Proceedings Article | 30 August 2013
KEYWORDS: Imaging systems, Imaging spectroscopy, Target recognition, Analytical research, Databases, Charge-coupled devices, Agriculture, Data storage, Statistical analysis, Reflectivity
In order to identify and classify horticultural crops rapidly, it is great importance of selecting effective characteristic wavebands from a large number of imaging data. Extracting effective characteristic wavebands can nearly represent holistic information of the research target from rich imaging spectral data, mainly used for rapid characteristic identification, cluster analysis and establishing database, especially in high spectral sensing to recognize targets for a long distance. The experiment chooses radish leaves and rice leaves as research samples so as to obtain spectral information from the surface of samples by interval of 5nm based on LCTF imaging; then, the standard deviation and correlation coefficient of the gray images are calculated for these two kinds of leaves; next, we calculate the value of waveband index according to standard deviation and correlation coefficient, and extract the effective characteristic wavebands for radish leaves and rice leaves through the sorting of waveband index. By those, the experimental results show there are six ideal wavebands at 530nm, 550nm, 555nm, 715nm, 510nm and 565nm for radish leaves, 645nm, 675nm, 685nm, 670nm, 690nm and 660nm for rice leaves separately. Further, according to the principle of Euclidean distance, we also give an assessment of classification accuracy for these two samples by comparing characteristic wavebands with full wavebands, and the classification accuracy of radish leaves and rice leaves is 80.00% and 86.67% respectively. Therefore, choosing these wavebands can be used as effective characteristic wavebands for radish leaves and rice leaves.
