毕业论文

打赏
当前位置: 毕业论文 > 电子通信 >

基于LARK算法的多光谱夜视目标鲁棒识别

时间:2018-12-03 19:56来源:毕业论文
使用自适应核回归的非训练目标特征提取方法。相对于图片中各点的灰度值、图片中具体的内容变化,该方法对图片的基本数据结构有着更高的敏感度

摘要目标识别是计算机人工智能的一个重要领域,目前常用的是基于训练学习的分类器识别方法。但是,基于训练学习的分类器识别方法有一个明显的缺点,就是使用者需要为分类器搜集非常多的样本,并通过大量的训练和学习过程来确定分类器的相关参数,对模板和训练样本的依赖性很高,并且对一般的目标识别的泛化能力弱。     多光谱夜视目标识别,本文提出了一种使用自适应核回归的非训练目标特征提取方法。相对于图片中各点的灰度值、图片中具体的内容变化,该方法对图片的基本数据结构有着更高的敏感度,因此即使源图像存在严重扭曲也能很好地描述图形原本的特征。 相比其他识别方法,本文提出的方法对复杂夜视场景下人体和车辆等典型目标具有稳健的识别能力。31019
毕业论文关键词  目标识别,局部匹配,相似度图像,LARK 
Title    Robust recognition of multi-spectral night vision target  based on LARK algorithm
Abstract Target identification is an important area of artificial intelligence, the most commonly used identification method is based on the classification of training to learn. However, classification based training and learning identification method has a significant drawback ,it requires a lot of training samples ,and need to determine the parameters of classifier through a lot of training and learning process, has a high dependence of the template and the training samples, and the generalization ability for general infrared target recognition is weak. Multi-spectral night vision  object recognition. Classification based identification method of training and learning has a clear disadvantage, so we propose a non-adaptive kernel regression and no-training target feature extraction method. Use this method can effectively capture the basic data structures, it is not sensitive to the image gray value of each point and specific things in the picture, it is more sensitive to the relative changes in gray values of the picture and the changes in graphics, even if there are serious distortions in the source image ,it also can be well described in graphic original features. Compared with other identification methods, the proposed method has the sound recognition of typical targets vehicles under the complex night vision scene, such as body and car.
Keywords    target recognition, local matching, similarity image, LARK 
目次
1绪论.1
1.1研究背景及意义1
1.2图像目标识别的系统结构和思路2
1.2.1图像目标识别的系统结构2
1.2.2图像目标识别的两种思路3
1.3图像目标识别方法研究现状3
1.3.1基于训练和学习的图像目标识别方法4
1.3.2基于非训练的图像目标识别方法5
1.4本文主要内容7
2基于非训练的图像目标识别8
2.1图像预处理8
2.1.1噪声处理.9
2.1.2图像增强.10
2.2图像特征提取12
2.2.1灰度共生矩阵.12
2.2.2尺度不变特征变换(Scale-invariantfeaturetransform)13
2.2.3基于核回归(KernelRegression)的特征提取15
2.3图像特征相似性匹配17
2.4本章小结18
3基于LARK算法的结构特征提取方法.19
3.1LARK计算原理.19
3.2LARK权值矩阵.21
3.3使用LARK特征提取方法构成的目标识别检测系统.23
3.4本章小结24
4局部相似结构统计匹配(LSSSM)模型26
4.1简单模板集的局部结构分析27
4.1.1构建简单模板集.27
4.1.2简单模板集的降文.28
4.2LSSSM模型原理.32
4.2.1相似性匹配.32
4.2.2生成相似度图像.34
4.3根据相似度图像获取目标信息35
4.4本章小结36
5实验与测试375.1测试步骤37
5.2测试结果41
5.3本章小结43
结论44 基于LARK算法的多光谱夜视目标鲁棒识别:http://www.youerw.com/tongxin/lunwen_27034.html
------分隔线----------------------------
推荐内容