Method for estimating aboveground biomass of rice based on multi-spectral images of unmanned aerial vehicle

A method for estimating the aboveground biomass of rice based on multi-spectral images of an unmanned aerial vehicle (UAV), comprising:

• step 1:
• collecting rice canopy multi-spectral image data by a UAV, and randomly and destructively sampling rice plants to obtain rice plant samples;
• step 2:
• preprocessing the rice canopy multi-spectral image data, specifically comprising the steps of removing noise, correcting halation, correcting lens distortion and correcting radiation to obtain reflectivity data of a rice canopy multi-spectral image;
• step 3:
• performing data processing on the rice canopy multi-spectral image reflectivity data, specifically comprising the steps of extracting reflectivity, calculating texture feature parameters, calculating a vegetation index and calculating a texture index, and simultaneously measuring biomass data of the rice plant samples;
• step 3-1:
• extracting reflectivity of each region according to the region of interest (ROI), where the region is defined as an area of 5 m×6 m;
• step 3-2:
• calculating the vegetation index correlated to biomass estimation according to the reflectivity in step 3-1;
• step 3-3:
• calculating 8 texture feature parameters based on a gray-level co-occurrence matrix on a reflectivity image;
• step 3-4:
• constructing a normalized difference texture index NDTI(T1,T2), NDTI(T1, T2)=(T1−T2)/(T1+T2) according to a single texture feature parameter, where T1 and T2 are respectively a certain texture feature parameter at an arbitrary band;
• the single texture feature parameter is one of the 8 texture feature parameters based on the gray-level co-occurrence matrix, and a certain texture feature parameter is one of the 8 texture feature parameters based on the gray-levell co-occurrence matrix;
• step 3-5:
• rooting and washing the rice plant samples, performing fixation and drying in an oven, and weighing the rice plant samples to obtain the aboveground biomass data of the rice plant samples;
• step 4:
• performing a correlation analysis on a spectral index, the single texture feature parameter, the texture index and the aboveground biomass, where the spectral index is one of the vegetation indices calculated according to the reflectivity in step 3-2;
• at the same time, performing a stepwise multiple linear regression analysis by using a data analysis software SPSS and combining the spectral index and the texture index to obtain a rice biomass estimation model, and calculating a coefficient of determination R2;
• and evaluating performance of different regression analysis models on rice biomass estimation by comparing R2;
• step 5:
• according to field test data, verifying accuracy of the rice biomass estimation model by a cross-validation method, and calculating a relative root mean square error (RMSE):
• RMSE
• =
• 1
• n
• ⁢
• ∑
• i
• =
• 1
• n
• ⁢
• (
• P
• i
• -
• O
• i
• )
• 2
• where Pi, and Oi, are predicted and measured values of the ith sample respectively, and n is a total number of samples;
• and
• step 6:
• by comparing the RMSE of different estimation models, obtaining an optimal rice aboveground biomass estimation model:
• AGB (t ha−1)=e(4.179×NDTI(Mean800, Mean550)−2.559×NDTI(Mean800, Mean720)+1.924×MTVI2−0.644)
• where t ha−1 is a biomass unit, NDTI (Mean800, Mean55o) is a normalized difference texture index constructed by Mean texture feature parameters at an 800 nm band and a 550 nm band, NDTI (Mean800, Mean720) is a normalized difference texture index constructed by the Mean texture feature parameters at the 800 nm and 720 nm bands, MTVI2 is a modified triangular vegetation index, and AGB is the aboveground biomass of rice.