Vehicle target detection algorithm in complex environment based on hybrid attention mechanism

Aiming at the problems of significant background interference, variable target scale and insufficient real-time performance in vehicle target detection in complex environments, this paper proposes an improved model based on hybrid attention mechanism and polarisation filtering (PTRDarknet-POSA). Firstly, the PTRDarknet module is embedded into the YOLOv5 backbone network<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">[1]</sup>, which adaptively enhances the channel characterisation and spatial localisation of the target critical region by fusing the SE channel attention and spatial attention mechanisms, and effectively suppresses the complex background interference. This hybrid attention mechanism enables the model to identify the target more accurately and reduce the false detection rate.Secondly, a polarised filtered feature pyramid network (POSA-FPN) is designed, and a dynamic filtering kernel is introduced to optimise the multi-scale feature fusion process, which improves the detection accuracy and efficiency. On the KITTI dataset, The model [email protected] reached $88.2 \%$, a $1.6 \%$ improvement over the original model was achieved. This significant improvement verifies the superior performance of the model in complex scenarios. Experimental results show that the method significantly improves the robustness of vehicle detection in complex situations such as rain, fog, and occlusion, providing an effective solution for autonomous driving systems. This innovative design not only enhances the detection performance, but also improves the real-time response capability, making the autonomous driving technology more reliable and efficient in complex environments. By optimising the feature extraction and fusion strategies, the model demonstrates excellent adaptability and stability, laying a solid foundation for the future development of autonomous driving technology.