基于实时更新神经滤波及其光学实现的复杂零部件特征并行检测算法  

The Complex Parts Feature Parallel Detection Algorithm Based on Real Time Updated Neural Filtering and its Optical Implementation

作　　者：刘向娇[1] 毛红阁[1] 

机构地区：[1]南阳师范学院软件学院,河南南阳473061

出　　处：《组合机床与自动化加工技术》2014年第9期105-108,112,共5页Modular Machine Tool & Automatic Manufacturing Technique

摘　　要：复杂零部件作为制造领域中常用零件,其特征提取质量对其自动化精密加工有着重要影响,特别是其边缘轮廓以及表面纹理。而当前的特征提取算法难以应用于复杂零部件,其检测效果不理想;且检测效率不佳,是一种非并行模式,增加了其检测成本。对此,为了提高复杂零部件特征的检测精度与效率,文章提出了基于实时更新神经滤波及其光学实现的复杂零部件特征检测算法。基于Hopfield神经网络,构造自适应滤波掩码,并嵌入2D卷积运算,设计了实时更新神经滤波技术及其学习算法,利用学习算法更新滤波系数与偏差值;并建立了该技术的光学联合转换结构,完成神经滤波的光学实现,提取复杂零部件特征,并提高其检测效率。仿真实验结果显示:与当前边缘检测机制相比,文章算法良好的检测精度,曲面连接处角点丰富,轮廓连续完整,拓扑凸显后,部件表面形貌清晰可见;且该算法的检测效率更高。Thread is an indispensable coupling member in the field of machinery, and its detection accuracy was determined by the tooth edge type extraction. In order to improve the quality and efficiency of thread tooth edge extraction, the thread image feature extraction algorithm based on real time updated neural filte-ring and its optical implement was proposed. The real time updated neural filtering technology was designed to extract thread tooth edge by basing on Hopfield neural networks and embedding adjustable filter mask and 2D convolution operation;and also the phase-only joint transform correlation was built to implement the dy-namic neural filtering for improving the detection speed. The simulation results showed that：compared with current thread edge extraction mechanism, the detection quality and efficiency of this algorithm was higher.

关 键 词：实时更新 神经滤波 复杂零部件 特征检测 

分 类 号：TH113[机械工程—机械设计及理论] TG65[金属学及工艺—金属切削加工及机床]