녹화면에 녹피율 산정 분석방법에 따른 정밀도에 관한 연구
김연미, 구원용, 남미아 · 한국생태환경건축학회 학술발표대회 논문집 · 2012
In ecological analysis and assessment, Green coverage is to consider as a quantitative indicator like frequency or density of the vegetation. Green coverage is closely related to the quantity of biomass, so that it indicates the aboveground storage capacity of CO2. Therefore there is a need to estimate accurately the green coverage of the analysed plot using a convenient methods like photograph imagery processing. The green coverage is a value in percentage measured by a vertical covered surface of the vegetation. In this study, a accuracy of green coverage by proposed image processing model is compared to that of optical assessment by vegetation experts, ImageJ image analysis software and manual operational calculation using CAD. As a methodological approach to recognize green spots, image processing and pattern recognition are applied. Firstly, according to KNN(K-Nearest Neighborhood) theory, the areas of plant and non-plant are differentiated optically and the masks for green and non-green are created in order to recognize between green and non-green. For model building, from the 50 applied sample image of 1 x 1m vegetation plots are derived 50 G-space image set and 50 NG-space image set which are used as KNN-learning data. The results show that the model has the 82% accuracy compared to CAD calculation while experts estimate the green coverage with ±90% accuracy. Taking into consideration that the proposed model uses exclusively RGB feature vector, it can be expected to enhance the accuracy through additional feature elements and improvement of the model.
Cited by 0 works.