面向人类活动识别的量子生成对抗网络数据增强与分类优化OA

To address the problem of classifier performance degradation caused by data imbalance in human activity recognition,traditional methods such as oversampling and cost-sensitive learning can alleviate the class bias,but they have drawbacks such as overfitting,information loss,and poor adaptability to extreme imbalance scenarios.This paper combines quantum computing and generative adversarial networks to design a joint optimization framework.Its generator uses a Variational Quantum Circuit(VQC),which implements implicit modeling of high-dimensional sensor data through parameterized quantum gate stacking with low parameter quantity;The discriminator uses a classical neural network and introduces the Quantum Wasserstein Generative Adversarial Network-Gradient Penalty(WGAN-GP)to constrain the discriminator,in order to enhance the diversity of generated samples and the stability of training.Experimental results based on the CASAS household activity recognition dataset show that the quantum hybrid generator only requires about 13%of the parameters of the classical model to achieve better generation effects.The improved quantum generative adversarial network not only improves the convergence speed but also enhances the accuracy of the final results.