摘要
卷积神经网络被广泛应用在图像处理领域,不同算法对网络识别率有较大的影响。基于此,引入小波分解理论,通过BP传播算法以及空间向量理论证明得到,相互独立的特征更能表达原图像的信息。通过小波分解去除卷积核之间的相关性,用较少的卷积核提取图像更独立、全面的特征,以提高网络的识别性能。在MNIST、CIFAR-10和CK标准数据库上进行分类识别实验,实验结果表明,此算法能在不同核函数尺寸的条件下取得较高识别率,且达到与传统算法相同识别率的前提下,所需的训练迭代次数更少,训练时间更短。最后,将该算法应用到绝缘子故障识别中,并取得了良好的效果。
Convolution neural network has been widely used in image processing field. Meanwhile,different algorithms have different impact on network recognition rate. Based on this,we introduced wavelet decomposition theory and got that independent characteristics can express the original image more clearly,which was proved by BP propagation algorithm and space vector theory. Because wavelet decomposition reduced the correlation between the Kernel and extracted more independent,comprehensive features with less convolution Kernel,thus the network performance was improved. Recognition experiments are conducted on MNIST,CIFAR-10 and CK standard database,the results show that the algorithm proposed in this paper can achieve higher recognition rate under the condition of different Kernel size and can obtain the recognition rate as the traditional algorithm with fewer iteration times and shorter training time.At last,this algorithm was applied in the fault identification of insulators and achieved good results.
出处
《电测与仪表》
北大核心
2017年第21期30-36,共7页
Electrical Measurement & Instrumentation
关键词
卷积神经网络
图像分类
核函数相关性
绝缘子
故障识别
convolution neural network, image classification, the correlation between the Kernel, insulator,fault identification
