Hassanein Mohamed, Lari Zahra, El-Sheimy Naser
Department of Geomatics Engineering, University of Calgary, 2500 University Dr NW, Calgary, AB T2N1N4, Canada.
Leica Geosystems Ltd.; 245 Aero Way NE, Calgary, AB T2E6K2, Canada.
Sensors (Basel). 2018 Apr 18;18(4):1253. doi: 10.3390/s18041253.
Over the last decade, the use of unmanned aerial vehicle (UAV) technology has evolved significantly in different applications as it provides a special platform capable of combining the benefits of terrestrial and aerial remote sensing. Therefore, such technology has been established as an important source of data collection for different precision agriculture (PA) applications such as crop health monitoring and weed management. Generally, these PA applications depend on performing a vegetation segmentation process as an initial step, which aims to detect the vegetation objects in collected agriculture fields’ images. The main result of the vegetation segmentation process is a binary image, where vegetations are presented in white color and the remaining objects are presented in black. Such process could easily be performed using different vegetation indexes derived from multispectral imagery. Recently, to expand the use of UAV imagery systems for PA applications, it was important to reduce the cost of such systems through using low-cost RGB cameras Thus, developing vegetation segmentation techniques for RGB images is a challenging problem. The proposed paper introduces a new vegetation segmentation methodology for low-cost UAV RGB images, which depends on using Hue color channel. The proposed methodology follows the assumption that the colors in any agriculture field image can be distributed into vegetation and non-vegetations colors. Therefore, four main steps are developed to detect five different threshold values using the hue histogram of the RGB image, these thresholds are capable to discriminate the dominant color, either vegetation or non-vegetation, within the agriculture field image. The achieved results for implementing the proposed methodology showed its ability to generate accurate and stable vegetation segmentation performance with mean accuracy equal to 87.29% and standard deviation as 12.5%.
在过去十年中，无人机（UAV）技术在不同应用领域得到了显著发展，因为它提供了一个特殊平台，能够结合地面和航空遥感的优势。因此，该技术已成为不同精准农业（PA）应用（如作物健康监测和杂草管理）数据收集的重要来源。一般来说，这些PA应用依赖于将植被分割过程作为初始步骤，其目的是在收集的农田图像中检测植被对象。植被分割过程的主要结果是一幅二值图像，其中植被以白色显示，其余对象以黑色显示。使用从多光谱图像导出的不同植被指数可以轻松执行此过程。最近，为了扩大无人机图像系统在PA应用中的使用，通过使用低成本RGB相机来降低此类系统的成本变得很重要。因此，开发用于RGB图像的植被分割技术是一个具有挑战性的问题。本文提出了一种针对低成本无人机RGB图像的新植被分割方法，该方法依赖于使用色调颜色通道。所提出的方法基于这样的假设：任何农田图像中的颜色都可以分为植被颜色和非植被颜色。因此，开发了四个主要步骤，使用RGB图像的色调直方图来检测五个不同的阈值，这些阈值能够区分农田图像中占主导地位的颜色，无论是植被还是非植被。实施所提出方法所取得的结果表明，其能够生成准确且稳定的植被分割性能，平均准确率等于87.29%，标准差为12.5%。
