Intensified fuzzy clusters for classifying plant, soil, and residue regions of interest from color images

Fuzzy excess red (ExR) and excess green (ExG) indices and clustering algorithms: fuzzy c-means (FCM) and Gustafson-Kessel (GK) were studied for unsupervised classification of hidden and prominent regions of interest (ROI) in color images. Images included sunflower, redroot pigweed, soybean, and velvet leaf plants, against bare clay soil, corn residue and wheat residue, typical of the Great Plains. Indices and clusters were enhanced with Zadeh's (Z) fuzzy intensification technique. Enhanced ROIs were sorted by degree of fuzziness, and recombined into labeled, false-color class images. ROIs with the lowest degree of fuzziness were consistently found to be plant clusters with some of the methods. The ZExG index only classified plant ROIs correctly at 76% (newly emerged) and 74% (young plants) for soil backgrounds, 55-65% for corn residue, and only 12% for young plants with wheat straw. The ZExR index failed for almost all categories, except bare soil. The ZFCM clustering algorithm correctly classified plants from 10 to 69% in bare soil, but failed for plants in corn and wheat residue. The ZGK algorithm classified plants from 16 to 96% in bare soil, and corn residue plants as high as 95%, and wheat straw plants as high as 99%, depending on age category and the relative pixel area of plants within the image. The ZGK algorithm could be potentially useful for remote sensing, mapping, crop management, weed, and pest control for precision agriculture. (C) 2003 Elsevier B.V. All rights reserved.