The increasing diversity of wireless systems and the proliferation of radar and communication signals have made accurate identification of signal type and modulation a fundamental requirement for intelligent signal processing systems. Under such conditions, conventional approaches based on handcrafted classical features do not provide satisfactory performance when facing nonstationary signals in high-noise environments. In this study, a hybrid framework that combines classical features with deep learning is proposed for blind discrimination and classification of radar and communication signals. In the proposed structure, a preprocessing stage based on the continuous wavelet transform (CWT) and the extraction of fourth- and sixth-order statistical measures is first performed to obtain an informative time-frequency representation and noise-robust statistical features. To enable a more thorough evaluation and analysis, the proposed framework is examined under three different scenarios using the extracted data and features to train two deep architectures, namely an improved LSTM model and a proposed ResNet model. In the first scenario, the networks are trained solely on the communication dataset RadioML2018.01A to assess performance on communication signals; the improved LSTM model achieves an average accuracy of 79.39%, and the ResNet model achieves 81.33%. In the second scenario, the networks are trained on the radar dataset DeepRadar2022 to evaluate performance on radar signals; the average accuracy reaches 79.23% for the improved LSTM model and 83.05% for the proposed ResNet. In the third scenario, a combination of the radar and communication datasets is used, which causes a performance drop for the LSTM model, whereas the ResNet model maintains stable performance and attains an accuracy of 98.2% at SNR=2 dB. Overall, the results indicate that combining CWT-based time-frequency analysis with higher-order statistical features and deep architectures such as ResNet can provide an accurate and reliable framework for blind classification of radar and communication signals in noisy environments.