Research on Remote Sensing Image Detection Algorithm Based on YOLOv8

Remote sensing image target detection utilizes remote sensing technology to extract ground information and is widely used in urban planning, environmental monitoring and other fields. Its characteristics include high dimensionality, large scope and complexity. Aiming at the current problem of remote sensing image target detection where target features are not obvious and information is lost, an improved small target detection YOLOv8 algorithm is proposed in this paper. The first, the ADown framework single-channel convolution is introduced into the backbone to replace the original convolution module conv for reducing the information loss in feature extraction and optimize the efficiency of computation; The second, the SE attention mechanism structure is engaged into the neck to weight the extracted features with features to enhance the detailed feature and positional feature representation in the complex background. The RSOD dataset is used for the experimental training and testing of this paper, and the results indicate that the improved algorithm improves the mAP@50 by 1.5% and 0.4%, and with a reduction of 0.2 GFLOPs of computation compared to the original YOLOv8 network, which can be seen that the enhanced algorithm reduces the computational volume and improves the accuracy of detection simultaneously. The improved algorithm compared with other mainstream remote sensing target detection algorithms is more effective and advanced such as the YOLOv8 algorithm and SSD algorithm.