Research On Crop Fine Classification Based On Feature Selection And Machine Learning

The North China Plain(NCP),a major agricultural area in China,plays an important role in China ’ s grain production.Timely and accurate crop information for NCP is very important to China’s food security and sustainable development.Due to high variability of the temporal profiles of vegetation indices,classification models using temporally aggregated remote sensing data often exhibit suboptimal performance for multi-crop classification in the NCP with complex cropping patterns.To solve the problems,it is necessary to combine multi-source remote sensing data to construct a multi angle classification feature set,search for the optimal feature set and classification model suitable for the region,and achieve efficient and accurate crop recognition.In this study,we used Sentinel-2 imagery to map the distribution map of main crops in Handan,a typical winter wheat production city in the southern North China Plain,in 2020 and2022.The NDVI time series,texture,phenology,topography and red edge time series characteristics of main crops in the study area were taken as input characteristics,and supervised and unsupervised feature selection methods were used for feature optimization.Then,three machine learning classifiers,random forest,support vector machine and artificial neural network,are used to evaluate the classification accuracy of different feature sets.The main conclusions are as follows:1)Combining multi-source remote sensing features can improve the accuracy of crop classification.Combining the results of the two types of special selection,texture features and time series features contribute significantly to crop recognition,followed by phenological features,while the importance of terrain features lags behind.Unsupervised feature selection algorithms obtain similar feature selection results to supervised selection without prior knowledge,providing new solutions for areas where sample information acquisition is difficult.In addition,by comparing the feature selection results before and after the addition of red edge time series features,the importance of red edge features in fine crop classification was verified.2)Remote sensing features of key temporal phases contribute more to crop identification.For the five main crops in Handan City,there are four key phenological periods,including the winter wheat rejuvenation period from February to March,the tree rejuvenation period from April to June,the main crop development period from July to August,and the crop maturity period from September to November.The phenological and texture features from February to March and April to May are the most effective features in crop type classification in this region.3)Different classifiers have different abilities to analyze multi-source features.In addition,the classification accuracy of random forest preferred feature set is almost the same as that of the full feature set in the experiment.The combination of feature selection and machine learning methods can be effectively applied to agricultural areas with complex planting patterns.In the final 10 meter resolution crop distribution map,the overall accuracy of the main crops in 2020 was 93%,with a kappa coefficient of 0.9.The overall accuracy of major crops in 2022 is 90.8%,with a kappa coefficient of 0.89.