基于深度神经网络和迁移学习的高精度车辆识别系统研究

Research on high precision vehicle recognition system based on deep neural network and transfer learning

近年来,我国大力推进智慧城市和智慧交通建设,在车牌识别领域所要求的识别精度、场景适用性和反映灵敏性也越来越高。文章提出了一种基于深度神经网络算法的高精度车辆识别系统,利用开源图形化视觉处理库OpenCV和数据分析处理库NumPy对车牌进行图像预处理。基于预处理后的数据,利用深度神经网络学习框架TensorFlow进行学习训练,实现了对车牌的快速精准识别。系统首先对车牌所在位置进行定位,其次对锁定后的车牌图像进行切割,再次将车牌背景和文字通过像素点移位算法由彩色图像转换为灰度图像,最后实现字符的切割与识别,得到所要识别的车牌数据。实验结果表明,与传统识别系统相比,基于深度学习的识别系统准确率更高,识别速度更快。

In recent years,China has vigorously promoted the construction of smart cities and smart transportation,and the recognition accuracy,scene applicability,and reflection sensitivity required in the field of license plate recognition are also increasing.The article proposes a high-precision vehicle recognition system based on deep neural network algorithm,which utilizes the open-source graphical visual processing library OpenCV and data analysis processing library NumPy to preprocess the image of the license plate.Based on the preprocessed data,a deep neural network learning framework TensorFlow was used for learning and training,achieving fast and accurate recognition of license plates.This system first locates the location of the license plate,then cuts the locked license plate image.Then,the license plate background and text are converted from a color image to a grayscale image through a pixel shift algorithm.Finally,character cutting and recognition are achieved to obtain the license plate data to be recognized.The experimental results show that compared with traditional recognition systems,deep learning based recognition systems have higher accuracy and faster recognition speed.