UAV Small Target Detection Algorithm Based on YOLOv8 Lightweight

Despite significant progress in the lightweight design of UAVs, the limited computational resources that can be carried, and the existence of small targets with small sizes, inconspicuous features, and susceptibility to background interference and occlusion, UAV small target detection is still challenging. To maintain a certain accuracy while significantly reducing the model size, the article proposes a small target detection algorithm based on the YOLOv8s network. Firstly, the C2f module in the Backbone part is replaced by the DynamicConv module, which introduces adaptive convolution kernel parameters to enhance the feature extraction capability while reducing the redundancy of parameters; secondly, the weighted bidirectional feature pyramid BiFPN is utilized as the Neck-end network, which operates its highly efficient feature fusion, redundancy-reducing connection and weighted feature fusion mechanisms to improves the model accuracy, reduces the number of parameters and computation in the convolution process; finally, the up-sampling method of the original model is replaced with CARAFE and the loss function is replaced with WIoU, which enhances the model's ability to extract features in small targets, and improves the mAP50% by 1.4% and reduces the model volume by 69.8% compared to the VisDrone2019 dataset. The experimental results show that the improved model can more accurately recognize small targets in aerial photography, and can greatly reduce the volume of the model while maintaining higher accuracy.