论文部分内容阅读
卷积神经网络在图像识别时,会出现参数调节速度慢,需要迭代次数多,以及小样本数据分类效果较差的问题。为解决上述问题,我们提出了局部调节卷积神经网络的方法。其首先通过调节需求的大小把参数对应的神经元分为关键部分和非关键部分,其次采用动态学习率和局部关键点修正,实现参数快速调节。最终通过在mnist、ORL、CIFAR-10和LFW上的实验结果表明,局部调节卷积神经网络的参数调节比其他方法更快,在图像识别中达到相同识别精度所需要的时间更少,而且整体的识别率上也比其他方法高。
Convolution neural network in image recognition, there will be slow parameter adjustment, the need for more iterations, and small sample data classification is less effective. In order to solve the above problems, we propose a method to locally adjust the convolution neural network. Firstly, the neurons corresponding to the parameters are divided into the key part and the non-key part by adjusting the size of the demand. Secondly, the dynamic learning rate and the local key point correction are adopted to realize the rapid adjustment of the parameters. Finally, the experimental results on mnist, ORL, CIFAR-10 and LFW show that the parameters of local tuning convolution neural network are much faster than other methods and less time is needed to achieve the same recognition accuracy in image recognition. Moreover, The recognition rate is also higher than other methods.