Automated Cover Estimation Using Ground Digital Images

This paper reports the development of a method to automate classification of digital images acquired from digital still-frame cameras. As the images to be processed record extensive details of the ground, it is difficult and inaccurate to classify such images without considering pattern structures of features such as bare soil, shrubs, grass and gravels. To overcome the problem, a statistical label relaxation algorithm was developed and implemented. The results from this method were encouraging showing high accuracy in the resulting classification which in turn provides essential statistics on cover types of testing images, with minimal human-machine interaction. Together with the development of field instruments, this classification method has made automated and accurate field sampling possible for rangeland vegetation investigation.