基于局部特征的遥感图像目标检测方法研究

发布时间：2018-08-03 08:03

【摘要】：近年来,随着遥感卫星技术的发展,遥感图像在军事和民用领域的应用范围逐渐扩大,针对遥感图像目标检测方法的研究越来越受到各国学者的关注。通常情况下,人们只关心整幅图像或整段视频中的很小一部分,单一地利用全局特征进行目标检测在实际应用中受到了诸多限制,具有一定局限性,而局部特征在对信息描述性能上的优越性,为复杂背景下的遥感目标检测提供了一条有效途径。为了提高遥感信息自动化处理的效率和可靠性、增强军事侦察和情报信息收集的能力,本文围绕遥感图像中团块目标、阵列目标和港口目标的检测和识别问题,将油罐、飞机、舰船作为具体的研究对象,以目标在人类视觉感知系统中所表现出的特性和其局部特征的空间关系作为切入点,系统地研究了上述目标的检测识别关键技术,并在此基础上提出了针对遥感图像中不同类型目标的检测识别方法,提高了遥感目标检测识别系统的效率和适应性,取得了较有价值的研究成果。本文主要研究基于视觉局部特征的遥感图像典型目标检测方法,开展了如下工作:1.针对遥感图像目标边缘特征检测结果准确率低,继而影响图像匹配、目标跟踪等图像处理分析精度的问题。论文先分析了人类视觉生理结构及遥感目标特性,系统研究了遥感图像目标解译过程,对解译的要素、方法及发展趋势进行了总结梳理,为全文的遥感目标检测识别研究打下了坚实的理论基础,然后提出了一种基于视觉感知的高低阈值遥感图像目标边缘特征检测方法。通过对视觉感知系统进行理论研究及分析其特性,挖掘出视觉感知与遥感目标边缘之间的潜在关系。实验验证了本文提出的基于视觉感知的高低阈值法对遥感图像目标边缘特征检测的有效性,并且通过与其它算法对比分析,证明了本方法可有效提高各边缘特征描述子的准确性。2.针对遥感图像的分辨率逐渐增加,图像内容趋于复杂化,目标受阴影干扰导致识别率下降,目标检测的准确些面临巨大困难和挑战这一问题。论文提出了一种基于类圆特征的团块目标检测方法,重点研究了遥感图像油罐目标的检测识别问题。实验结果表明,本文提出的遥感图像油罐目标检测方法相较于其他方法,其检测结果的准确率得到了有效提高,并且通过检测结果能够快速实现油库区域定位,算法适用于不同分辨率的遥感图像。3.针对在实际的遥感图像中,飞机目标受到自身阴影、建筑物遮挡以及地物干扰等因素的影响,使得检测结果中易将飞机目标的阴影轮廓误检测为飞机目标,降低了检测的准确率,导致实际飞机目标的定位和特征提取精确度下降这一问题。论文提出了一种基于不变特征的阵列目标检测识别方法,重点研究了机场背景中飞机目标的检测识别问题。实验结果表明,本文方法实现过程简单、相比于其它检测方法,其对目标背景的干扰影响具有较好的鲁棒性、并且运算量较小、检测结果准确率得到有效提高。4.针对遥感图像中的舷靠舰船目标自身灰度和纹理特征与港口相近,传统检测方法不易将目标与港口进行有效分离,检测准确率低的问题,论文提出了一种基于局部显著特征的遥感图像舷靠舰船检测方法。通过实验验证,本文所提出的检测算法对遥感图像中复杂背景环境下的舷靠舰船目标检测效果较好,并且算法不受舰船停泊位置及阴影的影响,目标识别率较高,鲁棒性较强。
[Abstract]:In recent years, with the development of remote sensing satellite technology, the application range of remote sensing images in military and civil fields has been gradually expanded. The research on remote sensing image target detection methods has attracted more and more attention from scholars in various countries. In general, people are concerned only with a small part of the entire image or the whole segment of the visual frequency, and use the global features in a single way. The target detection has been limited in practical application, which has some limitations, and the advantage of the local feature on the performance of information description provides an effective way for remote sensing target detection under complex background. In order to improve the efficiency and reliability of remote sensing information processing, the military reconnaissance and information collection will be enhanced. In this paper, the ability of the collection, focusing on the detection and identification of the mass target, the array target and the port target in remote sensing images, takes the oil tank, aircraft and ship as the specific research object, and studies the above objectives systematically with the characteristics of the target in the human visual perception system and the spatial relationship of its local characteristics. On the basis of this, the detection and recognition method for different types of objects in remote sensing images is proposed, and the efficiency and adaptability of the remote sensing target detection and recognition system are improved. The research results are obtained. This paper mainly studies the typical target detection method of remote sensing images based on the visual local features. The following work is as follows: 1. in view of the low accuracy of the target edge detection results of remote sensing images, then the image matching, target tracking and other image processing analysis precision problems. The thesis first analyzes the human visual physiological structure and the characteristics of remote sensing targets, systematically studies the interpretation process of remote sensing image, and the factors, methods and development of the interpretation. The trend is summarized and summarized, which lays a solid theoretical foundation for the research of remote sensing target detection and recognition in the full text. Then a method of detection of target edge features of remote sensing image based on visual perception is proposed. The visual perception and remote sensing targets are excavated through the theoretical study and analysis of the characteristics of the visual perception system. The experimental verification of the effectiveness of the high and low threshold method based on visual perception to the detection of edge features of remote sensing images is verified by experiments, and by comparing with other algorithms, it is proved that this method can effectively improve the accuracy of the descriptors of each edge feature descriptor.2. to increase the resolution of remote sensing images. In addition, the image content tends to be complicated, the target is affected by the shadow interference and the recognition rate is reduced, and the accuracy of the target detection is faced with the great difficulty and challenge. A new method of mass target detection based on the characteristic of the circle is proposed, which focuses on the detection and recognition of the target of the remote sensing image oil tank. The experimental results show that this paper is proposed in this paper. Compared with other methods, the accuracy of the detection results has been effectively improved, and the location of the oil storage area can be realized quickly by the detection results. The algorithm is suitable for the remote sensing image.3. of different resolution. The target of the aircraft is shadowed by its own shadow and building occlusion in the actual remote sensing image. As well as the influence of ground object interference, the shadow outline of the aircraft target is mistaken for the aircraft target, which reduces the accuracy of the detection and leads to the reduction of the accuracy of the actual aircraft target location and feature extraction. In this paper, an array target detection and recognition method based on the invariant features is proposed. The detection and recognition of the aircraft target in the airport background is studied. The experimental results show that the implementation of this method is simple. Compared with other detection methods, it has good robustness to the interference effect of the target background, and the computation is small. The accuracy of the detection results can effectively improve the.4. target ship's target in remote sensing images. The gray and texture features of the standard are close to the port. The traditional detection method is not easy to separate the target from the port, and the detection accuracy is low. In this paper, a method of ship detection on the starboard of remote sensing images based on local significant features is proposed. In the environment, the effect of target detection is better, and the algorithm is not affected by the berthing position and shadow of the ship. The target recognition rate is higher and the robustness is stronger.
【学位授予单位】：长春理工大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP751

【相似文献】