Object detection in drone aerial images has always been a hot topic of research.Compared with standard images,im-age backgrounds have many small objects and large object scale variations.Therefore,traditional object detection algorithms are not suitable for direct use in drone images.To address these issues,a YOLOv8-based object detection algorithm was studied.First,in order to improve the accuracy of multi-scale object detection,a C2F-L structure based on Large Selective Kernel Network(LSKNet)was proposed.By dynamically adjusting the receptive field of the network,the contextual information of the detected ob-ject changes can be more effectively processed.The Slim-neck structure was introduced to reduce the number of parameters and im-prove the model detection efficiency.Finally,the WIoU loss function was used to improve the generalization ability and overall per-formance of the network model.Experiments on the VisDrone2019 dataset show that the mAP@0.5 of the improved algorithm reaches 33.4%,which is 1.1 percentage point higher than the original YOLOv8 and the computational complexity is reduced by 7.4%.It has been proven that the improved algorithm can effectively improve the object detection accuracy of aerial images.
万方数据