中包含有大量全景设备区成像。鉴于全景海域图像的特殊性和背景的复杂性,本文的基本研究思路为:首先在全景图像中对海天线提取和目标检测有不良影响的全景设备区干扰进行提取和剔除,然后进行全景海天线提取,最后在海天线区域检测小目标。因此论文研究的三大核心内容为:全景设备区提取、全景海天线提取以及海天线区域的小目标检测。首先,进行了全景设备区提取方法研究。根据全景设备区属于人造物体,其纹理特征与海面、天空等自然景物有明显区别的特点,设计了基于分形维数的全景设备区提取算法;根据全景图像中的全景设备区干扰比较明显,更易引起视觉注意的特点,设计了基于视觉显著图的全景设备区提取算法;并通过实验验证了上述算法的有效性。其次,进行了基于全景视觉的海天线提取算法研究。在抑制了全景设备区干扰的基础上,针对全景海天线呈近似圆形的特点,设计了基于改进梯度Hough圆变换的海天线提取算法;针对全景海天线具有明显边缘轮廓特征的特点,提出了基于改进主动轮廓模型的海天线提取算法;针对全景海天线在全景图像中的梯度能量相对较大的特点,提出了一种基于改进Seam Carving的海天线提取算法;最后通过实验验证了这三种算法的有效性,并与现有文献中的算法进行了实验对比分析。最后,进行海天线区域的小目标检测算法研究。在海天线提取的基础上,将小目标视为图像信号的奇异点,设计了一种基于提升小波互能量的海天线区域小目标检测算法;利用暗通道理论对目标的放大作用,提出了一种基于暗通道先验理论的海天线区域小目标检测算法。最后通过大量实验验证了算法的有效性,并与现有文献中的算法进行了实验对比与统计分析,验证了算法的优越性。
[Abstract]:China has a long coastline and vast maritime territory, the task of maritime rights enforcement and maritime search and rescue is heavy. It is of great significance to develop advanced marine monitoring equipment and to study the detection methods of marine targets. The use of panoramic vision system in the area of sea area monitoring can meet the needs of large field of view, wide range and long distance monitoring in marine environment, effectively reduce the number of monitoring equipment and reduce the cost of hardware. But this system lacks the mature target detection technology support at present. Distant and near-term targets will first appear on the sea and sky lines. Detection of small targets in the sea antenna area can provide more reaction time for such work as taking evidence. In this paper, the small target detection algorithm based on panoramic sea antenna extraction is studied. Because of the particularity of the imaging principle of the reflected panoramic vision system, the linear sea antenna in the conventional vision is not only elliptical in the panoramic image, but also contains a large number of panoramic imaging equipment in the panoramic image. In view of the particularity of panoramic sea area image and the complexity of background, the basic research ideas of this paper are as follows: firstly, in panoramic image, the interference of panoramic equipment which has adverse effects on antenna extraction and target detection is extracted and eliminated. Then the panoramic sea antenna is extracted and the small target is detected in the sea antenna area. Therefore, the three core contents of this paper are: panoramic equipment region extraction, panoramic sea antenna extraction and small target detection in sea antenna region. Firstly, the extraction method of panoramic equipment area is studied. According to the fact that panoramic equipment area belongs to artificial object and its texture features are obviously different from those of sea surface and sky, an algorithm of extracting panoramic equipment area based on fractal dimension is designed. According to the obvious interference of panoramic equipment area in panoramic image, which is more easy to attract visual attention, a panoramic equipment area extraction algorithm based on visual salient image is designed, and the validity of the algorithm is verified by experiments. Secondly, the algorithm of sea antenna extraction based on panoramic vision is studied. On the basis of suppressing the interference of panoramic equipment, aiming at the characteristic that the panoramic sea antenna is approximately circular, the algorithm of extracting sea antenna based on improved gradient Hough circle transform is designed, and aiming at the characteristic of obvious edge contour of panoramic sea antenna, A sea antenna extraction algorithm based on improved active contour model is proposed, and a sea antenna extraction algorithm based on improved seam carrying is proposed, which is based on the relatively large gradient energy of panoramic sea antenna in panoramic images. Finally, the validity of the three algorithms is verified by experiments, and compared with the existing algorithms. Finally, the small target detection algorithm in the sea antenna region is studied. On the basis of the sea antenna extraction, the small target is regarded as the singularity of the image signal, and a small target detection algorithm based on lifting wavelet mutual energy is designed, which uses dark channel theory to amplify the target. A small target detection algorithm based on dark channel priori theory is proposed in this paper. Finally, the validity of the algorithm is verified by a large number of experiments, and the superiority of the algorithm is verified by comparison and statistical analysis with the existing algorithms in the literature.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U675.79;TP391.41

【参考文献】