兽类学报 ›› 2019, Vol. 39 ›› Issue (4): 450-457.DOI: 10.16829/j.slxb.150329

• • 上一篇    下一篇

三江源大型食草动物数量无人机自动监测方法

吴方明1 朱伟伟1 吴炳方1 赵新全2,3   

  1. (1中国科学院遥感与数字地球研究所遥感科学国家重点实验室,北京 100101)
    (2中国科学院西北高原生物研究所,西宁 810001)
    (3中国学科院三江源国家公园研究院,西宁 810001)
  • 出版日期:2019-07-30 发布日期:2019-07-25
  • 通讯作者: 吴炳方 E-mail: wubf@radi.ac.cn
  • 基金资助:
    中国科学院科技服务网络计划(STS计划)项目(KFJ_STS-ZDTP-013-02);青海省科技计划项目(2019-SF-155)

Method of automatic counting of large herbivores from UAV images in the Source Region of Three Rivers

WU Fangming1, ZHU Weiwei1, WU Bingfang1, ZHAO Xinquan 2,3#br#   

  1. (1 State Key Laboratory of Remote Sensing Science, Institute of Remote Sensing and Digital Earth,Chinese Academy of Sciences,Beijing 100101, China)
    (2 Northwest Institute of Plateau Biology, Chinese Academy of Sciences, Xining 810001,China)
    (3 Three-River-Source National Park Research Institute, Chinese Academy of Sciences, Xining 810001,China)
  • Online:2019-07-30 Published:2019-07-25

摘要:

应用无人机进行动物调查的研究越来越多,但是缺乏有效的自动处理图像方法,使得目视解释的工作量较大。为研究快速与准确估算三江源区域大型食草动物数量的方法,本文使用两种无人机进行7个站点的遥感影像采集。首先对无人机影像进行灰度化,使矩阵维数下降并且梯度信息仍然保留,运算速度大幅度提高;其次对影像开展高斯滤波,高斯滤波将数据进行能量转化,排除掉属于低能量部分的噪声;第三开展阈值处理得到二值化图像;采用样本中动物形态学特征开展形态运算,先用开运算消除小物体尽可能排除干扰,同时不误删牲畜,再用闭运算排除小型黑洞将同一对象连通不重复计数;最后从二值图像中检索轮廓,并返回检测到的轮廓的个数;从而自动获得主要大型食草动物物种数量与空间分布。精度检验方法为手工计数与自动计数结果比较,相对误差3.1%~6.5%,大多数情况均可达
到此水平。采用计算机自动处理图像后,每张图像处理和计数的平均时间小于3 s。无人机影像的自动处理方法可为今后大规模进行藏羊、牦牛、藏野驴和藏原羚等动物调查提供了一种有效、可靠的技术途径。

关键词: 三江源, 大型食草动物, 无人机遥感, 阈值分割, 自动计数

Abstract:

There is more and more research on the application of UAV in animal surveys, but there is no effective automatic image processing method, which makes the work of visual interpretation heavy. In order to quickly and accurately count the number of large herbivores in the source area of the Three Rivers, two UAVs were used to collect images from seven sites. Initially, the UAV image is grayed to reduce the matrix dimension and retain the gradient information, which greatly improves the operation speed. Secondly, the image is filtered by Gauss filter, which transforms the data into energy and eliminates the noise belonging to the low-energy part. Thirdly, threshold processing is carried out to get the binary image. Then the morphological characteristics of animal husbandry in the sample are used to produce the shape. The state operation first eliminates the interference of small objects as much as possible without deleting livestock by mistake, and then excludes small black holes by closed operation to connect the same object without repeated counting. Finally, the contour is retrieved from the binary image and the number of detected contours is returned. Thus, the number and spatial distribution of the main livestock species are automatically obtained. Comparing manual counting with automatic counting, the relative error is between 3.1% and 6.5%, which can be reached in most cases. The average time of image processing and counting is less than 3 seconds after automatic image processing by computer. The automatic processing method of UAV image can provide an effective and reliable technical way for large-scale investigation of Tibetan sheep, yaks, Tibetan wild ass and Tibetan antelope in the future.

Key words: Source Region of Three Rivers, Large Herbivores, UAV Remote Sensing, Thresholding segmentation, Automatic counting