Lightweight Target Detection in High Resolution Remote Sensing Images

Deep learning network becomes more and more complex and demands higher computing resources. However, embedded devices not only demands for high performance, but also for computing efficiency such as real-time computation. Therefore, it is important to compress the classification model. Based on YOLOv3 network, this paper uses lightweight networks to identify aircraft, oiltank, overpasses and other targets in high resolution remote sensing images of RSOD-Dataset. MobileNet was taking as the backbone network of YOLOv3. In the experimental results, the mAP of MobileNet-YOLOv3 are 58.34. The experiment shows that MobileNet-YOLOv3 model has good performance and fast running speed, which improves the efficiency of target detection, and it is suitable for embedded devices.