Extraction Of Crop Planting Areas Using Rapideye Remote Sensing Image

The information of the sown areas and outputs of the crops is the main basis to formulategrain policies and economic plans in our country. The sown areas of crops can reflect thesituation of the use of agricultural production resources in the space, which are the effectiveways to understand the important information of the types and the distribution characteristics ofagricultural products and also the basis for the adjustment of agricultural structure. It is of greatsignificance to timely comprehend the sown areas, growth and yield of crops, which contribute sto strengthen crop production management and assist the government departments to formulatescientific and rational food policy. The traditional yield estimation approaches mainly make useof statistical methods, according to the statistics obtained from bottom to top. The yield dataacquired is not only of low efficiency and large human influence but also cannot guarantee theinformation accuracy. Remote sensing with the characteristics of large coverage area, shortdetection cycle, large amounts of data and low cost provides a new scientific method to estimatecrop yield fast and accurately. The paper applies remote sensing method to attempt to estimatecrop area fast and accurately in Wu Cheng.The paper takes Wu Cheng as the study area, using RapidEye image and land use statusvector as the main data source. It estimates the crop area of the study area to use object-orientedand knowledge rules combination classification method based on eCognition, ENVI and ArcGISsoftware platform. Firstly, in order to build all kinds of objects, the paper uses the RapidEyeimage pixel’s brightness, texture, color and other characteristics in the study area based on theobject-oriented multi-scale image segmentation theory, and then create a variety of knowledgerules including spectral features such as NDVI for non-vegetation extraction, and finally removethe interference of linear features and correct the obvious wrong polygons through visualinterpretation. As a result, the crop distribution map and crop area estimation are obtained.In order to prove the feasibility of this research method, the paper uses the method ofrandom sampling to extract the object to test the accuracy, randomly selecting 300 test points inthe study area to record their feature types. The results obtained from object-oriented method arecompared with that of maximum likelihood classification. The results show that, the use ofobject-oriented classification based on knowledge rules has higher extraction precision.The paper studies conclude as follows:First, object oriented classification is more suitable for wheat area extraction than maximumlikelihood classification. Object oriented results are closer to artificial digital results, but usemuch less time. Object oriented classification is based on objects which are relative homogeneity,while maximum likelihood classification is based on pixels which exists foreign body in thesame spectrum and the same content in different spectrum phenomenon. Object orientedclassification reduces influence of noise and overcomes the salt and pepper phenomenon, at thesame time using many characteristics such as spectral, spatial, context and so on, so it has highclassification accuracy.Second, the results between object oriented classification method and maximum likelihoodclassification have great difference, but the distribution characteristic is almost the same. Third, object oriented classification method combined with GIS auxiliary data can improvethe classification precision. In the paper, object oriented classification plus vector data in thestudy area based on block segmentation and classification, can keep the object not broken andscattered and be able to maintain the integrity of features, resulting in a high precision.Fourth, phase selection for object oriented classification method is of great significance.The study on the extraction of cotton and corn cannot reflect the true conditions of planting.There is a large deviation. It is easy to confuse with surrounding crops such as cotton and corn.Fifth, it is crucial to choose the right segmentation scale for object oriented classification.Segmentation scale selection is the starting point of the segmentation. Its choice is closely relatedto the classification results.Sixth, in the paper, the application of eCognition software offer vector polygons andattribute tables in favor of the graphs and provides a very good opportunity for furtherintegration with GIS.