To reconstruct the target-induced attenuation image, the existing radio tomographic imaging techniques often search the solution in the attenuation signal space and regularize the solution to be consistent with explicitly pre-defined prior knowledge of the attenuation signal. However, the performance will inevitably deteriorate when the prior knowledge fails to be consistent with the signal, especially in complicated scenarios. To address this issue, this paper explores the use of generative model in building an attenuation image. The logic behind of using generative model technique here is to learn the inherent latent structure of the attenuation signal from the data itself rather than its prior knowledge. Specifically, an estimate of the unknown signal will be learnt by means of an untrained generative model and it will be kept consistent with the observed measurement data. In the context of device-free localization of interest here, the L1 regularization is incorporated into the learning process, which will promote the sparsity of the solution. The experimental studies on device-free localization with radio tomographic imaging are presented to validate the effectiveness of the proposed approach.

为了重建目标诱发的衰减图像，现有的放射线断层成像技术经常在衰减信号空间中搜索解并将其正规化，以与衰减信号的明确预先定义的先验知识相一致。但是，当先验知识与信号不一致时，性能将不可避免地下降，尤其是在复杂的情况下。为了解决这个问题，本文探讨了生成模型在构建衰减图像中的用途。这里使用生成模型技术的逻辑是从数据本身而不是先验知识中学习衰减信号的固有潜在结构。特别，未知信号的估计值将通过未经训练的生成模型学习，并将与观察到的测量数据保持一致。在此处关注的无设备本地化的情况下，将L1正则化合并到学习过程中，这将促进解决方案的稀疏性。提出了利用射线断层成像技术对无设备定位进行实验研究，以验证所提出方法的有效性。