Acoustic tomography for multi-physics fields measurement using multi-scale dense connected networks combined with channel attention mechanisms

To enhance the reconstruction quality of temperature and velocity fields in acoustic tomography, we developed a novel collaborative reconstruction algorithm. This algorithm integrates a Multi-Scale Dense Connected Network with a Channel Attention Mechanism (MSDCN-CAM). The proposed model consists of three key modules: a multi-scale feature fusion module, a dense connection module, and a channel attention module. The reconstruction process is divided into two stages: first, a coarse distribution of the temperature and flow fields is obtained using the Algebraic Reconstruction Technique (ART). Subsequently, a high-resolution reconstruction under a fine grid is achieved by MSDCN-CAM algorithm, which effectively combines multi-scale dense connectivity with a channel attention mechanism. The model utilizes a dual-input architecture to construct a temperature field with multiple peaks and a flow field with multiple vortices. Numerical simulations demonstrate that our algorithm outperforms conventional methods (ART and Tikhonov regularization) and other neural network approaches (DenseNet and 2D-CNN) in terms of noise resistance and the collaborative reconstruction of temperature and velocity fields. Furthermore, experimental validation confirms the practicality and effectiveness of our method, showing a relative error of less than 20% for the flow field and below 5% for the temperature field.