基于BBO-MLP和纹理特征的图像分类算法

发布时间：2018-06-05 13:29

本文选题：纹理特征 + 特征矩　；参考：《光电子·激光》2016年11期

【摘要】：为了提高图像分类的准确率,解决多层感知器(MLP)收敛速度缓慢等问题,提出了一种基于生物地理学优化-MLP(BBO-MLP)和纹理特征的图像分类算法。首先,从图像库中选取3类不同的图片,对图像分类算法运行环境进行建模;其次,选取角二阶矩(UNI)、熵(CON)、惯性矩(ENT)和相关性(CDR)4个纹理参数构建一个四维特征矩,根据用户提供的类别号和图像的纹理特征向量生成训练样本文件;然后,将提取的数据作为MLP的输入数据,为MLP定义一个评估栖息地的误差适应度函数并对适应度函数进行全局优化,利用BBO算法训练MLP,得到分类模型;最后,利用训练好的MLP对图像进行分类,并引入二次反馈机制进一步提高算法性能。实验结果表明,与PSO、GA、ACO、ES和PBIL等优化算法相比,本文的BBO-MLP算法具有较高的分类正确率。
[Abstract]:In order to improve the accuracy of image classification and solve the problem of slow convergence of multilayer perceptron, an image classification algorithm based on biogeographic optimization and texture features is proposed. Firstly, three different kinds of images are selected from the image database to model the running environment of the image classification algorithm. Secondly, four texture parameters are selected to construct a four-dimensional feature moment, which are angular second-order moment UNI, entropy conformal, moment of inertia (moment of inertia) and correlation CDR (). The training sample file is generated according to the class number provided by the user and the texture feature vector of the image. The extracted data is then used as the input data of the MLP. This paper defines an error fitness function for evaluating habitat for MLP and optimizes the fitness function globally. The MLP is trained by BBO algorithm to obtain the classification model. Finally, the trained MLP is used to classify images. And the quadratic feedback mechanism is introduced to further improve the performance of the algorithm. The experimental results show that the BBO-MLP algorithm has a higher classification accuracy than the PSO-GAA ACO-ES and PBIL optimization algorithms.
【作者单位】：西安翻译学院工程技术学院;西安电子科技大学空间科学与技术学院ICIE研究所;
【基金】：国家自然科学基金项目(61105066) 中央高校基本科研业务费专项资金(JB141305)资助项目
【分类号】：TP391.41

【相似文献】