论文部分内容阅读
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。
In this paper, we propose a hybrid neural network method for particle image detection. The hybrid neural network is composed of an ad-hoc competition sub-network (ASCSNN) for clustering feature extraction of binary input patterns of edge candidate images and a BP sub-network (BPSNN) for obtaining granular image edge vector information. The edge candidate images Is obtained by using edge candidate pixels extracted based on the gray minimum algorithm. The neural network uses the binary pattern of edge candidate pixels and their neighborhood pixels in the edge candidate image as training samples. Experiments on noise-contaminated images show that the edge image obtained by this method has good closedness, real edge description and strong anti-interference ability, which is suitable for the edge detection of particle images.