No abstract

No abstract

As the size of the radar hardware platform becomes smaller and smaller, the cost becomes lower and lower. The application of indoor radar-based human motion recognition has become a reality, which can be realized in a low-cost device with simple architecture. Compared with narrow-band radar (such as continuous wave radar, etc.), the human motion echo signal of the carrier-free ultra-wideband (UWB) radar contains more abundant characteristic information of human motion, which is helpful for identifying different types of human motion. In this paper, a novel feature extraction method by two-dimensional variational mode decomposition (2D-VMD) algorithm is proposed. And it is used for extracting the primary features of human motion. The 2D-VMD algorithm is an adaptive non-recursive multiscale decomposition method for nonlinear and nonstationary signals. Firstly, the original 2D radar echo signals are decomposed by the 2D-VMD algorithm to capture several 2D intrinsic mode function (BIMFs) which represent different groups of central frequency components of a certain type of human motion. Secondly, original echo signals are reconstructed according to the several BIMFs, which not only have a certain inhibitory effect on the clutter in the echo signal, but can also further demonstrate that the BIMFs obtained by the 2D-VMD algorithm can represent the original 2D echo signal well. Finally, based on the measured ten different types of UWB radar human motion 2D echo analysis signals, the characteristics of these different types of human motion are extracted and the original echo signal are reconstructed. Then, the three indicators of the PCC, UQI, and PSNR between the original echo signals and extraction/reconstruction 2D signals are analyzed, which illustrate the effectiveness of 2D-VMD algorithm to extract feature of human motion 2D echo signals of the carrier-free UWB radar. Experimental results show that BIMFs by 2D-VMD algorithm can well represent the echo signal characteristics of this type of human motion, which is a very effective tool for human motion radar echo signal feature extraction.

Radar-based human motion recognition has received extensive attention in recent years. Most current recognition methods generate a heat map of features through simple signal processing and then feed into a classification-based neural network for recognition. Such an approach can only identify a single action. When a set of data contains information about multiple movements, it can also only be recognized as a single movement. Another point that cannot be overlooked is that continuous action recognition methods are able to recognize continuously changing actions but ignore the issue of whether continuous actions are legitimate or not (continuous actions obtained by stitching together multiple current actions do not conform to real time). In this paper, we propose a continuous action recognition method based on micro-Doppler features and transformer, which translates the micro-Doppler features of continuous actions into machine translation tasks and uses the idea of natural language processing (NLP) to identify continuous action. In order to judge whether the continuous action is legal or not, we also design the action state transition diagram as a constraint condition to strictly control the forward and backward actions. The experimental results show that the method proposed in this paper achieves good recognition accuracy for the recognition of a single action and can also effectively segment and recognize continuous actions.

Utilizing the polarization analysis in underwater imaging can effectively suppress the scattered light and help to restore target signals in turbid water. Neural network-based solutions can also boost the performance of polarimetric underwater imaging, while most of the existing networks are pure data driven which suffer from ignoring the physical mode. In this paper, we proposed an effective solution that informed the polarimetric physical model and constrains into the well-designed deep neural network. Especially compared with the conventional underwater imaging model, we mathematically transformed the two polarization-dependent parameters to a single parameter, making it easier for the network to converge to a better level. In addition, a polarization perceptual loss is designed and applied to the network to make full use of polarization information on the feature level rather than on the pixel level. Accordingly, the network was able to learn the polarization modulated parameter and to obtain clear de-scattered images. The experimental results verified that the combination of polarization model and neural network was beneficial to improve the image quality and outperformed other existing methods, even in a high turbidity condition.