IDENTIFIKASI PLAT NOMOR KENDARAAN MENGGUNAKAN JARINGAN SARAF TIRUAN BACKPROPAGATION

NUGROHO ANJAR SETYARSO, NIM. 13650002 (2018) IDENTIFIKASI PLAT NOMOR KENDARAAN MENGGUNAKAN JARINGAN SARAF TIRUAN BACKPROPAGATION. Skripsi thesis, UIN SUNAN KALIJAGA YOGYAKARTA.

Preview	Text (IDENTIFIKASI PLAT NOMOR KENDARAAN MENGGUNAKAN JARINGAN SARAF TIRUAN BACKPROPAGATION) 13650002_BAB-I_V_DAFTAR-PUSTAKA.pdf - Published Version Download (2MB) | Preview
	Text (IDENTIFIKASI PLAT NOMOR KENDARAAN MENGGUNAKAN JARINGAN SARAF TIRUAN BACKPROPAGATION) 13650002_BAB-II_Sampai_BAB-IV.pdf - Published Version Restricted to Registered users only Download (4MB)

Abstract

There are many challenges in implementing the vehicle license plate identification system which associated with the image acquisition process. Among them are the lighting when taking the image of the plate, the illumination of the image, the condition and the material of the plate used. For that we need appropriate identification methods and reliable features that can survive and not be affected by these conditions. In this research, the researcher tries to use a feature extraction method called Histogram of Oriented Gradient to extract character trait of plate characters, as well as Artificial Neural Network Backpropagation for its classification method. Various HOG property scenarios are applied in the ANN training and testing process to find the optimal HOG structure that can produce the highest character recognition plate accuracy. In addition to HOG, the hidden layer ANN structure applied in every training process is also always changing with a limited number of supervised. This is done to get the best structure of ANN so it can produce the best character recognition accuracy. There are two neural networks used in this study, the first network used to detect the character plate, while the second network is used to perform the recognition (classification) of plate characters. After obtaining the HOG property structure and optimal ANN structure, the system testing will be done by identifying the character plat series from the vehicle plate image using the previously trained Neural Network. This process is done by using sliding window to scan the image to find the positions of each plate characters and then sent on the character recognition network to be recognized. The results are then evaluated. From the training and network testing processes that have been done, the highest recognition accuracy that character recognition network can generate is 95.0749% of the 468 test images consisting of 13 images per character. While on the character detection network, the resulting accuracy reaches 99.8601% of the 3575 test images that are derived from 468 character images and 3107 non-character imagery. The system test results obtained an average recognition accuracy rate of 80.13716%.

Share this knowledge with your friends :

Actions (login required)

View Item