Study On The Extract Methods Of The Early Rice Planting Area Based On The Mesoscale Rs Image

The early rice is one of the important components in the grain production of the country. And more importantly, its production stabilizes the planting of double-cropping late rice, which has great significance to promote grain production and ensure food security. Currently, the planting area is being shrinked in some regions because of the subject and object factors. Accordingly, it is crucial to investigate the area with RS technics, hold the background and analyse the dynamic changes and its causes. Then we can nail down the responsibilities and carry out measures to level off and impulse the production of the early rice.For remote sensing, different resolutions have different applicabilities. For the moment, researches of rice planting area based on large scale RS datas such as NOAA and MODIS account for a higher proportion. But the precision is restricted by their low resolutions and mixed pixels. It is feasible to monitor the planting area of the grain with mesoscale RS images such as CBERS, TM, SPOT, LISS and so on. The images which have less cloud and multi bands can satisfy the requirements entirely. Presently, some methods like pixel un-mixing, neural network, SVM, and evidential theory have enhanced classification accuracy to some extent, with defects of complex algorithms and hard maneuverabilities, they still can not be popularized. So it is necessary to adopt appropriate means contraposed different areas.The data source of RS used in the paper is IRS-P6 LISS3 with mesoscale resolution. Considering the limitation of the RS data, the article combined with maps of paddy field distributing, land use and the data of GPS investigations in the field to idenify and extract the information of the rice planting area. The ways adopted respectively are Maximum Likelihood, Threshold Segmentation and Optimal Iteration Unsupervised Classification which are simpler algorithms and more exercisable.According to the qualitative and quantitative verifications, the conclusions are as follows:(1)After the fusion of RS and GIS datas, the precisions were advanced distinctly compared with the traditional supervised classfication. (2)In the aspect of the qualitative verifications, Classification of Maximum Likelihood has the highest user accuracy and Kappa coefficient:90.63% and 0.8919, then is the Optimal Iteration Unsupervised Classification:88.89% and 0.8701, and the lowest is Threshold Segmentation with 87. 50% and 0.8512. On the other hand, Optimal Iteration Unsupervised Classification has the best productive accuracy and then are Threshold Segmentation and Classification of Maximum Likelihood. They are 86.49%,85.37% and 85.26% separately which are increased by 9.79%,9.07% and 0.1232 accordingly.In the aspect of the quantitative verifications, the interpret outcome of ALOS AVNIR-2 with 10 meter resolution was used as the quasi true value to identify the results of LISS data. And the results are as follows:Classification of Maximum Likelihood has the least error of -6.35%, then are Optimal Iteration Unsupervised Classification and Threshold Segmentation with-6.62% and -12.10%. Compared with the traditional way, its quantitative accuracy are improved by 20.61% at least. (3)In general, the extraction based on fusion enhances accuracy of the early rice planting area. Classification of Maximum Likelihood and Optimal Iteration Unsupervised Classification which have the higher precision can be used practically as the referenced ways.