Predicting missing field boundaries to increase per-field classification accuracy

Anew technique for predicting missing field boundaries was developed to increase the accuracy of per-field classification. This technique is based on a comparison of within-field modal land-cover proportion and local variance. Analysis was performed on 4-m and 20-m spatial resolution imagery derived from Compact Airborne Spectrographic Imager (CASI) data, to simulate the difference in land-cover classification accuracy between multispectral Ikonos and Satellite Pour l'Observation de la Terre (SPOT) High Resolution Visible (HRV) imagery. Initially, per-pixel classification was performed, followed by per-field classification. The technique for detecting missing boundaries was then implemented, and per-field classification was carried out a second time using updated field boundary data. Finally, an accuracy assessment was performed. The results demonstrate that classification was significantly more accurate when the missing boundary flag was used, and that simulated Ikonos imagery was considerably more accurate for this purpose than simulated SPOT HRV imagery.