面向特定目标的特征挖掘与提取方法研究与实现

发布时间：2018-06-08 04:03

本文选题：目标识别 + 特征提取　；参考：《北方工业大学》2014年硕士论文

【摘要】：随着遥感技术的迅速发展,如何从遥感图像中有效提取目标特征,并对其目标特征进行挖掘成为遥感图像处理的一个重要课题。由于道路的重要性,道路目标的特征提取和特征挖掘在社会发展的各个方面具有重大的意义。本文主要的研究工作如下：图像在获取或传输过程中不可避免地存在一定程度的噪声干扰,噪声恶化了图像质量,给图像分割、分析判断等工作带来了困难。为此,提出了一种基于希尔伯特黄变换的图像去噪方法,采用二维经验模式分解对遥感图像进行多尺度分解,分解出各个本征模式函数,对本征模式函数进行二维希尔伯特变换,保留并加强图像细节信息、削弱噪声,最终实现图像的去噪,并利用客观评价指标评价去噪效果。在图像特征提取阶段,提取了图像的颜色、纹理和形状特征。在边缘特征提取时,本文提出了用Canny算子与小波变换相结合的边缘提取算法。该方法结合了Canny算子提取边缘的优点和小波变换在检测图像突变点中的优点,避免了Canny算子提取边缘道路边缘不连续现象和小波变换提取边缘时出现的虚假的边缘信息,有效地对图像进行边缘检测。利用BP神经网络对图像的特征进行训练,由于BP神经网络有其自己的缺陷,当给一个训练好的BP神经网络提供新的学习记忆模式时,将使已有的连接权值被打乱,导致已记忆的学习模式的信息消失,使得训练效果一般。在对BP神经网络优化时,主要的方法就是添加动量因子,或者改变其激励函数,在本文中,利用BP神经网络与模拟退火法相结合,提高了神经网络学习效果,缩短了陷入局部极小值的时间。
[Abstract]:With the rapid development of remote sensing technology, how to extract target features effectively from remote sensing images and mine them has become an important task in remote sensing image processing. Because of the importance of road, feature extraction and feature mining of road goals are of great significance in all aspects of social development. The main research work of this paper is as follows: there is inevitably a certain degree of noise interference in the process of image acquisition or transmission, which results in the deterioration of image quality and brings difficulties to image segmentation, analysis and judgment. In this paper, a new image denoising method based on Hilbert-Huang transform is proposed, in which two dimensional empirical mode decomposition is used to decompose the remote sensing image at multiple scales, and the intrinsic mode functions are decomposed. The intrinsic mode function is transformed into two dimensional Hilbert transform, which preserves and strengthens the detail information of the image, weakens the noise, finally realizes the denoising of the image, and evaluates the denoising effect by using the objective evaluation index. The color, texture and shape features of the image are extracted. In this paper, an edge extraction algorithm based on Canny operator and wavelet transform is proposed. This method combines the advantages of Canny operator in edge detection and wavelet transform in detecting image mutation points, and avoids the discontinuity of edge road edge by Canny operator and false edge information when wavelet transform is used to extract edge. The BP neural network is used to train the image features. Because BP neural network has its own defects, when a trained BP neural network is provided with a new learning and memory mode, It will cause the existing connection weights to be disturbed, resulting in the loss of the information of the learning mode, which makes the training effect less effective. In the optimization of BP neural network, the main method is to add momentum factor or change its excitation function. In this paper, BP neural network is combined with simulated annealing method to improve the learning effect of neural network. Shortens the time to get into a local minimum.
【学位授予单位】：北方工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】