| Canola is one of the main oil crops in China,and timely and accurate acquisition of canola planting area is of great significance for agricultural production management,macro-policy control and food security.The fragmentation of canola plots in China is high,and Sentinel images with high spatial resolution have become an important data source for canola remote sensing identification.At present,the following shortcomings exist in the remote sensing identification research of canola:(1)In the remote sensing identification research of canola using high spatial and temporal resolution images on a large spatial scale,it often faces the problem of difficulty in processing massive data.(2)Some winter crops have very similar phenological characteristics,and the recognition accuracy of canola will be limited in areas with serious mixed planting phenomenon.Therefore,this paper relies on Google Earth Engine cloud platform to provide an effective solution for massive remote sensing data processing,and carries out accurate automatic identification of canola based on Sentinel optics and microwave images.(1)When the images of canola flowering period are sufficient,the canola flower index is constructed based on a single Sentinel-2 optical image to enhance the image features of canola and realize the automated mapping of canola in Hubei Province with an overall accuracy is 94.20% and the Kappa coefficient is 0.88,among which the producer accuracy of canola is 93.01% and the user accuracy is 92.69%.(2)When the optical image of canola flowering period is missing,an algorithm for enhancing canola image features is designed based on a single Sentinel-1microwave image,and the effective time window for identifying canola is determined and Sentinel-1 image is synthesized for automatic mapping of canola in Hubei Province,with the overall accuracy of 87.73% and Kappa coefficient of 0.76,among which the producer accuracy of canola is 93.37% and the user accuracy is77.29%.(3)In order to solve the problem of low user accuracy of canola when using sentinel-1 microwave image alone,a research scheme of coupling sentinel optical image is proposed.The results showed that the overall accuracy was 92.51%,and kappa coefficient was 0.86,among which the producer’s accuracy was92.16% and the user’s accuracy was 91.39%.Compared with using Sentinel-1 microwave image alone,the canola identification accuracy of coupled Sentinel-1/2 image increased by 4.78%,the user accuracy of canola increased by 14.10%,and the Kappa coefficient increased by 0.10.In this paper,the key technologies of accurate identification of canola are studied from two aspects according to whether the optical image of canola flowering period is missing or not,and the problems faced by remote sensing identification of canola on a large scale are solved.The main results are as follows:(1)The canola flower index(CFI)was constructed to enhance the image information of canola and to realize the automatic mapping of canola.Based on Sentinel-2 optical image,the Fisher values of canola and winter wheat,forest,bare land,construction land and water in different periods were calculated.The period with the maximum Fisher value is the best period for identifying canola,and accordingly,the flowering period of canola is determined to be the best period for identifying canola.Based on Sentinel-2 images of canola flowering period,we analyzing the spectral reflectance characteristics of different features and found that the NDVI value of canola is higher than that of non-vegetation(construction land,bare land and water),the sum of reflectance of canola in red and green bands is higher than that of winter wheat and forest,and the difference of spectral reflectance between green and blue bands is higher than that of winter wheat and forest.Accordingly,we constructed the Canola Flower Index(CFI),which significantly enhanced the image characteristics of canola.CFI image has only one layer,while Sentinel-2 original image has four layers with a spatial resolution of 10 m.CFI significantly reduces the dimension and data volume of remote sensing data.The threshold classification method based on CFI images improved the classification accuracy of canola compared with other canola indices.(2)Synthesize Sentinel-1 microwave image enhance the image characteristics of canola and solve the data redundancy.The flowering period of canola in Hubei Province is short in March,and the optical image of canola may be missing during the flowering period due to the influence of cloudy and rainy weather.Sentinel-1 microwave image is not affected by the weather,and it is very sensitive to the plant structure.By analyzing the time series Sentinel-1 microwave spectrum curve of canola and other ground objects,it is found that the VH band of canola rose rapidly and reached the peak around April,and then began to decline gradually,while other ground objects did not show similar characteristics.According to this characteristic,three effective time windows for identifying oilseed canola were identified,from January 19 to February 18,2020,April 8 to April 28,2020,and May 28 to June 17,2020,respectively.The mean values of VH bands were calculated within each of the three time windows,and finally the three VH bands within the three time windows were combined into one layer to obtain Sentinel-1 synthetic images,which effectively reduces data redundancy and enhances the canola image features.(3)Coupling Sentinel-1/2 active and passive images to improve the accuracy of remote sensing recognition of canola.Sentinel-1 microwave image is easily affected by speckle noise,and its actual spatial resolution is5 m × 20 m,which makes it easy to mistake the non-canola ground objects for canola at the boundary of canola plot,which limits the recognition accuracy of canola.This paper solves this problem by coupling Sentinel-2 optical images.Firstly,based on the phenological information and time series MODIS-NDVI data of overwintering crops(canola,winter wheat)and other ground objects,the effective time window for identifying overwintering crops was determined,and then the maximum NDVI images were synthesized in the high NDVI time window(December 30,2019 to March 19,2020),and the minimum NDVI image were synthesized in the low NDVI time window(October 1-October 31,2019 and May 3-June 12,2020).Based on the synthetic image and decision tree classification model,the distribution data of overwintering crops,that is,the potential distribution data of canola,are obtained.Then,the intersection operation is performed with Sentinel-1 microwave recognition results.That is,canola identification under the coupling of Sentinel-1/2 active and passive images is completed. |
PDF Full Text Request