KLASIFIKASI CITRA KOMPONEN SEPEDA MOTOR MENGGUNAKAN ALGORITMA CNN DENGAN ARSITEKTUR MOBILENET

Abstract

Image recognition is a sub-category of computer vision technology used to classify images into specific categories. The purpose of this research is to create a CNN model with the MobileNet architecture to classify motorcycle component images and measure the accuracy level produced by the model. The creation of the deep learning CNN model uses the TensorFlow library. The initial data for the training process consists of 50 images divided into 5 categories: spark plugs, brake pads, bearings, regulators, and roller housings. These data undergo augmentation techniques such as rotation, shifting, and image flipping. This research successfully developed a CNN model using the MobileNet architecture that can classify motorcycle component images. The MobileNet model was tested using 20 test data, with 10 of them subjected to a motion blur filter. The test results showed that the accuracy performance of the CNN model with the MobileNet architecture in classifying motorcycle component images is 85%, and the accuracy of image classification did not significantly decrease when the motion blur filter was applied.