A Lightweight YOLOv5 Model Integrating GhostNet and Attention Mechanism

Aiming at the problems of complex structure of neural network, large number of parameters and difficult deployment to terminal devices, this paper proposes an improved lightweight object detection model for the purpose of building a model with high classification accuracy and low complexity. First, the one-stage detector yolov5 is used as the base network. Considering the excessive redundant features of feature extraction, Ghost module is used to replace the traditional convolution, effectively allocate computing resources, to reduce the size of the model and improve the speed of model detection. When designing model bottleneck structure, Normalization-based Attention Module (NAM) is added to improve training efficiency and reduce model complexity. Experimental results show that compared with YOLOv5, the calculation speed of the proposed model is increased by 17% and the accuracy is reduced by less than 2%, which is of great significance in the environment of low-performance computing equipment.