中文摘要：

利用安装在无人机平台上的双光谱相机所获取的可见光和近红外遥感图像,采用改进的加权支持向量数据描述多分类算法,实现病害松树识别。首先根据不同内容信息图像的特点,提取双光谱相机所获取的可见光图像和近红外图像各颜色分量作为相应像素点的颜色特征,再通过提取加窗图像块的灰度共生矩阵得到中心像素点的纹理特征,然后利用权重系数为每类样本分别作加权支持向量数据描述,实现松树状态的多输出分类识别,其中权重系数是通过建立关于训练样本中心距离的权重函数所确定。与传统的人工、航空和卫星遥感识别方法不同,利用无人机平台和双光谱相机获取遥感图像,具有可操作性强、费用低廉等优势。试验结果表明,相比传统的支持向量机和支持向量数据描述算法,改进的加权支持向量数据描述多分类算法更能准确地进行病害松树识别。

英文摘要：

An improved multi-classification algorithm of weighted support vector data description （WSVDD） was applied for the recognition of infected pine by utilizing the visible and near-infrared images acquired by the double spectrum camera fixed on the unmanned aerial vehicle （UAV） platform. Each color component for visible and near-infrared images acquired by the double spectrum camera was extracted as the color feature of the corresponding pixel on the basis of the difference of content information. Then the texture feature of the central pixel was acquired by extracting the gray level co-occurrence matrix of the adding window image block. The weight coefficient was used for the WSVDD of each kind of sample in order to realize the multi-classification and recognition of pine state. Here the weight coefficient was determined by building the weight function on the center distance of the training sample. Compared with the other methods such as manual work, aerial and satellite remote sensing, this method for acquiring the remote sensing image by using the UAV platform and the double spectrum camera was more operable, more low-cost etc. The experiment results showed that the WSVDD multi- classification algorithm could recognize the infected pine more accurately than the traditional methods of support vector machine （SVM） and support vector data description （SYDD）.