Abstract <p>Carbon fiber reinforced composites are susceptible to multiple types of damage during service in practical engineering applications, with complex interactions often occurring among different damage mechanisms. Reliable identification and characterization of these damage types are essential for ensuring the structural integrity and safe operation of composite components. In this study, acoustic emission (AE) techniques and deep learning methods are employed to classify damage modes in carbon fiber reinforced composites. A novel model, termed the efficient channel attention gated residual network (ECAG-ResNet), is developed by integrating gated convolutional layers and the efficient channel attention (ECA) mechanism into a residual network framework. The proposed model’s performance is systematically compared against five conventional baseline models. The raw acoustic emission signals are first processed using continuous wavelet transform (CWT), followed by data augmentation to enhance the model’s generalization capability during training. The wavelet-transformed and augmented data are then used as inputs to the neural network. The results indicate that the ECAG-ResNet model achieved a damage recognition accuracy of 93%, significantly outperforming traditional neural networks and effectively mitigating overfitting. This presents a novel approach for utilizing acoustic emission techniques in damage identification of composite materials.</p>