Perpustakaan Digital ITB

The development of computer vision have a purpose to to create autonomous system which could perform visual tasks that even might surpass human’s performance with the emerging popular approach is using convolutional neural network (CNN). In aerospace engineering field, there are wondrous possibilities of computer vision application, like for maintenance, research, military, and aircraft control. Utilizing CNN, this work would solve one engineering case, which is recognizing crack in concrete surface, and one non-engineering case, which is recognizing and classifying types of tumor. The model would compromised of established convolutional architectures combined with the author’s proposed classifying architecture. The work compare four different convolutional architectures performance, which are VGG16, DenseNet201, InceptionV3, and EfficientNetV2S. This work also investigate the feature highlight of the model using SmoothGrad saliency map and Grad-CAM. The result of this work is that for the binary classification of the concrete crack case, the best model to use is the one using VGG16 while for the categorical classification of the brain tumor case the best model to use is the one using EfficientNetv2S. The consideration in these recommendation is the quantitative classifying performance and the model’s feature highlight.