YOLOv8 Improved Network for Remote Sensing Video Small Target Detection

This paper proposes a remote sensing video small target detection technology based on improved YOLOv8(You Only Look Once), which is mainly used for the detection accuracy of continuous frame small targets in remote sensing video data. For small target objects that appear continuously, the neck network on the YOLOv8 algorithm is used to increase the feature map up-sampling to prevent the feature size of the small target and enhance the feature information. At the same time, LSTM(Long Short Term Memory) is introduced in the backbone network to mine small target features from long-term frames and short-term frames, while suppressing useless context information to further improve the detection accuracy of small targets. Finally, the improved model is applied to the real-time detection of remote sensing small targets. The final experimental findings demonstrate an increase in detection accuracy of more than 11.4% for the revised YOLOv8-LSTM model in the remote sensing small target video dataset.