The paper considers a lightweight modified version of the YOLO-v5 neural network, which is used to recognize road scene objects in the task of controlling an unmanned vehicle. In the proposed model, the pooling layer is replaced by the ADown module in order to reduce the complexity of the model. The C2f module is added as a feature extraction module to improve accuracy by combining features. Experiments using snowy road scenes are presented and the effectiveness of the proposed model for object recognition is demonstrated.

Keywords: road scene object recognition, YOLOv5, Adown, C2f, deep learning, pooling layer, neural network, lightweight network, dataset