基于YOLOv5的遥感图像小目标检测

针对遥感图像中存在对物体检测尺寸较小、检测不准确等问题,该文提出将YOLOv5算法应用到遥感图像目标检测中,用于解决上述问题。选取DOTA数据集中2 000张图片组成训练数据和测试数据,利用YOLOv5对模型进行训练。用测试数据进行测试,YOLOv5能快速识别遥感图像中舰船、汽车等小目标。实验结果表明,m AP值为94.35%,检测精度为93.68%。与YOLOv3相比,YOLOv5用于遥感图像小目标检测方法在检测精度和速度上具有更好的表现。

Aiming at the small size and inaccurate detection of objects in remote sensing images, this paper proposes to apply YOLOv5 algorithm to target detection in remote sensing images to solve the above problems. 2 000 photos from DOTA data set are selected to form training data and test data, and YOLOv5 is used to train the model. Using the test data, YOLOv5 can quickly identify small targets such as ships and cars in remote sensing images. The experimental results show that the map value is 94.35% and the detection accuracy is 93.68%. Compared with YOLOv3, the YOLOv5 for small target detection in remote sensing images has better performance in detection accuracy and speed.