Research on Target Extraction System of UAV Remote Sensing Image Based on Artificial Intelligence

In many mountainous areas, geological disasters are prone to occur, such as landslides, debris flows and landslides, which are devastating to the affected areas. However, the rescue of the disaster area is very difficult. Because it is difficult to accurately identify the complex terrain, it is impossible to accurately grasp the specific information of the disaster area in a short time. unmanned aerial vehicle (UAV) has the advantages of all-weather, high-resolution and long-distance photography, and is widely used in geological hazard monitoring. With the development of artificial intelligence technology, the accuracy of classification and recognition of remote sensing images is continuously improved, and the deep learning algorithm can be well applied to the target extraction system of unmanned aerial vehicle remote sensing images. In this paper, the Faster Region-Convolutional Neural Network (R-CNN) and YOLOv3 algorithms are compared for target detection. The results show that YOLOv3 algorithm has better vehicle target extraction accuracy compared with Faster R-CNN algorithm. The recognition accuracy of station wagon, off-road vehicle, pickup truck and engineering vehicle is 91%, 92%, 89% and 93% respectively. Therefore, applying YOLOv3 algorithm to target extraction system of UAV remote sensing image can improve the accuracy of target extraction.