Mapping Summer Maize By Using Spatiotemporal Characteristics Of Backscattering Coefficients

Summer maize is an important food crop and feed source.Timely and accurate acquisition of planting area can provide scientific basis for relevant departments to formulate food policies and adjust agricultural planting structure.The existing methods of mapping summer maize based on optical images could be easily affected by cloudy and rainy weather,it is hard to obtain continuous time series data.Moreover,these methods mainly focus on small regions(city-level and below),only considering the change of backscattering coefficients of summer maize over time and ignoring their spatial differences,which cannot be fully applicable in large scale(province-level and above),and the classification accuracy needs to be further improved.In order to solve these problems,this paper took Henan Province as an example,with the help of multi-temporal dual-polarization Sentinel-1A SAR images,calculated the backscattering coefficients of summer maize in the whole growth cycle,then analysed their spatial and temporal characteristics,and studied a method suitable for mapping summer maize in large scale.The main research contents and results are as follows:(1)Analysis of the spatiotemporal characteristics of backscattering coefficients of summer maize.In terms of time,the backscattering coefficients of summer maize first increase and then decrease with the growth and development of crops.They vary in different periods,with time differences.In space,due to the different climatic conditions and sowing time,the change rules of backscattering coefficients of summer maize in different regions are not the same,with spatial differences.(2)A comparison of three classification algorithms for SAR images.RF(random forest),SVM(support vector machine)and GBDT(gradient boosting decision tree)were used to classify summer maize,and their classification results were compared.It is found that SVM has the lowest classification accuracies and the slowest running speed.GBDT runs fast,but the user accuracy is low and the classification effect is unstable.In contrast,the classification accuracies of RF are maintained at a high level,the classification effect is the most stable,and it is easy to set the parameters.So the RF is chosen as the basic algorithm for summer maize classification.(3)An algorithm for SAR image classification,considering the spatial neighbourhood features,was studied and implemented.In order to reduce the influence of speckle noise on the classification results,a classification algorithm RF-MRF was proposed by combining RF with Markov random field(MRF),which took into account the spatial neighbourhood features.The classification accuracy of RF-MRF is significantly improved compared with the basic algorithm.The producer accuracy of summer maize is up to 95.78%,and the user accuracy is up to 93.81%,which shows that the algorithm has excellent classification effect and can be widely used in SAR image classification.(4)Based on the spatiotemporal characteristics of backscattering coefficients,the planting area of summer maize in large scale was extracted.Based on the spatiotemporal characteristics of the backscattering coefficients of summer maize,the classification algorithm RF-MRF was used to map the summer maize in the whole Henan Province in 2020.The final extraction accuracy is 92.25%,which verifies the feasibility and practicability of this method.In this paper,the spatial and temporal characteristics of backscattering coefficients of summer maize are used as the basis for spatial division and optimal phase selection for mapping summer maize in large-scale.Through the experimental comparison of three commonly used classification algorithms,it is found that the RF performs the best in classifying summer maize.Then based on RF,a classification algorithm RF-MRF which takes into account the spatial neighbourhood features is proposed,and further improves the classification accuracies.Finally,through the experiment of mapping summer maize in the whole Henan Province,the feasibility and practicability of the method proposed in this paper are proved.This study can provide technical support for relevant departments to grasp grain planting structure in time.