基于神经网络学习控制的图像挖掘算法

针对传统神经网络算法进行图像分类识别时收敛速度慢,学习过程中可能出现震荡甚至收敛于局部极小值的情况,提出了一种小波变换融合神经网络的图像分类识别方法.利用高斯小波基函数取代神经网络隐含层中的隐节点函数,采用小波神经网络参数初始化方法和改进的模拟退火算法自适应调整学习过程中的网络权值参数,从而解决了神经网络的学习效率低等情况.结果表明,本文方法对5类动物图片的正确分类识别率为84.0%,较传统神经网络和稀疏表示的正确分类识别率提高了4.2%和6.1%.

Image mining algorithm based on neural network learning and control

Aiming at the situation that the convergence speed of traditional neural network algorithm is slow, the oscillation may appear in the learning process, and even the algorithm may converge to the local minimum value, an image classification recognition method based on wavelet transform fusion neural network was proposed. The Gaussian wavelet basis function was used to replace the hidden node function in the hidden layer of neural network. The network weight parameters in the learning process were adaptively adjusted with the wavelet neural network parameter initialization method and the improved simulated annealing algorithm. Therefore, such problem as the low learning efficiency of neural network can be solved. The results show that the correct classification and recognition rate of the proposed algorithm for five kinds of animal images is 84.0%, which increases by 4.2% and 6.1% than that of traditional neural network and sparse representation, respectively.