深度学习在复杂环境下车牌定位算法中的应用  

Application of deep learning in license plate locating algorithm in complicated environment

作　　者：赵伟[1] 张南楠 ZHAO Wei;ZHANG Nannan(Mechanical and Electrical Engineering Institute,Northeast Forestry University,Harbin 150040,China)

机构地区：[1]东北林业大学机电工程学院

出　　处：《现代电子技术》2019年第17期38-42,48,共6页Modern Electronics Technique

基　　金：国家自然科学基金(61405045)~~

摘　　要：为了使车牌识别技术适应复杂环境以及实际工程应用,将深度学习理论和算法融于复杂环境下的车牌定位改进算法中。首先通过处理速度快的Canny边缘检测算子,通过阈值设定得到边缘细节完整的车牌图像;然后结合形态学处理进行大范围的车牌疑似区域提取;最后利用改进的深度学习算法,即采用Alex Net卷积神经网络法去除伪车牌进行车牌精定位,并输出最后的精确定位结果。实验结果表明,该方法定位准确性、定位效率高,而且资源开销较少,此方案可持续发展性强,在未来的工程应用中具有一定的实用价值。In order to make the license plate recognition technology adapt to the complex environment and practical engineering application,the deep learning theory and algorithm are integrated into the algorithm of license plate positioning in the complex environment. The license plate image with complete edge details is obtained by setting the threshold value on the basis of Cannny edge detection operator with high processing speed. A wide range of license plate suspected region is extracted in combination with morphology processing. The improved deep learning algorithm,that is Alex.net convolution neural network method,is used to remove the false license plate for the license plate accurate positioning. And the final accurate positioning result is outputted. The experimental results show that this method has more accurate position,higher positioning efficiency and less resource cost. There is a certain practical value in future engineering application.

关 键 词：车牌定位 深度学习 CANNY边缘检测 形态学处理 疑似区域提取 复杂环境 

分 类 号：TN911.7334[电子电信—通信与信息系统] TP391.41[电子电信—信息与通信工程]