Abstract: To remedy the problems of current YOLOv5s, including complex target detection network, many parameters and high configuration for training, a lightweight target detection algorithm-YOLOv5s_GCB, is proposed in this paper. First, the algorithm employs GhostNet as the backbone feature extraction network, taking full advantage of its less amount of computation and non-redundancy of feature graphs and thus reducing the complexity of the algorithm and improving its detection efficiency. Second, CA (Coordinate Attention) mechanism is introduced to effectively integrate the spatial coordinate information with the attention map, allowing the network to quickly extract useful features and further enhance the feature extraction ability of the algorithm. Finally, the bi-directional feature pyramid network (Bi-FPN) structure replaces the path aggregation network structure of the original algorithm, a new lightweight network model YOLOv5s_GCB is built with the fusion of multi-scale features. Compared with the original algorithm, the upgraded one maintains the target detection accuracy, but it significantly reduces the model parameters and lowers the hardware requirements for running YOLOv5 algorithm. In the VOC2007 data set, the average accuracy (mAP) of the YOLOv5s_GCB algorithm reaches 74.2%, the model volume 10.6 MB, and the amount of floating-point computation 11.3 GFLOPs (Giga Floating-point Operations Per Second). Compared with the original algorithm, the proposed one cuts the number of parameters by 30% and the weight model by 20%. The experimental results show the YOLOv5s_GCB algorithm not only maintains the detection accuracy, but also allows the model to become lightweight. Therefore, it lays some theoretical foundation for its deployment and application on the under-performing embedded platforms.