基于ISAR图像的舰船目标识别研究

发布时间：2018-08-19 13:46

【摘要】：第二次世界大战以来,雷达在现代战争中扮演着越来越重要的角色。逆合成孔径雷达(ISAR)由于其丰富的信息获取能力,为雷达的自动目标识别(ATR)技术提供了良好的应用场景。岸基ISAR可以对舰船目标进行远距离二维成像,获取目标的形态结构等特征,因此,基于ISAR图像的舰船目标识别技术具有着广阔的发展前景和应用价值。本课题基于ISAR实测数据,对舰船目标自动识别技术进行了研究。研究内容主要包含以下几个方面：首先,针对舰船目标ISAR图像中典型的噪声和结构残缺问题,提出了包含图像分割去噪和形态学方法等的完整的图像预处理流程,获取了目标清晰完整无噪点的图像。其次,研究了舰船目标的形态特征提取方法,并基于实测数据,研究了形态特征对横向定标和样本分布等实际问题的稳健性,为形态特征的实际应用提供了依据。此外,提出了几种舰船目标上层结构曲线的自适应分段算法。均值梯度法在视觉上最合理的位置对曲线进行分段,而均值逼近法和聚类法则通过追求分段后段内曲线分布更加接近来实现曲线的最优分段,而基于拐点的聚类法则结合了这两个角度。对实测数据进行实验的结果证明了自适应分段算法对上层结构编码特征区分度的有效提升。最后,本文将模糊集理论引入,提出了基于上层结构编码的模糊识别技术。该技术通过将目标表示为全体编码论域上的模糊集,扩展了原有上层结构编码的描述能力,并采用最大贴近度原则,对目标进行识别。该技术通过一种两步式的计算方法,利用分段相对均值特征,得到所有编码对目标的隶属度,从而将目标表示为模糊集。随后,利用最大贴近度原则,将目标识别为贴近度最大的类别。最终,在实测数据的识别实验中,比较了三种贴近度定义——海明贴近度、欧几里得贴近度和格贴近度在本应用中的效果,并验证了本文提出的模糊识别技术对现有其他基于上层结构的舰船识别技术的提升。
[Abstract]:Since World War II, radar has played an increasingly important role in modern warfare. Inverse synthetic Aperture Radar (ISAR) provides a good application scenario for automatic target recognition (ATR) technology because of its rich information acquisition ability. Shore-based ISAR can be used to image ship targets at a long distance and obtain the characteristics of target morphology and structure. Therefore, the technology of ship target recognition based on ISAR image has a broad development prospect and application value. Based on the measured data of ISAR, the automatic recognition technology of ship target is studied in this paper. The research mainly includes the following aspects: firstly, aiming at the typical noise and structural defects in ship target ISAR images, a complete image preprocessing process including image segmentation and denoising and morphological methods is proposed. The image of clear and complete target without noise is obtained. Secondly, the extraction method of ship's morphological features is studied. Based on the measured data, the robustness of morphological features to the practical problems such as lateral calibration and sample distribution is studied, which provides a basis for the practical application of morphological features. In addition, several adaptive segmentation algorithms for upper structure curves of ship targets are proposed. The mean gradient method segments the curve in the most reasonable position, while the mean value approximation method and the clustering method achieve the optimal segmentation of the curve through the pursuit of closer distribution of the curve in the segment after the segment. The clustering method based on inflection point combines these two angles. The experimental results on the measured data show that the adaptive segmentation algorithm can effectively improve the classification of the upper structure coding features. Finally, the fuzzy set theory is introduced and a fuzzy recognition technique based on upper structure coding is proposed. By representing the target as a fuzzy set on the whole coding domain, the technique extends the description ability of the original upper structure coding, and adopts the principle of maximum closeness to identify the target. By using a two-step calculation method and the feature of piecewise relative mean, the membership degree of all the codes to the target is obtained, and the target is represented as a fuzzy set. Then, the target is recognized as the category with the largest degree of closeness by using the principle of maximum closeness. Finally, in the recognition experiment of measured data, the effects of three definitions of closeness, Heming closeness, Euclidean closeness and lattice closeness in this application are compared. It also verifies that the proposed fuzzy recognition technology improves the existing ship recognition technology based on upper structure.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TN957.52;E925.6

【相似文献】