The Disquisition On Crop Classification Based On Temporal And Spectrum Information

Crop classification is a significant topic among remote sense application. Many domestic and foreign scholars have studied remote sensing classification algoriths. Due to diversity of classification and complexity of remote sense data, no algorithmcan be applied without limitation at the present time. In addition, crop classification from remote sense image becomes more complexible because of pixel mixture. Crop classification was sometimes far from perfect precision, if only spectral data were employed. Consequently, many classification algorithms were brought out. These algorithms can improve classification precision. Among these algorithms, some are difficult to realize due to inconvenience or complexity. Some of them are still at researching period and need more validation. It is important that the function of temporal information was ignored in these classification algorithms. These classification algorithms didn't integrate temporal and spectrum information. The crops' spectral characteristics of remote senseimage are similar if their growth stage are almost the same. The temporal and spectral information has influence upon classification precision, and this can't be ignored in crop classification. Therefore, crop classification based on temporal and spectral information can take full advantage of phenological growth stage difference and spectral characteristic, and avoid the intersection of the different crops growth stages. Consequently the classifications can overcome the interference factor among different crops and avoid the phenomenon that the same crop has different spectrum characteristics and different crops have same spectrum characteristics. And crop classification can be improved based on temporal and spectrum information. Crop classification based on temporal and spectral information was studied using different remote sense data and different classification methods in the thesis. And precision evaluation was carried out for the classified image by confusion matrix, the result confirmed the feasibility of this method. Different remote sense data were acquired for different research region. Different classification methods were designed and applied according to the difference of remote sense data and research region. For the studying region of Beijing, multi-temporal TM images were selected for crop classification. The widely planted crops were classified through the algorithms of the decision tree classification based on the multi-temporal and multi-spectral images in Beijing region. For the studying region in Heilongjiang, the temporal series MODIS data were ordered. In order to remove cloud contamination and improve the quality of the Modis Normalized difference vegetation index (NDVI) data, the temporal series NDVI data were decomposed and reconstructed by the Harmonic Analysis of time Series (HANTS) algorithm. This thesis also designed and tested a classification method to combine different spatial resolution images. The feasibility was verified by the main crop classification image of Heilongjiang region. According to characteristics of different resolution image, the spatial advantage of high space resolution image was utilized to improve middle resolution image quality.