递归神经网络记忆存储器的两个应用
发布时间：2019-06-06 06:20
【摘要】：随着生物神经网络系统的研究和智能化信息处理技术的发展,人工神经网络成为科学研究者们探索的热点.其中递归神经网络被广泛地应用在图像处理、联想记忆、平行计算、信号处理、模式识别等领域,因此关于递归神经网络的实际应用研究很有现实意义. 本文提出了一种基于Cohen-Grossberg神经网络的图像解密消噪的方法,彩色图像在标准的RGB三色空间中表达,图像加密方式采用Arnold变换.Cohen-Grossberg神经网络将加密后的图像数字矩阵作为网络的平衡点进行存储,以实现解密前消除噪声的功能.消除噪声的加密图像数字矩阵通过执行正确的Arnold变换迭代次数实现解密.仿真实例验证了提出方法的有效性,实现了消除传输噪声的功能. 另外本文还用训练单层前向神经网络的方法来实现Hopfield神经网络记忆存储的功能,并将此Hopfield神经网络应用在产品质量分类中.仿真实例表明Hopfield递归神经网络的分类效果比较理想,可以为产品优化、市场决策提供有效信息.
[Abstract]:With the research of biological neural network system and the development of intelligent information processing technology, artificial neural network has become the focus of scientific researchers. Recurrent neural network is widely used in image processing, associative memory, parallel computing, signal processing, pattern recognition and other fields, so the practical application of recurrent neural network is of great practical significance. In this paper, a method of image decryption and denoising based on Cohen-Grossberg neural network is proposed. The color image is expressed in the standard RGB trichromatic space. Arnold transform is used in image encryption. Cohen-Grossberg neural network stores the encrypted image digital matrix as the balance point of the network in order to eliminate noise before decryption. The encrypted image digital matrix which eliminates noise is decrypted by performing the correct number of iterations of Arnold transform. The simulation example verifies the effectiveness of the proposed method and realizes the function of eliminating transmission noise. In addition, this paper also uses the method of training single-layer forward neural network to realize the memory storage function of Hopfield neural network, and applies the Hopfield neural network to product quality classification. The simulation example shows that the classification effect of Hopfield recurrent neural network is ideal, and it can provide effective information for product optimization and market decision.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP333
本文编号：2494134
[Abstract]:With the research of biological neural network system and the development of intelligent information processing technology, artificial neural network has become the focus of scientific researchers. Recurrent neural network is widely used in image processing, associative memory, parallel computing, signal processing, pattern recognition and other fields, so the practical application of recurrent neural network is of great practical significance. In this paper, a method of image decryption and denoising based on Cohen-Grossberg neural network is proposed. The color image is expressed in the standard RGB trichromatic space. Arnold transform is used in image encryption. Cohen-Grossberg neural network stores the encrypted image digital matrix as the balance point of the network in order to eliminate noise before decryption. The encrypted image digital matrix which eliminates noise is decrypted by performing the correct number of iterations of Arnold transform. The simulation example verifies the effectiveness of the proposed method and realizes the function of eliminating transmission noise. In addition, this paper also uses the method of training single-layer forward neural network to realize the memory storage function of Hopfield neural network, and applies the Hopfield neural network to product quality classification. The simulation example shows that the classification effect of Hopfield recurrent neural network is ideal, and it can provide effective information for product optimization and market decision.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP333
【参考文献】
相关期刊论文 前1条
1 廖晓昕,肖冬梅;具有变时滞的Hopfield型神经网络的全局指数稳定性[J];电子学报;2000年04期
，本文编号：2494134
本文链接：https://www.wllwen.com/kejilunwen/jisuanjikexuelunwen/2494134.html
