Light‐weight deep learning method for active jamming recognition based on improved MobileViT

Abstract

In recent years, there has been significant research focused on identifying jamming signals using neural networks to improve identification accuracy. However, traditional networks require large‐scale datasets for training, which is challenging in small sample scenarios such as actual battlefield jamming signal data. Additionally, the large number of parameters in traditional models presents deployment challenges for devices. To address these issues, this study proposes a lightweight neural network model called MobileViT_CA, an improved version of MobileViT, for jamming recognition in small sample scenarios. MobileViT_CA adopts a coordinate attention mechanism suitable for lightweight networks, which further improves the classification performance of the network. The model converts one‐dimensional jamming signals into two‐dimensional time‐frequency distribution images through time‐frequency analysis and then uses the network for recognition. Compared with traditional neural network models, MobileViT_CA significantly reduces the network parameter volume while maintaining high recognition accuracy, accelerating network inference speed, and reducing hardware requirements. Experimental results show that MobileViT_CA performs well in small sample data. The model's fast learning ability has also been verified by using transfer learning, and the use of Grad‐CAM and t‐SNE to analyse MobileViT_CA demonstrates its advantages in feature recognition and extraction. Overall, this approach has the potential to significantly improve the reliability and effectiveness of jamming recognition systems in battlefield scenarios.

Compound Jamming Recognition Based on a Dual-Channel Neural Network and Feature Fusion

“…For a fair comparison, several recognition methods based on neural networks and feature images are employed as comparison methods, namely, the JRNet [12], the MBv2 [14], and the IResNet [15]. The overall accuracy (OA) of each method for each compound jamming is listed in Table 2, where the OA is defined as:…”

“…Simulation results indicate that the proposed CNN achieves more than 99% accuracy when the JNR is −3 dB. In [14], a lightweight improved MobileViT for TFIs is employed to recognize six kinds of jamming signals and the method can effectively reduce the computational complexity. Similarly, using time-frequency images (TFIs) by the STFT, an inverse ResNet enhanced by a channel attention module is designed to recognize eight kinds of jamming signals in [15], where features in the time-frequency domain and image channel domain are combined to promote recognition performance.…”