Abstract: Problems such as inaccurate positioning,and location uncertainty in traditional license plate detection were alleviated by license plate detection algorithms Faster R-CNN and Inception ResNet v2. Accurate location of license plate is now achieved with transfer learning. License plate processing entails pixel counting to effectively extract single character. Single character recognition is then done by mLeNet5 convolutional neural network. Character images are collected under complex conditions including partial occlusion or tilt. Experiments show that this algorithm is effective and accurate.