深度学习用于无人机影像树种识别研究进展

Progress on Tree Species Identification from Unmanned Aerial Vehicles

提高遥感数据质量和利用新的图像处理方法是提高遥感图像识别精度两个主要方向.树种识别是林业遥感领域国际性关注的学术问题,深度学习方法用于无人机影像树种识别初见端倪.介绍无人机类型、外业飞行参数,在总结无人机高光谱影像树种识别现状以及传统机器学习方法在无人机树种识别的基础上,分析卷积神经网络与迁移学习方法在无人机树种识别的国内外研究现状.研究指出无人机树种识别数据集构建与共享迫在眉睫;逐步建立卷积神经网络在无人机树种识别的标准与规范;采用并行计算方式,加速卷积神经网络训练时间;拓展无人机森林资源调查因子提取,进行森林健康实时监测,从而改进和完善我国森林资源监测体系.

There are two ways to improve classification accuracy of remote sensing image recognition. One is to improve the quality of remote sensing data, the other is to adopt new image processing methods. Tree species identification is an international academic issue in the field of forestry remote sensing. It is the beginning to identify tree species from UAV images by deep learning. Firstly, the UAV type and parameter design of field flight were briefly introduced. Then, cutten status of tree species recognition in UAV hyperspectral images and the traditional machine learning methods used in UAV tree species recognition were summarized. Furthermore, the research status of convolution neural network and transfer learning method applied in UAV tree species recognition was analyzed. The research further points out that it is urgent to construct and share UAV tree species identification data set. It is necessary to establish the standards and norms of convolutional neural network in UAV tree species identification. To improve China’s forest resources monitoring system, more forest resources investigation factors were extracted from UAV and parallel computing will be adopted to accelerate the training time of convolutional neural network for real-time monitoring.