Implementasi Mask R-CNN Dan CNN Untuk Identifikasi Merek Mobil Berdasarkan Citra
Total View This Week0
Total View This Week0
Institusion
Institut Teknologi Telkom Purwokerto
Author
Vandy, Ahmad Misry Ar Razy
Subject
T Technology (General)
Datestamp
2022-03-21 08:19:17
Abstract :
Multiple computer vision applications are applied to identify the type of vehicle. However, only few applications are able to distinguish one brand from another. The problem that have been encountered is regarding the various car brands distributing in Indonesia have a similar shape and features. The objective of this study was to determine the accuracy of the car brand detection model using the Mask R-CNN algorithm. This study used 213 images which had beed taken using a redmi 6A cellphone camera with a resolution of 12MP. The learning model used 5 models of neural network architecture with different layers, namely 4,5,6,7,8 layers. In first scenario, five architectures were applied to the image dataset taken from the front and back angles. In second scenario, the model architecture is implemented on the image dataset with the shooting angle from front, rear, and side. The experimental results show that the first scenario had the highest accuracy of 87% with a 5-layer CNN architecture. However, the percentage that obtained in layer 4 was 81%, layer 6 was 76%, layer 7 was 51%, and layer 8 was 48%. Meanwhile, the second scenario had accuracy of 49% with 5 layers. Layer 4 is 33%, layer 6 is 45%, layer 7 is 41% and layer 8 is 46%. The capture angle and number of layers affect the accuracy. so, it can be concluded that front and rear angles is the best images. While the best architecture is the 5 layer architecture. Keywords: Car brand,Computer vision ,CNN, Mask R-CNN,
Multiple computer vision applications are applied to identify the type of vehicle. However, only few applications are able to distinguish one brand from another. The problem that have been encountered is regarding the various car brands distributing in Indonesia have a similar shape and features. The objective of this study was to determine the accuracy of the car brand detection model using the Mask R-CNN algorithm. This study used 213 images which had beed taken using a redmi 6A cellphone camera with a resolution of 12MP. The learning model used 5 models of neural network architecture with different layers, namely 4,5,6,7,8 layers. In first scenario, five architectures were applied to the image dataset taken from the front and back angles. In second scenario, the model architecture is implemented on the image dataset with the shooting angle from front, rear, and side. The experimental results show that the first scenario had the highest accuracy of 87% with a 5-layer CNN architecture. However, the percentage that obtained in layer 4 was 81%, layer 6 was 76%, layer 7 was 51%, and layer 8 was 48%. Meanwhile, the second scenario had accuracy of 49% with 5 layers. Layer 4 is 33%, layer 6 is 45%, layer 7 is 41% and layer 8 is 46%. The capture angle and number of layers affect the accuracy. so, it can be concluded that front and rear angles is the best images. While the best architecture is the 5 layer architecture. Keywords: Car brand,Computer vision ,CNN, Mask R-CNN,
Download
book
BibTex
Latex, Jabref
cloud_download
Original Resource
url resource Institution
