Chinese herbal medicine (CHM) enjoys extensive applications in healthcare and medicine. However, current CHM recognition methods primarily rely on traditional image processing techniques and simplistic machine learning algorithms. Due to the scarcity of CHM data resources and limitations in handling image variations and noise interference, the recognition performance is often suboptimal. This paper proposes a novel Convolutional Neural Network (CNN) recognition algorithm based on Randomly Synergistic Data Augmentation (RSDA) technology to enhance model accuracy and generalization. RSDA, through the probabilistic combination of augmentation techniques, effectively addresses the issue of data scarcity, significantly improving model generalization compared to conventional methods. Experimental validation demonstrates an average increase of 9.36% in model recognition accuracy, with the accuracy of the EfficientNetB0 model trained using RSDA reaching 96.17%. These results signify a significant advancement in CHM recognition tasks, offering valuable insights and effective means for improving CHM recognition technology.