This article proposes a fingerprint database enhancement model (FASRGAN-DAE) that combines denoising autoencoder (DAE) and super-resolution generative adversarial network (FASRGAN) to address the problems of low positioning precision, poor environmental adaptability, and high computational complexity of model retraining in large-scale areas in traditional fingerprint database methods for indoor wireless positioning. Indoor positioning, as a key technology in intelligent engineering construction, directly affects the overall application effect in terms of its performance. Therefore, the method proposed in this article effectively enhances the original sparse fingerprint database by constructing high fidelity and high-density virtual fingerprint data, thereby improving the positioning precision. Specifically, the model introduces the DAE structure into the Generative Adversarial Network (GAN) framework, which enhances the ability to extract and reconstruct wireless signal features by extending the hidden layer and output layer, guiding the generator to learn and synthesize fingerprint data with key spatial features more accurately. The simulation results show that compared with traditional methods, FASRGAN-DAE significantly improves the positioning precision and exhibits stronger robustness and generalization ability in various typical indoor scenarios.
Zhao et al. (Sun,) studied this question.