Research On The Classification And Recognition Of Remote Sense In Southern Rice Based On RF-MDA Algorithm

In southern China,the degree of fragmentation of cultivated land is high.the planting structure of crops is more common,and the factors such as the complexity and diversity of the planting structure of crops seriously affect the accuracy of remote sensing identification and area estimation of southern crops.At present,some scholars use multi-temporal images for feature extraction and classification based on crop phenology,although they can improve crop extraction accuracy.However,superimposing temporal phase feature images and blindly extracting feature types can easily lead to Hughes phenomenon in dimension disasters.The RF-MDA feature selection model adopted in this paper can not only select feature optimization subsets from a large number of feature parameters to solve the dimensional disaster problem,but also improve the classification accuracy compared with the traditional SVM-RFE-CV feature selection model.This article takes Nanchang Jinxian as the experimental area.According to the phenological characteristics of the lower-season rice in Jinxian County and the image quality of the experimental area,the WFV multi-spectral image?16m resolution?of the Sanjing GF₁ was selected to construct a time-series image collection.The feature set data set and the spectral feature set are extracted based on the time series image set.SVM-RFE-CV model and RF-MDA model were used to extract the SVMRFE-CV model feature selection subset and the RF-MDA model feature selection subset from the feature type data set.The four datasets were used to classify rice and compare the accuracy of the results.Feasibility of Rice Extraction in Southern China Using RF-MDA Characteristic Selection Models.According to the accuracy of the classification results,the main conclusions are as follows:???Extracting and selecting multiple features of image data can improve the classification accuracy.Two feature selection model comparison experiments were conducted in this paper:In the four groups of rice classification,Based on the RF-MDA model,the overall accuracy of the optimized subset is 95.03%;The Kappa coefficient is 95.03%.This set of experiments is the highest in the four groups.The experimental results show that the RF-MDA model is feasible in the classification of southern rice.???In southern rice extraction,spectral characteristics and vegetation index characteristics are the two most important features of all feature types.There are 8 features in the SVM-RFE-CV model feature optimization subset,among which there are 3 spectral features and 4 vegetation index features.There are 40 features in the optimal subset of RFMDA model features.Among these features,there are 12 spectral features and 11 vegetation index features.The two feature types have the highest number of all the classification parameters.