一种混合神经网络在颗粒图像边缘检测中的应用

A Hybrid Neural Network and Its Application on Edge Detection of Globular Material

本文提出了一种应用混合神经网络进行颗粒图像检测的方法。混合神经网络由用于对边缘候选图像的二值输入模式进行聚类特征提取的自组织竞争子网络（ＡＳＣＳＮＮ）和用于获取颗粒图像边缘矢量信息的ＢＰ子网络（ＢＰＳＮＮ）组成，边缘候选图像是通过采用基于灰度极小值算法提取的边缘候选象素获得。神经网络以边缘候选图像中的边缘候选象素及其邻域象素的二值模式作为训练样本。对经过噪声污染的图像进行实验表明，该方法获得的边缘图像封闭性好、边缘描述真实，抗干扰能力较强，适用于颗粒图像的边缘检测。

This paper proposes an edge detection method of globular material using hybrid neural network. The proposed network consists of self-organizing competitive netal network used for feature extraction from the binary input pattern of the dge candidates image and the BP network network used for deducing edge vectro and a logical judgement algorithm is used for getting edge candidate image' Edge pixel pixel scandidates and its neighbor pixels constitUte the binary samples of the hybrid neural network. The expermments on image corrupted with Gaussian noise show that the image segmented by this method has good edge closedness and true edge and are suitable for the segmentation of the cumulate particle image.