Multi-scale Segmentation Of Crop Condition Parameters Inversion Study

To achieve the overall regional scale crop growing real-time and non-destructive monitoring of the growth of crops we use the data obtained from the large satellite-air plane-ground remote sensing in the experiments in Heihe River Basin in June to July 2012 and the observed data of the ground as the data source. The eCognition image processing software is used as a platform and the multi-scale segmentation techniques are used to extract the information from hyperspectral remote sensing image. Based on the above methods we can calculate the growing parameters LAI, Cab, N, etc. to explore the affection of multi-scale object-oriented segmentation of Hyperspectral remote sensing on crop parameters inversion, in order to obtain information about the agricultural application of hyperspectral data and experience that new results from the theory and methods. This paper mainly focuses on the following three points in depth study:First, based on the object-oriented multi-scale segmentation techniques which are combined with progressive visual interpretation and trial and error method, and under the evaluation of overlap of the relative areas to screened the optimal segmentation scale parameters, namely the weighting factor of each band is 1, the weighting factor of shape is 0.4, the tightness is 0.5 and the segmentation scale size is 70.Secondly, based on the considering of the CASI image features and priories of previous studies, we statistically describe and analysis the relativity of the growth parameters of the crop and parameters of vegetation and choose some parameters which have better inverse ability and stability to establish a multilinear remote sensing estimation model.Finally, according to the characteristics of object-oriented and multiscale segmentation image and combining with crop growth inversion model which correspond with the actual condition of the study area we contrast analysis the method which inverse the parameters after the segmentation and the method which segment the image after the inversion and this method realized the remote sensing monitoring of the whole growth on a area scale crop.The experiment results show that the method which inverse the condition of growth after segmentation is better than the method which segment the image after inversion. The accuracy of the estimation in the former method of the parameters LAI, Cab, N are all over 80% and the decision coefficient R had pass through the 0.95 reliability level of significance tests. This shows the unique advantages of the object-oriented multi-scale segmentation techniques applied to crop parameters inversion and makes the inversion results of the overall field crops more objective, accurate and general. The proposed method realized the efficient extraction of growth of crops in lager areas based on gyperspectral remote sensing image.