Method Study On Extraction Of Farmland Ecosystem's Information Based On High-resolution Remote Sensing

Indicators such as coverage,type configuration and spatial distribution of farmland ecosystem play great roles in ecosystem services and environmental quality assessment.Accurate estimation and extraction of farmland ecological information can support the construction of "precision agriculture" in the aspect of data.With the advancement of observation technology,the spatial resolution of remote sensing has been continuously improved,making extracting fine ecological information of farmland possible.However,owing to the complicity of background and great diversity of ecosystem configuration resulted from the combination farmland ecotypes and the natural environment,the accurate extraction of farmland ecological information based on high-resolution images is subject to the severe problem of "the same object with different spectra characteristics and different object with the same spectra characteristics".High-resolution images not only provide spectral information,but also inform much on shapes,textures and context,etc.The combined application of these features can improve the accuracy of extracting agricultural ecological information enormously.In this case,the traditional method of extracting the impervious surface information of middle and low resolution images based on spectral information is far from enough to meet requirements of high-resolution images.Therefore,starting from the description of the spectral and spatial features of remote sensing images integrated,this paper explores the atlas features of images comprehensively with a classification model of cooperative utilization of atlas built to extract high-resolution farmland ecological information at regional scale precisely.The main contents and innovations of this paper are as follows:(1)A multi-scale and multi-level remote sensing image segmentation approach for farmland ecosystems based on mathematical morphology is proposed for the purpose of addressing the serious problem of "over-segmentation" and"under-segmentation”,as the traditional single-scale segmentation approach fails to meet the needs of multi-scale segmentation of complicated farmland ecosystem surface features.This paper proposes a multi-scale and multi-level segmentation approach based on mathematical morphology to integrate the pixel-level image classification results with the first-level large-scale image segmentation results,to assist in segmentation of the second-level multi-scale(different surface features)images and to apply suitable segmentation scale parameters of self-adaption surface features,with the requirements of different surface feature segmentation scale parameters met.According to the experimental results,compared with the single-scale approach,the undersegmentation rate of the multi-scale segmentation method based on relative entropy(KL)divergence is between 12.5% and 13% at different scales,the approach proposed in this paper can access to more homogeneous and complete patches of surface features,which is conductive to the extraction of morphology and features of feature objects at various scales and the identification of subsequent attributes.(2)A joint feature extraction approach based on remote sensing subject indicators,color correlogram and gray-gradient co-occurrence matrix is proposed;moreover,the original image features are extracted further by means of embedded feature selection in the category of feature extraction.The objective is to space dimensionality of features effectively and eliminate possible correlations between features,to decrease redundant information in features without changing the original feature space and feature values.With the original surface feature information retained,the construction of a forecast and classification model with less time and memory consumed can be secured.Research results show that the Contourlet transform and the final match point number changes very little SURF combination algorithm,the matching accuracy remain above 94%,proved based on the wave of Contourlet transform and accelerate the steady characteristic(SURF)the combination of remote sensing image feature extraction algorithm can achieve high scores of farmland ecosystem characteristics of remote sensing mapping accuracy and rapid extraction.(3)A high-resolution farmland ecosystem sorting approach based on improved SVM algorithm is proposed.In view of the shortcomings of the traditional vector superposition multi-feature fusion method in the application of farmland surface classification with high spectral heterogeneity,this paper,starting from the collaborative utilization of the spectral and spatial characteristics of remote sensing images,apply the improved SVM algorithm with Remote Sensing Based Ecological Index(RESI)integrated to the extraction of farmland ecological information;moreover,with the function reflecting the sample-feature space obtained by studying the labeled samples of the existing images,this paper leverages atlas feature combinations and application modes that are conducive to the current image classification profoundly to achieve high-precision classification of ecosystems,so as to extract accurate farmland ecological information.According to the experimental results,the proposed approach can improve the classification accuracy of ecosystems significantly and lay a solid foundation for the subsequent high-precision extraction of farmland ecological information.Experimental results show that based on sequential minimal optimization(SMO)algorithm and the ecological index(RSEI)of the remote sensing classification of support vector machine(SVM)optimization algorithm the overall accuracy is 86.07%,the Kappa coefficient 0.752,field validation figure spot extraction accuracy is 95%,the number of types of feature extraction accuracy is 86.13%,extraction area of accuracy is 85.89%,prove that the method can improve classification accuracy of farmland ecosystem,for after the implementation of farmland ecosystem information extraction,management,and analysis of comprehensive study to lay a solid foundation.(4)A set of Spark-based processes for distributed extraction of farmland ecosystem information is proposed.Owing to the great diversity of configuration and complicity of background of farmland ecosystems,the high-resolution and precise farmland ecological information extraction algorithm is rather sophisticated and incapable of meeting requirements of extracting farmland ecological information at the regional scale rapidly and accurately.With Spark,a cloud platform in the computer realm,applied to extracting high-resolution remote sensing farmland ecological information,this paper,focusing on the characteristics of remote sensing images and high-resolution farmland ecological extraction algorithm,designs and implements a Spark-based automatic partitioning and merging strategy for high-resolution images and a work-flow system of high-resolution farmland ecological information extraction algorithms.Last but not least,distributed and rapid extraction of high-resolution farmland ecological information is achieved by means of partitioning remote sensing image data or data sets.According to the experimental results,this approach proposed in this paper can increase the extraction rate of high-resolution farmland ecological information significantly without affecting the accuracy.The experimental results show that the distributed processing of overall accuracy was 84.27%,the Kappa coefficient of 0.7433,shows the image in the process of data divided by overlapping data partition is to a certain extent resolve the calculation problem of the object pixels,the proposed approach can effectively expand the object of study of space and time scale and greatly improve the rate of information extraction,and will not reduce the accuracy of information extraction.It is beneficial to the ecological orientation and network research methods of long-term research.