UAS for Remote Sensing - Image Classification

We use object-based image analysis (OBIA) (eCognition software) for image classifications. OBIA is more suitable than pixel-based classification for high and very high resolution imagery. The classification scheme is hierarchical and employs a rule-based masking approach, where an image is first classified into easy to define classes (bare/vegetated, shadow/non-shadow) using rules, followed by further classification to the species level with a nearest neighbor classification. Decision trees can be used to select suitable features for classification. We use texture features and hue, saturation, and intensity to compensate for the low spectral and radiometric resolution in the Canon imagery. Future classifications of the multispectral imagery will take advantage of the greater spectral and radiometric resolution.

In a study conducted in southwestern Idaho, we used OBIA to classify UAS imagery of 50 m x 50 m plots measured concurrently on the ground using standard rangeland monitoring procedures.

UAS images and classifications of six 50m x 50m plots

  Correlations between image- and ground-based estimates of percent cover for bare ground, shrubs, and grass/forbs resulted in r-squared values ranging from 0.86 to 0.98. Time estimates indicated a greater efficiency for the image-based method. Classification accuracies for rangeland vegetation maps commonly range from 78% to 92%, depending on the number of classes.

UAS image mosaic (center), enlarged area of mosaic (left), classification of image (right)

  Due to the relatively large file size of image mosaics, we develop the classification rule or process trees on small areas and transfer the rule base to the larger mosaic.