The intelligent capture and analysis model for highway toll evaders is an important component of maintaining the order of highway operation. Accurate vehicle recognition can effectively promote the intelligence of vehicle capture. The existing vehicle re-identification algorithms still suffer from inaccurate recognition in complex recognition environments. Therefore, a model for intelligent capture analysis of highway evasion vehicles based on the vehicle re-identification method is proposed. Multi-dimensional self-attention is combined with a multi-dimensional feature fusion network for optimization and finally verified through simulation experiments. The experimental results showed that the multidimensional self-attention and multi-dimensional feature fusion network used in the study achieved the best performance compared to existing methods. After combining multi-dimensional self-attention and multi-dimensional feature fusion networks, the model indicators were further improved on the VehicleID dataset. The small test set had a 6.56% increase in average accuracy. On the VeRI-776, the performance indicators for vehicle identification were further improved. Especially after combining multi-dimensional self-attention with multi-dimensional feature fusion, the mAP reached 84.69%, the Rank-1 reached 97.64%, and the Rank-5 reached 98.15%. The intelligent capture and analysis model for highway toll evaders proposed in the study is significant for optimizing highway operation management and promoting the construction of smart highways.