照相法测量冬小麦覆盖度的图像处理方法研究

Image processing methods to determine the fractional vegetation cover of winter wheat using digital camera

植被覆盖度是重要的农学和生态学参数,对水文、生态、全球变化等研究具有重大意义。数码照相法作为一种方便快捷、精度较高的地面测量方式,在科学研究中得到了广泛的应用。目前照相法测量植被覆盖度的图像处理方法可以分为彩色空间判别法和图像分类法两类。通过对冬小麦覆盖度的连续监测,对不同植被分类判据进行了对比分析,结果表明基于RGB和HLS彩色空间算法的判据具有一致性。基于景合成模型的最大似然分类方法突破了传统分类方法将图像仅分为植被与非植被两类的限制,将图像分为叶片耀斑、光照叶片、阴影叶片、光照土壤和阴影土壤5个组分,具有较高的估算精度,可以作为照相法进行覆盖度估算的首选方法。

The fractional vegetation cover（fCover） is an important agronomy and ecology parameter, which is essential in the sudies of hydrology, ecology and global change. Currently, eyeballing method, statical method, inter- pretation method using digital camera imagery and remote sensing method are the major methods for determination of fCover. Since interpretation of the digital camera imagery is an efficient and low-cost in situ measurement method, it is widely used in the scientific studies. The image processing methods can be categorized into 2 classes, one is Color Space Threshold Method （CSTM）, and the other one is Image Classification Method （ICM）. The image pro- cessing methods to determine the fCover are various, however, the comparison of the theory, robustness, accuracy and impact facors of different image processing methods are seldom disscussed. At the winter wheat field of State Key Laboratory of Earth Surface Processing and Resouce Ecology, Fangshan, China , an entire dataset of the winter wheat cover fraction was measured using digital camera （ Nikon Coolpix $500, Panasonic DMC - TZ3 and FUJIF- ILM IS - 1 ）. The data was collected once or twice a week in the field from Febrary 20th to May 20th, 2009. The RGB color space threshold method （RGB） , HLS color space threshold method （HLS） , revised HLS color space threshold method （rilLS）, ISODATA classification method （ISODATA）, Scene based Max Likehood Classification method （Max Likehood） and Segementation method using NIR imagery （NIR） are used to determine the fCover of winter wheat. The comparison of the above 6 methods shows that the criterions of RGB and HIS method are equiva- lent, and results of HLS and rilLS are similar. The scene based Max Likehood classification method using ROIs （region of interest） from 5 components of imagery： glint of leaves, sunlit leaves, shaded leaves, sunlit soil and shaded soil. The comparison result shows that it is very efficent to determine the fCover, and the scene based Max Likehood classification can be seen as the primary method to determine the fCover using digital camera. The RGB method, HIS method and rilLS method are usaually to overestimate the fCover when the true fCover is low, and the NIR method is always underestimate because it~ difficult to extract the information in the shadow. The mean values of 5 components of images are shown in RGB and HIS color space, and the distribution of 5 components in HLS color space shows that the scene can not be segemented by simple hue threshold because the irradiance of the scence is not homogenous. In future, a scene simulation model for evaluation of the different image processing methods is needed, and the white balance calibration, radiation calibration and the projection detoration of the dig- ital camera should be considered.