基于改进LeNet-5的车牌识别算法

传统的车牌识别算法包括模板匹配、特征统计等方法,但是这些算法依赖于人工提取图像特征,识别准确率低。卷积神经网络LeNet-5算法能够自动提取车牌图像的特征,提高车牌识别准确率。但是目前基于LeNet-5网络结构的车牌识别算法存在识别不完整,运算时间长等缺点。提出基于改进的LeNet-5网络的车牌识别算法,该算法将输入车牌字符图像归一化为32×16大小,并通过删除传统LeNet-5网络中的C5层、修改输出层中神经元个数等,将车牌字符按照汉字和数字/字母的形式识别输出。通过采集大量车牌数据进行训练验证,结果表明:与前人改进的LeNet-5网络结构相比,本文算法在识别率和时间效率上均得到了提高。

License Plate Recognition Algorithm Based on Improved LeNet-5

Traditional license plate recognition algorithms have template matching, feature statistics and so on, but these algorithms rely on manual extraction of image features, and the recognition rate is low. LeNet-5 structure of convolutional neural network can automatically extract the features of license plate image and improve the accuracy of license plate recognition. However, the current license plate character recognition algorithm based LeNet-5 network has shortcomings such as incomplete recognition and long operation time. Therefore, an improved LeNet-5 license plate recognition algorithm was proposed, which normalized the input license plate character image to size of 32*16, deleted the C5 layer in the traditional lenet-5 network, and modified the number of neurons in the output layer to output license plate characters in the form of Chinese characters and alphanumeric letters. The training verification is carried out by collecting a large amount of license plate data. The results show that compared with the LeNet-5 network structure improved by other authors, the improved license plate recognition algorithm of LeNet-5 network structure improves the recognition rate and time efficiency.