【摘要】 人工神经网络是特征识别的有力工具。在研究对驻极体麦克图像识别方法的基础上,本文提出了一种用改进的BP神经网络进行图像特征的识别和学习算法,并给出了动量系数和学习率的调整方法。对比传统方法测定的结果,使用改进的BP神经网络在识别不规则特征时:减少了输入信息冗余,网络结构相对简单;神经网络输出的各项指标明显提高了精度,对麦克图像特征的平均识别正确率达到92.7%;识别速度也满足在线实时检测的要求。理论分析和实验均表明该算法能实时有效地检测出驻极体麦克图像的特性。本文为研究图像不规则特征的识别提供了一种新的方法。更多还原

【Abstract】 Artificial neural network is a powerful tool for feature recognition. On the basis of studying recognition method for image of electret condenser microphone, this paper presents an improved BP learning algorithm for recognizing the image features using neural network, and the method for adjusting momentum vector and learning rate is discussed. Compared with the inspecting result of traditional methods, such as those based on Multilayer Perceptron (MLP) or Radial-Basis Function neural network (RBF), the ameliorated BP neural network recognition method presented in this paper has a relatively simple network structure and less input information redundancy for irregular feature recognition, the accuracy of the neural network output indexes is improved obviously. The average correct rate of microphone image recognition reaches to 92.7% and the recognition speed meets the requirement of online real-time detection. Theoretical evaluation and simulation experiments show that the improved BP neural network can effectively detect the image feature of electret condenser microphone. The paper provides a new approach for studying irregular image feature recognition.更多还原