Off-campus UNL users: To download campus access dissertations, please use the following link to log into our proxy server with your NU ID and password. When you are done browsing please remember to return to this page and log out.
Non-UNL users: Please talk to your librarian about requesting this dissertation through interlibrary loan.
A fuzzy logic approach for plant image segmentation and species identification in color images
This dissertation describes a fuzzy logic approach for segmentation of color images and classification of plants according to species. Models were derived using an adaptive neural-network fuzzy logic inference system. Digital images of grass, bare soil, corn stalks, and wheat straw were processed to obtain average color values and derived contrast indices. Apparent lighting source color temperature and surface illuminance were recorded for each image. Fuzzy classification accuracies were above 90%, while statistical discriminant analysis accuracies were only 50% to 60%. Rules based on individual color values yielded better classification results than rules based on color contrast indices. ^ Digital images of plants against bare soil, corn stalk, and wheat straw backgrounds were obtained under greenhouse conditions. A training set of 80 images was manually segmented and used to train the rules for two prototype fuzzy inference systems. A system based on derived color indices achieved a classification accuracy of 54% for a four-class system (plant, bare soil, corn stalks, and wheat straw). For a two-class system (plant and background), the overall accuracy rate was 99%. Using individual color values, a four-class system achieved a classification accuracy of 78%. The corresponding two-class system achieved an accuracy rate of 97%. ^ Four textural features were calculated for both plant and background regions. The feature values are based on the relative frequency of pixel intensities between neighboring pixels in the grayscale images. Because of the size of the generated system, only one feature could be modeled at a time. The fuzzy inference systems achieved classification accuracies ranging from 34% to 85%. Local homogeneity and entropy showed the most promise for developing an accurate plant species identification system. However, classification accuracy using local homogeneity dropped from 85% to 58% when image rotation was considered. The system was able to correctly classify over 90% of all plant regions as plants and background regions as background, indicating that texture can be of significant value in species identification given the proper parameters. ^
Hindman, Timothy W, "A fuzzy logic approach for plant image segmentation and species identification in color images" (2001). ETD collection for University of Nebraska - Lincoln. AAI3022633.