Abstract: With the development of 5G technology, its high bandwidth, low latency and high density access features have led to a change in the cloud computing model to a’cloud-management-end’ model, and edge computing as a key terminal technology has become critical to the deployment of AI algorithms on terminals with limited computing power. Taking the video retrieval algorithm for pine tree plant identification in nursery acceptance as an example, a lightweight algorithm for pine sapling detection and counting in nurseries suitable for terminal deployment of AI algorithms in proposed. The algorithm achieves network lightweighting by introducing the MobileNet v3 feature extraction mechanism on the basis of the YOLOv5 network, integrating the lightweight attention module in Squeeze-and-Excitation Networks（SENet） as a bneck basic block to improve the network’s sensitivity to feature channels and enhance the network’s feature extraction capability. The vector angles of the target and prediction frames are further considered on the IoU basis. The SIoU loss function is used as the prediction function and the associated loss function is redefined, thus making the nursery sapling prediction frame closer to the real frame. The results of the study show that the number of parameters of the improved model is significantly reduced, and the size of the improved network model is compared with the method in the comparison experiment, the model has a 21.48% improvement in frame rate（FPS） to reach to 71.43 frames per second with a 3.26% reduction in accuracy（Precision） and a 1.03% reduction in mean average precision（mAP）, and the computational effort is reduced from the original YOLOv5s reduced 148.44%, proving that the algorithm is highly efficient and lightweight, providing an algorithm prototype for the porting of artificial intelligence algorithms to edge computing terminals.