A lightweight object detection network based on YOLOv5 with multidimensional pruning

With the rapid development of edge computing and mobile intelligent devices, lightweight object detection networks have become a research focus in the field of computer vision. Aiming at the balance between "lightweight" and "high precision" of object detection models in resource-constrained scenarios, this paper proposes an efficient lightweight object detection network based on YOLOv5. The network retains classic lightweight technologies such as depthwise separable convolution and attention mechanism, and innovatively introduces a three-dimensional parameter importance evaluation system integrating "local-global-correlation" to achieve precise pruning of redundant parameters. Combined with a fine-tuning strategy, the network effectively maintains detection performance while reducing model complexity. This design not only solves the problem of excessive parameters and computational complexity of traditional models, but also improves the efficiency of feature extraction and utilization, providing a feasible solution for the deployment of object detection technology on edge devices.