Implementation of Image Classification Using Convolutional Neural Network (Cnn) Algorithm on Vehicles Images

Abstract

The use of surveillance cameras for most agencies only relies on video recordings and storing them for a certain time. The use of this surveillance camera can be applied to determine the type of vehicle even if the camera is not in the right position. Regarding the background of the problem, this research will use the Convolutional Neural Network (CNN) algorithm, which is part of Deep Learning with the help of Keras Library and TensorFlow, to carry out the learning process on videos captured by surveillance cameras so that it can detect images from 3 types of vehicles. The dataset used is 100 images of motorcycles, 100 images of cars, and 100 images of buses. The method used is the Image Classification Method, and the model used is the best model selected from several experiments. Researchers used training and test data distribution, namely 80% and 20%. The best results were obtained with an accuracy rate of 96.49% using epoch 100, learning rate 0.001, and batch size 32. Meanwhile, vehicle images produced image accuracy for motorcycle images when using test data from outside the dataset is 78.92%, car image is 81.71%, and bus image is 82.26%.