Multi-object recognition of vehicles based on multi-attention mechanism and improved Deep Neural Network

The advancement of intelligent transportation systems has made precise identification and tracking of multi-target vehicles one of the key technologies for improving road safety and traffic efficiency. However, the complex traffic environment and the diversity of vehicle motion characteristics pose higher challenges to recognition models. Therefore, this study first utilizes the idea of motion feature extraction to perform multi-target tracking and state updates on vehicles, in order to provide accurate target location information. Secondly, this study constructs a ResNet18 vehicle multi-objective recognition model based on multi-attention mechanism optimization by introducing a multi-level attention module. The results indicated that the research model could achieve a recognition accuracy of 94.33% and a minimum misidentification rate of 2.17% in complex traffic scenarios. Compared to traditional models, the proposed model had better real-time performance, with a minimum processing frame rate of 26.8 frames per second. The performance advantages of research methods in multi-target vehicle identification provide an effective technical means for vehicle recognition and tracking in intelligent transportation systems, which can promote the further development of traffic management systems in practical applications.