Cogongrass inventory and management.

A field study was conducted from 2005-2006 to test broad scale classification of cogongrass [Imperata cylindrica (L.) Beauv.] on Mississippi highway rights of ways with aerial imagery. Four mosaics of high resolution multispectral images of median and rights of way along Interstate 59 between Meridian and Laurel, Ms and rights of way along MS Hwy 528 between Bay Springs, MS and Interstate 59 were used for analysis and classification. The basis for this study was to test basic user classification methods on high resolution imagery for broad scale detection of cogongrass. The imagery was analyzed by supervised and unsupervised classification techniques based on a 5-class system in ERDAS imagine. The unsupervised classification technique began with 100 classes which were narrowed down to the five classes of interest, whereas the supervised classification technique trained the system for the five classes of interest. Near infrared (NIR), red, green, and blue spectral reflectance values for each known class area within the images, along with spatial patterns and expert knowledge, were analyzed and used to train and recode the classified image. Areas of the images suspected to be cogongrass, other roadside vegetation, road/bare soil, forest, and shadow/water were used to train the system for supervised classification and used to recode the unsupervised classification. A database of GPS points of known locations for each class within each image were used to test the accuracy of each classification. Overall accuracies for supervised classification of the images ranged from 85 to 95%, while unsupervised classification resulted in 75 to 90% accurate. Producers’ accuracies for the cogongrass class ranged from 54 to 71% with unsupervised techniques; however, supervised classification techniques resulted in 54-100% accuracy to depict cogongrass. Both classification techniques produced 100% cogongrass class user’s accuracies for all images. All other classes produced lower user’s accuracies. The results from this study show good results for cogongrass detection with basic knowledge classification techniques.