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 combined statistical-soft computing approach for classification and mapping weed species in minimum -tillage systems
This dissertation describes a combined statistical-soft computing approach for classifying and mapping weeds species using color images in minimum-tillage systems. A new unsupervised separation index (ExGExR) is introduced to distinguish plant canopies from different soil/residue backgrounds. Results showed that ExGExR was significantly improved for all species and all three weeks over the previously published excess green (ExG). ExGExR performed very well for separating both pigweed and velvetleaf from bare soil and corn stalk backgrounds during the first and second week after crop emergence. A new algorithm for individual leaf extraction was introduced based on fuzzy color clustering and genetic algorithm. Images of green canopies were segmented into fragments of potential leaf regions using clustering algorithm. Fragments were then reassembled into individual leaves using genetic optimization algorithm. The algorithm performance was evaluated by comparing the actual number leaves automatically extracted with the number of potential leaves observed visually. An overall performance of 75% for leaves correctly extracted was obtained. Elliptic Fourier method was next tested for characterizing the shape of hand selected young soybean, sunflower, red root pigweed, and velvetleaf leaves. Discriminant analysis of these shape coefficients suggested that the third week after emergence was the best time to identify plant species with a correct classification average of 89.4%. When leaves from the second and third week were analyzed a correct classification average of 89.2% was reached. An unsupervised method for plant species identification was finally tested. Elliptic Fourier descriptors not only provided leaflet shape information, but also a lamina boundary template, controlling where textural features were computed. Each lamina shape extracted was corrected such that all leaflets had the same orientation for texture extraction. SAS PROC DISCRIM procedure was performed to build a species classification model using selected Fourier coefficients and local homogeneity and entropy texture features. An overall success rate of 86% was obtained for plant species classification.
Camargo Neto, Joao, "A combined statistical-soft computing approach for classification and mapping weed species in minimum -tillage systems" (2004). ETD collection for University of Nebraska - Lincoln. AAI3147135.