Hurricane Damage Detection by Classic and Hybrid Classic-Quantum Neural Networks
Yevhenii Trochun, Zhen Wang, Oleksandr Rokovyi, Gang Peng, Oleg Alienin, Guoiming Lai, Yuri G. Gordienko, Sergii Stirenko · 2021
This article describes a deep hybrid convolutional neural network (HNN) that uses one quantum device for binary image classification. The quantum device is a Ry quantum circuit with one qubit. This HNN configuration was evaluated on a practical problem - detection of damaged buildings on pictures taken from a satellite in addition to our previous attempts on the simple MNIST, MNIST Fashion and more sophisticated CIFAR10 and CIFAR100 datasets. The performance of HNN was compared with the reference classical model and two pretrained models: VGG-15 and RESNET-50. The measured metrics (accuracy and loss) during these experiments showed some drawbacks compared with the reference model but still support our assumption about the feasibility of HNN application for multiclass classification problems.