摘要： 介绍了一种网络图像监管技术方案,在此基础上着重研究了基于Kohonen神经网络和BP网络的色情图像特征识别技术,而且通过引入分裂算法对Kohonen网赋初值、对BP网络采用随机样本输入以及动态调整学习率等方法,对识别技术进行了改进,使整个网络图像监管技术取得了较好的识别效果。

关键词: Kohonen神经网络, BP网络, SOM, 分裂算法

Abstract: This paper introduces a new method of image supervision , and focuses a study on the porn images’ character identification technology based Kohonen neutral network and BP network. To get the expectant identification effect, it improves Kohonen neutral network with setting initial value to it by split algorithm, and BP network with random sample inputting and adjusting learning rate automatically .

中图分类号: 

• TP391.41