Abstract: Accurate identification of traffic vehicle loads is one of the main tasks in the field of bridge monitoring.Considering the non-stationarity of bridge responses caused by vehicle loads,the instability of VGG16 model,and the scarcity of actual bridge vibration response data,this paper proposes a traffic vehicle load identification method based on the fusion of fractional Fourier transform (FRFT),optimized VGG16,and transfer learning.Firstly,to improve feature extraction from non-stationary signals,the response was decomposed using FRFT,and time-frequency analysis in the fractional Fourier domain yielded FRFT spectra as model inputs,thereby enhancing feature extraction accuracy.Secondly,to address the potential instability and overfitting problems in the VGG16 model training,the activation function and fully connected layer modules were improved to enhance the model identification accuracy.Subsequently,transfer learning was introduced to alleviate the scarcity of actual bridge vibration response data.The bridge response data simulated by the finite element method was used for model pre-training,and the model parameters were fine-tuned with a small amount of experimental data to achieve vehicle load identification.Finally,the effectiveness and superiority of the proposed method were verified through numerical simulation combined with laboratory examples.The results show that compared with methods without FRFT time-frequency preprocessing,unoptimized VGG16 models,and models with FRFT coefficients as input,the proposed method achieves higher accuracy.The influence of traffic vehicle moving speed and material error on the identification results is studied,indicating that the proposed method has high vehicle load identification accuracy.