In radar intrapulse modulation recognition through convolutional neural networks (CNN), the recognition rate vis-a-vis varying signal-to-noise ratios (SNRs) is utilized to analyze the performance of networks. However, three shortcomings exist in this realm, (1) Proportional increase in recognition rate up to 100%, due to SNR increase, leading to saturation. (2) The dependency of recognition rate on the test set. (3) Recognition rate failing to substantiate the increase in SNR and its manifestation in CNN performance enhancement. Foregone in view, this letter proposes an evaluation criterion to evaluate CNN performance to address the shortcomings above. We propose Grad-CAM position score-Internal Benchmark (GCPS-IB) to address (1), (2), and Grad-CAM position score-External Benchmark (GCPS-EB) to address (3). Finally, we utilize GCPS to evaluate and compare the performance of existing GoogLeNet and ResNet-18 to verify the efficacy. The proposed criterion establishes its efficacy by addressing the shortcomings mentioned above, hence augmenting the recognition rate.
Multi-function radar (MFR) work modes recognition is an important research component of the electronic reconnaissance field. When facing MFR systems equipped with complex mode-waveform mapping relationships and flexible beam scanning techniques, the intercepted work mode pulse sequences have a wide temporal range of feature distributions and variable durations, which bring significant challenges for accurate recognition. To address this issue, this study constructs a novel hierarchical MFR signal model with waveform multiplexing and waveform scheduling laws with spatial beam arrangement and proposes a work mode recognition method based on dual-scale feature extraction. The recognition method first obtains the variable-length sequence processing capability through pulse sequence segmentation. Then, a structure composed of convolutional neural network (CNN) and long short-term memory (LSTM) is followed to extract the deep time-series features at the internal-segment scale of segments, and the features of each segment are concatenated in the time dimension. Subsequently, an LSTM-Attention network is used to extract the external-segment-scale features while adaptively assigning a higher weight to important waveform segments. Ultimately, the work mode recognition results are obtained. The experimental results show that the proposed method’s performance is advantageous in recognizing work modes under the comprehensive MFR signal model.
7 August 2025 in Engineering Applications of Artificial Intelligence
To address the problem of aliasing caused by pulse data streams from multiple dense signal sources, traditional density-based clustering algorithms often exhibit degraded clustering performance, strong reliance on manually tuned parameters, and limited effectiveness of conventional association methods. This paper proposes a sorting algorithm based on adaptive root particle swarm optimization (PSO) and an association algorithm using Cauchy distribution matching. First, the K-mutual nearest neighbor algorithm is combined with DBSCAN to avoid misjudging core points through strict mutual neighbor determination and to prevent cross-cluster expansion using a proportional factor. Then, the optimal parameters of the K-mutual nearest neighbor DBSCAN are obtained through the adaptive root PSO algorithm, the adaptive root PSO algorithm incorporates an adaptive root growth mechanism. By dynamically adjusting inertia weights and individual and population factors, and by employing Levy flight and opposing learning strategies, it enhances search diversity and the ability to escape local optima. Compared to traditional PSO algorithms, it can more effectively address complex optimization problems. Finally, the Cauchy distribution matching algorithm is applied to associate the sorting results of two consecutive observation frames. Through simulation verification, under Dataset 1 conditions, when 20% interference pulses are added, the proposed algorithm achieves a 19.94% improvement in sorting accuracy compared to DBSCAN and attains an association accuracy of 83.875% in inter-frame association experiments.
In this paper, we propose a new type of time-frequency distribution, named double-sided exponentially forgetting transform (DSEFT). In addition, an algorithm for instantaneous frequency estimation using DSEFT is presented. It is shown that the DSEFT offers good time-frequency resolution and that stable reconstruction of a signal from samples of the DSEFT is possible. Compared with other forms of time-frequency representation, DSEFT is highly computational efficient, since iterative computation can be used. Analysis shows that DSEFT is superior to the exponential forgetting transform as an instantaneous frequency estimator. The bias and the mean square deviation of the estimator are discussed and its validness is verified by simulations.