Radar intelligence data recognition plays a vital role in both military and civilian applications. However, traditional methods often have difficulty with complex backgrounds, detecting small targets, and being robust against interference. To address these challenges, this study improves the Region-based Fully Convolutional Network (R-FCN) by incorporating EfficientNet, Mixup, and multi-scale convolution to enhance feature representation and generalization. In addition, the dictionary learning component of the sparse representation classification (SRC) network has been modified to include a dynamic adaptive neighborhood mechanism. A fusion framework that combines R-FCN and SRC has been proposed to integrate detection and classification. The experimental results demonstrated that the proposed method outperformed baseline models consistently across multiple datasets and noisy environments. This shows that the method is robust and can generalize across domains. This work provides an effective solution to enhance the accuracy and stability of radar intelligence data recognition, with promising applications in military defense and intelligent transportation.