摘要弱小目标检测研究无论是在军用领域还是在民用领域当中有着重要的地位和作用。但是,由于弱小目标所占像素点少,而且背景经常伴随着复杂的干扰,其检测手段一直效率较低。本文基于这个研究背景,提出了基于字典学习的弱小目标综合检测方法。本文主要的内容如下: 首先,针对弱小目标检测的研究背景和现状进行了调查和分析,列举出了现有的检测方法和正在进行的一些理论研究并简要的介绍了字典学习的一些基本知识。提出了针对弱小目标图像的形态学背景抑制方法。运用形态学滤波的基本知识和数学公式,实现了对背景的初步去噪。然后利用了基于块分类的稀疏编码模型和字典学习方法,通过大量的数学公式推算了字典块构造和稀疏编码的过程,并推导出字典学习算法,重构出所需背景。提出两种背景的融合方法,并利用灰度直方图双峰之间低谷设定阈值对经过背景抑制的图像进行二值化,筛选出候选目标。 本文还构造了验证所提出的方法所需要的数据集。展示了数据经过上述检测算法分步测试的结果和最终弱小目标的候选结果。利用了对基于字典学习的弱小目标检测方法的定量评估曲线和图表验证了算法的性能,提出了文章的一些不足之处和今后发展的方向。实验结果证明了该方法的合理和有效性。42401 毕业论文关键词 弱小目标检测 图像去噪 字典学习 稀疏编码
Title Research on small target detection using dictionary learning
Abstract Small target detection has an important position and role in the military field and the civilian field. However, due to the small target pixels, and the background is often accompanied by complex interference, most of detection methods now has been less efficient. Based on this background, this paper proposes a method of comprehensive detection of small and weak targets based on dictionary learning. The main contents of this paper are as follows: At the beginning, this paper investigates and analysis the background and current situation of the small target detection, lists the existing detection methods、some theoretical research and briefly introduces the basic knowledge of dictionary learning. A morphological background suppression method for small target images is proposed. The basic knowledge and mathematical formula of morphological filter are used to realize the initial denoising of the image’s background. Then by using the block classification of sparse coding model and the dictionary based on learning methods, through a large number of mathematical formula to calculate the dictionary block structure and sparse coding and dictionary learning algorithm is derived, based on that ,this paper reconstructs the required background. Two background fusion methods are proposed, and uses the gray level histogram to get the threshold. After that, this paper proposes a threshold adjusting method to screen the candidate target. This paper also constructs the data set required to validate the proposed method. Showing the results of the data after the test algorithm step by step and the final result of the final weak target. The performance of the algorithm is verified by the quantitative evaluation curve and the chart of the small target detection method based on dictionary learning. The experimental results show that the method is reasonable and effective.
Keywords Small target detection;Image denoising;Dictionary learning;Sparse coding