Research on the construction of target detection algorithm for traffic scene based on deep learning

In order to improve the accuracy and real-time performance of target detection in traffic scenes, the study constructs a lightweight multi-scale target detection framework and optimizes feature extraction, detection head design and computational acceleration strategies. Based on the improved GhostNet-ECA with cross-dimensional augmented attention module, the feature expression ability is improved in low-light environment; cross-level bidirectional weighted feature pyramid and dynamic sparse Anchor configuration are used to optimize the performance of multi-scale target detection; channel pruning, mixed-precision quantization, and operator fusion techniques are introduced to reduce the computational complexity and improve the inference efficiency. The experimental results show that the algorithm performs well in small target detection and occlusion target recognition, with an inference speed of 89FPS, which is suitable for the efficient deployment of intelligent traffic sensing systems.