The paper presents a novel method called GLIP (Generative Local Image Prior) for reconstructing electromagnetic field (EMF) exposure maps in an urban environment. The key highlights are:
The method uses only a deep generative network (encoder-decoder architecture) and does not require a large training dataset or full reference exposure maps, overcoming the limitations of previous GAN-based approaches.
The model is trained using sparse sensor data as the Local Image Prior (LIP), without the need for explicit training on full exposure maps.
Experimental results show that GLIP can accurately reconstruct EMF exposure maps, even when only 1% of the area is covered by sparse sensor measurements. The reconstruction quality improves as the sensor density increases.
Compared to using random input features, the LIP-based approach (GLIP) demonstrates significantly better performance, with mean squared error (MSE) as low as 2.68e-5 when using 100 sensors.
The method effectively captures the complex propagation dynamics in the urban environment, considering factors like building characteristics, without relying on simplistic signal propagation assumptions.
The proposed framework is computationally efficient and can be deployed in new areas without the need for time-consuming training on full reference maps.
לשפה אחרת
מתוכן המקור
arxiv.org
תובנות מפתח מזוקקות מ:
by Mohammed Mal... ב- arxiv.org 05-07-2024
https://arxiv.org/pdf/2405.03384.pdfשאלות מעמיקות