基于卷积神经网络的遥感图像目标分类
Remote Sensing Image Target Classification Based on Convolutional Neural Networks

近年来，大量对地观测遥感卫星被成功发射并在轨运行，无人机等新型遥感平台也在不断发展更新，传统的人工目视解译已不能满足遥感影像解译在效率和精度方面的需求，深度学习在遥感影像处理方面表现出较好的可靠性和高效性。基于深度学习的遥感影像地类识别技术数据处理及特征提取能力较强，能够有效提升识别精度，使地类信息获取更加智能化，因而被广泛应用于遥感影像地类处理。遥感影像应用的核心和关键是遥感影像解译，遥感影像大数据时代智能解译提供了新的解决方案，已经成为测绘遥感学科发展的重要驱动力量。本文就遥感影像分类开展研究，首先介绍了本文所使用的DOTA数据集和YOLOv5算法，其次，建立遥感影像识别的YOLOv5项目工程，并且设置算法的关键参数；最后，根据识别结果与可视化面板对模型进行分析。遥感图像信息容量大、各类地物交错复杂，在识别方面有一定的挑战。本文的研究结果对遥感技术更好的应用提供了基础，为人类生活提供更多帮助。
In recent years, Earth observation remote sensing satellites have been successfully launched and operated in orbit, and new remote sensing platforms such as unmanned aerial vehicles are con-stantly developing and updating. Traditional manual visual interpretation can no longer meet the efficiency and accuracy requirements of remote sensing image interpretation. Deep learning has shown good reliability and efficiency in remote sensing image processing. The remote sensing im-age land class recognition technology based on deep learning has strong data processing and fea-ture extraction capabilities, which can effectively improve recognition accuracy and make land class information acquisition more intelligent. Therefore, it is widely used in remote sensing image land class processing. The core and key to the application of remote sensing images is remote sensing image interpretation. In the era of big data in remote sensing images, intelligent interpretation provides new solutions and has become an important driving force for the development of survey-ing and remote sensing discipline. This article conducts research on remote sensing image classifi-cation. Firstly, relevant knowledge of the DOTA dataset and YOLOv5 algorithm used in this article were introduced. Secondly, the YOLOv5 project for remote sensing image recognition was estab-lished, and the key parameters of the algorithm were set. Finally, the model was analyzed based on the recognition results and visualization panel. Remote sensing images have a large amount of in-formation capacity and complex intertwining of various land features, posing certain challenges in recognition. The research results of this article provide a foundation for the better application of remote sensing technology and provide more assistance for human life.