A Non-Line-of-Sight Target Detection System Based on Acoustic-Electromagnetic Intermodulation

When detecting non-line-of-sight targets, the nonlinear characteristics of the target in a silent state are masked by the surrounding environment, leading to a decrease in detection accuracy. Based on this, this paper proposes an acoustic-electromagnetic detection system based on an improved convolutional neural network (CNN), using non-negative matrix factorization (NMF). Specifically, the system couples acoustic wave energy to the target via acoustic signals, causing noncontact vibration. The radar echo signal carries abundant characteristic information of amplitude modulation and micro-Doppler modulation induced by forced vibration of the target. Subsequently, non-negative matrix factorization is introduced to extract key features of the target from the data. Finally, the NMF-processed data is input into the improved CNN model for classification. In our experiments, we classified four different target materials—wood, copper, aluminum, and stainless steel—in non-line-of-sight scenarios. The experimental results show that the classification accuracy for wood and copper reaches 95%, while the recognition accuracy for aluminum and stainless steel reaches 91% and 93%, respectively.