As modern warfare places increasing demands on the accuracy of ammunition damage effect assessment (ADEA), traditional assessment methods, relying on expert experience and physical simulation, have problems such as strong subjectivity, low efficiency, and difficulty adapting to complex combat environments. This paper proposes a deep learning-based ADEA model, aiming to improve the intelligence and automation level of the assessment. First, a large-scale sample database containing multi-dimensional features such as ammunition type, target characteristics, hit parameters, and environmental variables is constructed. Then, a convolutional neural network (CNN) is used to model the spatial relationship of the input features, and a multi-layer perceptron (MLP) is integrated to extract high-dimensional features to achieve accurate judgment of the damage level. During the model training process, data augmentation and loss function weighting techniques are applied to improve the model’s generalization ability. The findings indicate that the proposed model achieves an average assessment accuracy of 93% on a public ammunition damage database, while also increasing its inference speed to 1,510 samples per second. The deep learning-based ADEA model can effectively improve assessment accuracy and efficiency and has broad engineering application prospects.