论文部分内容阅读
极端学习机因其学习速度快、泛化性能强等优点,在当今模式识别领域中已经成为了主流的研究方向;但是,由于该算法稳定性差,往往易受数据集中噪声的干扰,在实际应用中导致得到的分类效果不是很显著;因此,为了提高极端学习机分类的准确性,针对数据集样本中带有噪声和离群点问题,提出了一种基于角度优化的鲁棒极端学习机算法;该方法利用鲁棒激活函数角度优化的原则,首先降低了离群点对分类算法的影响,从而保持数据样本的全局结构信息,达到更好的去噪效果;其次,有效的避免隐层节点输出矩阵求解不准的问题,进一步增强极端学习机的泛化性能;通过应用在普遍图像数据库上的实验结果表明,这种提出的算法与其他算法相比具有更强的鲁棒性和较高的识别率。
Because of its fast learning speed and extensive generalization performance, extreme learning machine has become the mainstream research direction in the field of pattern recognition. However, due to the poor stability of the algorithm, it is often susceptible to noise in the data set. In practical applications Therefore, in order to improve the accuracy of the extreme learning machine classification, aiming at the problem of noise and outliers in the data set samples, an angle-optimized robust extreme learning algorithm is proposed This method uses the principle of optimization of robust activation function to reduce the influence of outliers on the classification algorithm, so as to keep the global structure information of the data samples and achieve a better denoising effect. Secondly, to effectively avoid hidden nodes Output matrix to solve the problem of uncertainty, to further enhance the extreme learning machine generalization performance; experimental results show that the proposed algorithm on the universal image database, compared with other algorithms has more robust and higher Recognition rate.