Information Extraction Of Continuous Cropping Of Cotton Fields In Alar Reclamation Area Based On Satellite Remote Sensing Image Classification

Cotton is one of the most important crops in the world.It is not only an indispensable raw material in modern industrial production but also a necessity for people’s daily life.As the most critical cotton-producing area in China,Southern Xinjiang occupies a pivotal position in the national cotton industry.Alar Reclamation Area is a representative producing area in Southern Xinjiang.Acquiring the spatial distribution information of cotton planting quickly and accurately is of great significance to estimating planting area and yield.It affects the formulation of national and local government policies.Remote sensing technology is an essential means of agricultural information monitoring,and its advantages of high efficiency,accuracy,and speed have shown great potential in crop identification and classification applications.There are few specific monitoring applications for cotton in southern Xinjiang,and there are few reports on long-term,large-scale monitoring and identification.This paper takes the Alar reclamation area as the research object uses the Landsat series,Sentinel series,GF series multi-source remote sensing data as the data source.Moreover,it combines the field survey data of the two core experimental areas in the research area to create a self-made crop label dataset using the support vector machine(SVM),random forest(RF),neural network(NN),and other methods to explore the critical parameters of cotton-specific identification and classification,such as the best classification time phase and the best classification spatial resolution,and analyze the advantages and disadvantages of different data sources,vegetation indices and classification methods,and finally make a Digital mapping of the spatial distribution and dynamic changes of cotton planting area in from 1990 to 2020.The main conclusions of this paper are as follows:(1)Through field investigation on the crops planted in 2018 in the two core research areas of Alar Reclamation District,the actual survey area was 313.8km~2,and 12,028 labels in 13 different crops were obtained.One vector label dataset and four raster datasets with different spatial resolutions were produced to meet the basic data needs of this research.(2)Through the analysis of the differences in the crop spectral curves of different sensors in the core test area A in July 2018,the cotton spectral curves are quite different from other crops in the"red-near-red"band.The cotton spectra between different sensors The curve consistency is good,and the MSI sensor has 3"red edge bands"added.It can better describe cotton spectral details.(3)The normalized difference vegetation index(NDVI)and simple ratio index(SR)index of cotton in2018 were calculated using MSI and OLI sensor data,and the vegetation index time series curve was calculated using the time series harmonic analysis method(HANTS).And the Savitzky-Golay method(SG)for filtering.The SR time series smoothed by HANTS filtering can better distinguish cotton from other crops in the time-series dimension.The best time to distinguish cotton is the seedling stage,followed by the post-harvest conditioning period and the flowering period.(4)Through the exploratory analysis of the 2018 annual data of different C-SAR sensor data,cotton can be distinguished from other crops on the time series scale.Due to the limitation of its sensor characteristics,the cotton time series curve has evident noise and cannot be harvested.The classification effect is perfect.The best time to use C-SAR data to distinguish cotton from other crops is May 4th(seedling stage),followed by August 14th(flowering stage).(5)Through the analysis and comparison of different data labelling modes,spatial resolutions and classification models,the one-hot encoding mode can obtain the best recognition effect at 15m spatial resolution,and the cotton recognition accuracy is 99.5%.At the same time,three different models were trained with 70%of the data labels in the self-made dataset at a spatial resolution of 30m.The overall recognition accuracy of the final BPNN model was 83.2%,and the recognition accuracy of cotton was 94.3%.(6)Using the time series data of Landsat series sensors from 1990 to 2020,the spatial distribution of cotton planting in the Alar Reclamation Area was identified,classified and counted.A digital map of the continuous cropping duration of cotton in the Alar Reclamation Area from 1990 to 2020 was finally drawn.