The modern electromagnetic environment is becoming increasingly complex. Traditional radar signal processing methods have significant limitations in dealing with strong clutter interference and stealth target detection, failing to meet the accuracy and timeliness requirements of military reconnaissance and defense systems for target identification. This research aims to construct a radar signal target detection and threat level evaluation system based on convolutional neural networks (CNNs) to achieve accurate target detection and threat assessment in complex environments. Method ologically, the CNN network architecture is first optimized (including hierarchical design, activation function selection, and optimization algorithm adaptation). Then, a diversified dataset covering simulation, experimental, and publicly available data is constructed, and data augmentation techniques are used to expand the sample. Simultaneously, a five-level threat evaluation index system based on carrier frequency, pulse width, and repetition rate is established, and a combined weighting method is used to improve the objectivity of the evaluation. Experimental results show that the proposed model achieves a Target Detection Accuracy (TDA) of 89.2%, a threat level evaluation accuracy of 87.6%, and a processing speed of 32.8 FPS, significantly outperforming traditional methods and classic CNN models. Furthermore, it exhibits good robustness in complex electromagnetic environments. This research provides an effective technical path for the intelligent upgrading of radar systems and can provide accurate threat assessment support for military decision-making, possessing significant engineering application value.

CNN, radar signal, target detection, threat level, deep learning