OIF-CBDN: Orientation Information Fusion CBD Network for UWB-IR Vehicle Classification

To address the problem of low SNR in ultra-wideband impulse radar (UWB-IR)-based vehicle target classification, this letter proposes an Orientation Information Fusion CNN-BiLSTM-DNN (OIF-CBDN) model. The OIF-CBDN model comprises a primary channel and an orientation-assisted auxiliary channel. In the primary channel, target echo information undergoes processing using CNN, BiLSTM, and deep neural network (DNN). The auxiliary channel focuses on encoding target orientation information, which is then integrated with the feature extraction and classification modules of the primary channel. Specifically, in the feature fusion part, orientation features are incorporated into CNN convolutional features using a residual fusion technique. Additionally, the classification module incorporates orientation data into the DNN classification layer via fully connected layers. Experimental results on a UWB ground vehicle target dataset demonstrate that OIF-CBDN significantly enhances recognition performance. Notably, in low SNR conditions (SNR = −10 dB), OIF-CBDN outperforms the existing state-of-the-art methods by 7.16%.