基于高阶神经网络的机构零件形状识别

提出了一种机械零件在线自动检测的形状识别系统，该系统以零件各边的长度、角度、圆心角和与邻边夹角4特征来表示零件的形状，并采用高阶神经网络（HONN），实现了零件的平移、惊讶和旋转不变怀识别，由于特征参数本身的平移、尺度不变性和循环移位性，可采用二阶HONN 造系统，解决了高阶神经网络中连接的组合爆炸问题，仿真验证了该系统对机械零件的不变性识别能力以及不同参数系统的性能和实用价值。

Shape Recognition of Machined Parts Based on Higher Order Neural Network

An automation shape recognition system of machined parts on line was presented. This system describes the machined parts by its four geometric features: its length, curvature, central angle and relative orientation. Besides, a higher order neural network (HONN) is used to achieve translation, scaling and rotation invariant (TSRI) pattern recognition effectively. Due to the translation, scale invariant and rotation variant of the primitives of machined parts, an HONN structure combining two order and first order is proposed to maintain the capability of TSRI pattern recognition of complete HONN, and also solve the problems of explosively increased weighs in complete HONN. The simulation result proves this system achieves the TSRI recognition of machined parts. It also shows its different performance and application value under different parameters.