Study On Extraction Of Plant System Of Paddy Rice From MODIS Data

Agriculture is the foundation of the development of a nation, and rice is one of Chinese most important food crops. Hold on the red line of1.2million hectares of croplands has been an important measure to ensure food security in China, paddy fields is particularly important. With the acceleration of the process of the country’s population growth and urbanization, there is an ever-increasing need in establish a long-term, effective and reliable monitoring system of rice to ensure food security and promote social stability. Rice monitoring system including the area of rice planted, the phenology of rice, the production of rice and the insect pests of rice. With the development of remote sensing technology and information processing technology, the combination of remote sensing data and information processing software make the large-area rice monitoring achieved.MODIS data detected by optical sensors, it has a high time resolution but a lower spatial resolution. What’s more, there has been a challenge that MODIS data is vulnerable to the areas covered by clouds. China has the large population, the few per capita croplands and there is a few croplands under modern management, the use of croplands is very complex. All of above, make it difficult to monitoring the rice by remote sensing. However, with the development and popularization of information processing technology, agricultural monitoring has a very significant effect combined with information processing technology.This paper is aimed to extract the rice paddy area and rice phenology of Jiangsu provinces using the signal separation technology and a variety of signal filtering methods based on multi-temporal MODIS imagery. To investigate the reliability of similarity index algorithm in dealing with rice paddy pixels extraction and the efficiency of Fast Independent Component Analysis (FastICA) algorithm in dealing with mixed pixels. In order to map the rice area, time series of Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI) were decomposed by FastICA algorithm and the sources of inaccuracy has been discussion. The paper also filtered EVI data with HANTS (Harmonic Analysis of Time Series) filtering and wavelet filtering, and then choose wavelet filtering combined with Matlab software to extract the phenology after compared the result of HANTS filtering with wavelet filtering. After all, investigate the accuracy of the result.In conclusion, the main development and results as follows:(1) The rice pixels extracted by similarity index algorithm match the actual distribution of rice, but there are also some limitations of this algorithm which bring inaccuracy of the result.(2) The effect by decomposition of the FastICA algorithm is very obvious which can decompose different rice growth curve of NDVI/EVI under different conditions. The method of abundance calculation works well. According to the analysis and comparison with the statistic data, the average accuracy of the obtained rice area is91.1%by EVI data and86.4%by NDVI data.(3)The inaccuracy of the process of abundance calculation has a relationship with the number of rice growth curve obtained by FastICA algorithm. The more of the numbers, the more inaccuracy of the result which is calculated by add all the results of different rice growth curves. Using only one curve to represent the whole area by reduce the Virtual Dimensionality can reduce the inaccuracy when the number is big.(4)Wavelet filtering is more effective than HANTS filtering in cloud-contaminated areas. The new method which extract the transplanting date and the wavelet filtering method which extract the other rice phenology can accurately reflect the real rice phenology.