| Summary: | 碩士 === 中原大學 === 機械工程研究所 === 107 === In recent years, deep learning has made great progress in both theory and application, leading the development of a new wave of artificial intelligence. In general, deep learning is a modified multi-layer neural network, where the most representative learning architecture is the convolutional neural network. Nowadays, there are many practical applications, especially in the field of image classification and identification. The purpose of this dissertation is to develop a new license plate recognition system using convolutional neural network based deep learning architecture. First, by taking a wide range of front-end images of various types of cars, the actual license plate image is collected, and training samples of each character are obtained. Second, a convolutional neural network using mini-batch gradient descent algorithm is developed for training the data set of all character images. Next, a graphical user interface is designed to provide a friendly operating environment. Finally, all 100 license plate images are used to test and verify the system. In practical operation, when the user selects a license plate image, it must go through the appropriate image pre-processing steps, including color space conversion, contrast enhancement, binarization, image morphology, labeling, area filtering, character image segmentation, and resizing. Then, each character image is substituted into the trained convolutional neural network to obtain the identification result. This dissertation tests 100 car images and the experimental results show that the recognition success rate is about 96%, where the main reason for the failure of identification may be that the sample size of the sample database collected is too small and the angle of image capture is too skewed. In conclusion, the license plate recognition system developed in this dissertation has achieved a good performace.
|
|---|
