Automated rangeland vegetation cover and density estimation using ground digital images and a spectral-contextual classifier

A method to estimate vegetation cover, density and background brightness parameters in a rangeland environment from low-altitude digital images is presented. A digital still-frame camera, mounted on a 5.2 m pole, is used to acquire images of the ground. The acquired images are then processed using an unsupervised spectral-contextual classifier to extract quantitative measurements automatically. The test results show that the extracted cover measures from the fully automated procedure provide an accuracy of 0.89 to 0.99, measured in kappa, compared with 0.36 to 0.58 and 0.79 to 0.95 from k-means clustering and maximum likelihood supervised classifications respectively. For the clump density measure, the proposed method had an error level ranging from 0 to -62%-hundreds of times less than those produced from both k-means and maximum likelihood classifications. The presented method overcomes human subjectivity inherent in other commonly used ground investigation methods for estimating vegetation cover. The results provide an accurate and objective reference for the calibration of models which relate the spectral reflectance recorded by remote sensors to quantitative measures of range condition.