<![CDATA[License plates are a unique feature to identify a vehicle in the combination between letters and numbers. Feature extraction needed to identify each letter and number in a digital image. There are several methods in feature extraction, one of them uses a gradient feature extraction. In this research, an application program to identify the license plate is a character gradient method and backpropagation neural network (BNN). First, the digital image is cropped to get a license plate then segmented to generate each character. The next step is the extraction feature using Character gradient to get a particular feature from each character. Backpropagation neural networks are used as data classification. This research consist of two types of testing: performance analysis based on hidden layers and feature quantity in training datasets and license plate data. From the result, we can conclude that the quantity of features affects the system performance. The highest performance rate in the first scenario test is feature 48 with 60 hidden layers, and in the second scenario, the highest is feature 108 with 60 hidden layers. The lowest performance rate is shown in feature 12 with 20 hidden layers. Keywords: License plate, character gradient method, backpropagation neural network, feature, hidden layer.]]>
Copyrights © 2020
