利用数据引力进行图像分类
发布时间：2018-07-26 15:55
【摘要】：提出了一种建立在数据颗粒引力基础上的图像分类方法,此方法使用的数据颗粒质量是图像的特征(如图像的分形维)。对于每一类别训练数据颗粒集,图像特征采用训练数据颗粒集中几幅图像特征的均值m_i,用训练数据颗粒集的像幅数w_i作为它的权重,那么第i个训练数据颗粒集的质量为w_im_i,而待检验数据颗粒是原子数据颗粒,其质量为1。假定待检验一幅图像数据颗粒j的特征为t_(m_j),那么第i个训练数据颗粒集与待检验一幅图像的数据颗粒j之间的距离为|m_i-t_(m_j)|。假定有3种不同类别的图像,从各类别中取出一部分图像组成3类训练数据颗粒集,求得每类数据颗粒集特征的均值和一幅待检验数据颗粒的特征值,按公式计算每类数据颗粒集对待检验数据颗粒的引力,3个引力中具有最大引力的类别即为待检验数据颗粒的类别。实验结果表明,基于数据引力的图像分类方法具有一定的优势。
[Abstract]:An image classification method based on the gravity of data particles is proposed. The quality of the data particles used in this method is the feature of the image (such as the fractal dimension of the image). For each type of training data particle set, the image feature is based on the average value of several image features in the training data particle set, and the image amplitude of the training data particle set is used as its weight. Then the mass of the first training data particle set is WSTIM _ I _ s, while the data particle to be tested is atomic data particle, and its mass is 1. 5%. Assuming that the character of the image data particle j is t _ (maugj), the distance between the first training data particle set and the data particle j of an image to be examined is mStuff i-tj. Assuming that there are three different classes of images, a portion of the images are taken from each category to form a set of three kinds of training data particles, and the mean value of the feature of each class of data particle set and the eigenvalue of a piece of data particle to be tested are obtained. According to the formula, the gravitation of each kind of data particle set towards test data particle is calculated. The class of three kinds of gravity with maximum gravity is the class of data particle to be tested. Experimental results show that the image classification method based on data gravity has some advantages.
【作者单位】： 武汉大学深圳研究院;武汉大学遥感信息工程学院;武汉大学电子信息学院;
【基金】：深圳市基础科研项目(JCYJ20150422150029095)~~
【分类号】：O314;TP391.41
，
本文编号：2146519
[Abstract]:An image classification method based on the gravity of data particles is proposed. The quality of the data particles used in this method is the feature of the image (such as the fractal dimension of the image). For each type of training data particle set, the image feature is based on the average value of several image features in the training data particle set, and the image amplitude of the training data particle set is used as its weight. Then the mass of the first training data particle set is WSTIM _ I _ s, while the data particle to be tested is atomic data particle, and its mass is 1. 5%. Assuming that the character of the image data particle j is t _ (maugj), the distance between the first training data particle set and the data particle j of an image to be examined is mStuff i-tj. Assuming that there are three different classes of images, a portion of the images are taken from each category to form a set of three kinds of training data particles, and the mean value of the feature of each class of data particle set and the eigenvalue of a piece of data particle to be tested are obtained. According to the formula, the gravitation of each kind of data particle set towards test data particle is calculated. The class of three kinds of gravity with maximum gravity is the class of data particle to be tested. Experimental results show that the image classification method based on data gravity has some advantages.
【作者单位】： 武汉大学深圳研究院;武汉大学遥感信息工程学院;武汉大学电子信息学院;
【基金】：深圳市基础科研项目(JCYJ20150422150029095)~~
【分类号】：O314;TP391.41
，
本文编号：2146519
本文链接：https://www.wllwen.com/kejilunwen/lxlw/2146519.html
