In response to the challenges of missing detection and false detection in small object detection,an improved lightweight small object detection algorithm NSCOT-YOLO is proposed.Firstly,a feature extraction module NSC-MFFM is designed,which uses non-strided convolution for feature extraction and multi-branch for feature fusion,to effectively avoid the important feature information of small object being ignored and improve the detection accuracy of the model,Experiments are conducted to determine optimal placement of the module.Then,a shallower detection branch and detection layer are added,while the deepest detection branch is removed,The C-Cot module,which incorporates rich contextual information and shared weights,is employed to replace the detection layer of YOLOv8s,so as to better separate small-scale object from noise background.Finally,a lightweight and efficient GhostConv module is introduced,leading to a significant reduction in the models parameter count without significant decrease in accuracy.The effectiveness of the proposed algorithm is evaluated on the published VisDrone2019-DET dataset.The experimental results show that the mAP@0.5 and mAP@0.5∶0.95 of NSCOT-YOLO algorithm is 38.3% and 22.0%,which is 5.5 and 3.2 percentage points higher than that of YOLOv8s algorithm,and the model parameters count are 6.8×106,which is 39% lower than that of YOLOv8s algorithm.It can be seen that the proposed algorithm significantly improves detection accuracy for small objects while ensuring the lightweight of the model.