Research On Crop Classification Method Based On GF-1 Remote Sensing Image

The basis for agricultural monitoring and yield estimation could be provided by rapid acquisition and accurate analysis of crop planting structure.It has important practical significance for guiding agricultural production,formulating agricultural policies,and adjusting agricultural planting structure.The extraction of traditional crop planting structure is mainly based on statistical reports and sample surveys.Problems of strong subjectivity,poor timeliness,and low operating efficiency are discovered by these methods.Large area crop classification extraction techniques were provided through remote sensing techniques.It has the characteristics of wide coverage and fast speed.Satellite remote sensing has unique advantages.Image data with different time and spatial resolution is provided.It has been widely used in the extraction of crop planting structure.Compared with other satellites such as Landsat and Sentinel-2,domestic GF-1 has higher spatial resolution and shorter revisit period.It has broad application prospects in crop planting structure extraction.In this research,the GF-1 remote sensing image was used as the data source.First,the vegetation index time series curve of NDVI,EVI,RVI and NDWI were constructed based on the field investigation and experimental data.The phenological characteristics of crops and spectral information were combined.The vegetation index with distinct distinction in different periods was selected as the feature set of classification.A support vector machine(SVM)model and a random forest(RF)model based on the pixel level were established.The applicability and recognition accuracy of the two classification models for crop classification were compared.Then,the adaptive mutation particle swarm optimization(AMPSO)algorithm was used to optimize the SVM classifier.Finally,an AMPSO-SVM classification model was constructed.And the classification effect of SVM model before and after optimization were compared.The research results of this article were as follows.(1)The NDVI,EVI,RVI and NDWI vegetation index time series curves of typical crops in the study area were constructed based on the GF-1 remote sensing image.By the time series combination of the 4 vegetation indexes,the dynamic change process of various objects could be description,and the objects could be effectively distinction.According to the changes of the vegetation index of various crop in different months,the 24 vegetation indexes spectral features corresponding to the 6 scene images from May to October were extracted as the feature set of the classification.(2)Two classification models SVM and RF were constructed.The classification of crops in the study area was achieved based on the extracted vegetation index features and combined with ground measured sample data.The results showed that the overall classification accuracy of the SVM classifier was 90.91% and the Kappa coefficient was 0.8851.The overall classification accuracy of the RF classifier was 93.58% and the Kappa coefficient was 0.9188.Generally speaking,RF classifier had a high classification accuracy,It was superior to SVM classifier in mapping accuracy and user accuracy for various crops.(3)The SVM model optimized based on the AMPSO algorithm was constructed.In order to solve the effect of hyperparameter setting on the classification effect of SVM classifier,The AMPSO algorithm was proposed to optimize parameters of SVM.The results showed that the classification effect of the AMPSO-SVM model was improved,the optimal kernel parameter was determined to be 0.135 and the optimal penalty factor was 221.67.The overall classification accuracy of the final model was 94.39%.It was 3.48% higher than the overall classification accuracy of the SVM classifier.The Kappa coefficient was 0.9287.It was increased by 0.0436.Compared with the RF model,the AMPSO-SVM model improves the overall classification accuracy by 0.81% and the Kappa coefficient by 0.0099.