Sugarcane Growth Monitoring Based On Multi-source Remote Sensing Images And Time Series Analysis

Sugarcane crop has the properties of both sugar and energy raw materials,and is one of the nine major crops in China.The related industrial chain around sugarcane plays an important role in agricultural economy.As the country attaches great importance to modern agriculture,it is of great significance for sugarcane yield estimation and the formulation of the government’s agricultural subsidy policy to use remote sensing technology to carry out accurate and real-time growth monitoring of sugarcane.Based on the planting of sugarcane crops and the actual climate and weather conditions in South China,aiming at the problem that some classical remote sensing parameters in the field of remote sensing crop inversion are not sensitive to the change of sugarcane growth,by preprocessing,matrix transformation and Cloude-Pottier target decomposition of the 23-scene time-series dual-polarization Sentinel-1A radar remote sensing images covering the whole growth cycle of sugarcane,The Dual-Pol Radar Vegetation Index(DPRVI)is obtained.In this paper,the dynamic changes of DPRVI and sugarcane growth parameters(plant height)with sugarcane growth are analyzed,and DPRVI and three classic remote sensing parameters are compared.The results showed that the performance of DPRVI was better than the other three parameters,and it could better reflect the change of sugarcane in different growth stages.Then,four classical empirical regression models(linear,quadratic polynomial,exponential,logarithm)were used to invert the plant height of sugarcane at different growth stages in the form of piecewise-function,and the best inversion model was established.The experimental results showed that the fitting model had the highest correlation before tillering stage,and the fitting effect of quadratic polynomial model was the best.The correlation coefficient R~2 and root mean square error(RMSE)reached 0.882 and 0.118,respectively.It was verified that DPRVI had a good performance in sugarcane growth monitoring,and provided a simple and effective method for the inversion of plant height during the whole growth period of sugarcane,which could provide more accurate reference information for the monitoring of sugarcane growth situation.Aiming at the problem of accurate identification of sugarcane planting areas in a large range,based on the preprocessed of three time phases Sentinel-2A data,the NDVI,BI2 and S2REP vegetation index features were calculated,and the optimal Homogenity texture features were selected through principal component analysis and gray level co-occurrence matrix.Then,based on the recognition model of the random forest algorithm,different combinations of single-phase single-feature,single-phase multi-feature,multi-temporal single feature,and multi-temporal multi-feature were established respectively,and the comparative experiment of sugarcane planting area identification was carried out in Nansha District,Guangzhou.It was found that the sugarcane planting region integrated with NDVI,BI2,Homogenity and temporal features had the best identification result.The boundaries between different ground features were clear,and the phenomenon of salt and pepper noise in the image was well controlled.The overall accuracy reached 0.974 and the Kappa coefficient reached0.931.In the result verification,the identification accuracy was 81.6%in Nansha District of Guangzhou City,which reached the expected research goal,and the errors generated in the experiment were analyzed.The proposed method is suitable for the identification of sugarcane planting areas in the Pearl River Delta region,and can provide effective reference information for the agricultural sector to estimate sugarcane yield and forecast price trend.