图像识别神经网络处理系统

目的　研制用于二值图像实时识别的神经网络处理器； 方法　采用数字电路实现的图像识别神经网络系统有３ 大模块：边缘提取模块以二值图像序列作为输入，用平面建筑物原理和８ 邻域原理提取图像的边缘特征；特征提取模块在边缘特征的基础上提取具有不变性的角特性和组块特征；模式分类模块采用４ 层特征映射神经网络实现图像模式识别； 结果　通过对图像信号发生器送出的二值飞机图像识别说明，训练样本数越多系统识别率越高； 结论　该系统能够快速、正确地实时识别二值图像序列；

Image Recognition Neural Network Processing Systems

Aim To develop the neural network processor used in real time binary image recognition. Methods The neural network system for image recognition was implemented by digital circuit, the system was divided into three modules: the edge extraction module made use of plane structure principle and 8 adjacent field principle to extract edge feature from the binary image sequence, the feature extraction module extracted invariable angle features and area features from the edge of the image, the pattern classification module was a four layers feature mapping neural network used to realize image pattern recognition. Results Binary plane images from the image gernerator were sent to this system to be recognized, the more training images the higher recognition rate could be attained. Conclusion This image recognition neural network system can immediately rightly recognize binary image sequence in real time.