基于分形脊波神经网络的遥感图像分类研究

在分形理论和脊波神经网络的基础上，综合利用彩色遥感图像的光谱、纹理和形状特征，提出了一种彩色遥感图像的分类新方法．该方法把彩色图像的蓝、绿、红波段作为3个光谱特征，由分形理论计算的DBC维和多重分形维数作为2个纹理特征，平均不变矩作为1个形状特征，并利用对曲线具有极强方向识别能力的脊波神经网络作为分类器．实验结果表明，提出的彩色遥感图像分类方法具有较高的分类准确率和较强的抗噪音能力．

Study on Remote Sensing Image Classification Based on Fractal Theory and Ridgelet Neural Network

Through comprehensively utilizing color spectrum, texture and shape feature of the remote sensing images,a novel method of image classification is presented based on the fraetal theory and ridgelet neural network. The bule,green and red band spectrums is used as three-band-spectral characteristics, the DBC dimension and muhi-fraetal dimension calculated by fractal theory as two texture characteristics, the average unchanged moment as one shape feature, and the ridgelet neural network with a strong ability to identify the direction of curve is used as classifier in the proposed method. The experimental results indicated that the method used in color image classification has a high accurate rate and a strong antinoise ability.