| Summary: | 碩士 === 國立臺灣大學 === 生物產業機電工程學研究所 === 89 === In this research, a multiresolution wavelet analysis with accumulative multi-dimensional decision (AMDD) model was used to classify the species of crops. The classification was based on weighting Bayes distance. The distance was derived from the dominant features extracted from the energy of feature images, and the weighting was determined by the geometry of leafs in crop images.
To eliminate the variation of the energy influenced by factors such as climate, plantation density, spread of leafs, planting stage and orientation of sunshine, a run length histogram from the geometry of leafs in a crop image was developed for the similarity estimation between crops. The weighting of Bayes distance was calculated based on the rum length histogram.
An accumulative multi-Dimensional Decision algorithm was devised for the robustness of classification. It effectively reduce the categorization error due to the variation of dominant features within the same crops.
The result of experiments showed that the classification accuracy for ten species of crop images acquired in three successive days under different circumstance was up to 97.2% by using the developed approach. In this study, we also investigated different process and parameter, including equalization, filtering, weighting, DC offset and proportional constant of weighting, in terms of the classification accuracy. The developed algorithm was proved to be very effective for the recognition of crop species.
|
|---|
