【摘要】：卷積神經(jīng)網(wǎng)絡(luò)被廣泛應(yīng)用在圖像處理領(lǐng)域,不同算法對(duì)網(wǎng)絡(luò)識(shí)別率有較大的影響�；诖�,引入小波分解理論,通過(guò)BP傳播算法以及空間向量理論證明得到,相互獨(dú)立的特征更能表達(dá)原圖像的信息。通過(guò)小波分解去除卷積核之間的相關(guān)性,用較少的卷積核提取圖像更獨(dú)立、全面的特征,以提高網(wǎng)絡(luò)的識(shí)別性能。在MNIST、CIFAR-10和CK標(biāo)準(zhǔn)數(shù)據(jù)庫(kù)上進(jìn)行分類識(shí)別實(shí)驗(yàn),實(shí)驗(yàn)結(jié)果表明,此算法能在不同核函數(shù)尺寸的條件下取得較高識(shí)別率,且達(dá)到與傳統(tǒng)算法相同識(shí)別率的前提下,所需的訓(xùn)練迭代次數(shù)更少,訓(xùn)練時(shí)間更短。最后,將該算法應(yīng)用到絕緣子故障識(shí)別中,并取得了良好的效果。
[Abstract]:Convolution neural network is widely used in image processing. Different algorithms have great influence on the recognition rate of the network. Based on this, the wavelet decomposition theory is introduced, and it is proved by BP propagation algorithm and spatial vector theory that the independent features can better express the information of the original image. The correlation between convolution kernels is removed by wavelet decomposition, and the image is extracted with fewer convolution kernels to extract more independent and comprehensive features in order to improve the recognition performance of the network. The classification and recognition experiments are carried out on MNIST,CIFAR-10 and CK standard databases. The experimental results show that the algorithm can achieve high recognition rate under the condition of different kernel function sizes, and achieve the same recognition rate as the traditional algorithm. The number of training iterations is less and the training time is shorter. Finally, the algorithm is applied to insulator fault identification, and good results are obtained.
【作者單位】： 華北電力大學(xué)電氣與電子工程學(xué)院;
【分類號(hào)】：TM216

【相似文獻(xiàn)】

相關(guān)期刊論文 前10條

1 謝慧敏;;淺析電機(jī)故障識(shí)別及診斷[J];電子制作;2014年05期

2 王雪峰,鄔建華,馮英浚,王建元;運(yùn)用樣本更新的實(shí)時(shí)神經(jīng)網(wǎng)絡(luò)進(jìn)行短期電力負(fù)荷預(yù)測(cè)[J];系統(tǒng)工程理論與實(shí)踐;2003年04期

3 文漢云;;硫化氫燃燒的神經(jīng)網(wǎng)絡(luò)PID控制及其仿真[J];自動(dòng)化與儀器儀表;2006年01期

4 張U，

本文編號(hào)：2488805