Research On Biomass Estimation Of Edible Forage Based On RGB And Multispectral Images Of UAV

The South Bank of Qinghai Lake in Hainan Tibetan Autonomous Prefecture is located in the northeast of Qinghai Tibet Plateau.It is not only an important base for Tibetan sheep husbandry,but also an extremely important ecological barrier.In recent years,due to the transitional grazing behavior of herders,the grassland grassland balance has been destroyed,resulting in the growth of palatable weeds and poisonous grass.This case had a significant impact on the ability of grassland stoic and hindered the sustainable development of animal husbandry.Above Ground Biomass(AGB)refers to the amount of organic matter accumulated on the unit area for photosynthesis.It can be used to characterize growth and yield estimation.As an important index of grassland livestock capacity,the biomass of edible herbivorous is an important basis for the balance of livestock on pastures.Therefore,estimating the biomass of grasslands that can feed herds can help to improve the utilization rate of pastures and achieve accurate zoning.The regeneration ability of Stellera chamaejasme is strong,and it has a strong ability of drought resistance and cold resistance.It is widely distributed in Hainan Province.In this paper,we first use the Parrot Bluegrass to collect the RGB and multispectral images of grassland.Then,image processing technology is used to automatically identify the distribution characteristics of poisonous weeds.On this basis,height and spectral characteristics are used to structure the estimation method of Achnatherum splendens.This paper mainly completes the following work:(1)Correction of UAV near earth remote sensing image data.For the problem of color deviation in RGB image,the pixel values in the areas of black,white and gray color correction plate are extracted,and then the pixel values obtained in the standard environment are used for polynomial regression analysis,so as to achieve the color correction of RGB image.At the same time,to solve the problem of spectral reflectance deviation of multispectral image,the pixel values in the area of the spectral correction plate are extracted,and then the input light intensity is calculated by combining the real reflectance information,so as to realize the spectral reflectance correction.(2)Image registration in different bands of spectral cameras.Parrot is equipped with an array structure multispectral camera.Registration in different bands is used to solve the problem that large position deviation of the corresponding pixels which in different bands of the same image.Firstly,the Gaussian filter is used to smooth the data of each band.And then the connected region method is used to extract the Region of interest in each band.Last,the AKAZE algorithm is used to solve the transformation matrix between the near-infrared band and the rest bands by detecting and matching the feature points,so as to achieve the registration target of different bands with high accuracy.(3)Image fusion with RGB images and multispectral imagesImage fusion is multi-source information fusion,which combines the relevant information of each type of image into one image to highlight the features to be analyzed.The image fusion of high-resolution RGB image and low-resolution multispectral image can combine the texture information of RGB image with the spectral reflectance information of multispectral image to obtain a fusion image with better distinguishing characteristics of poisonous grass.It can effectively improve the accuracy of identification and classification of poisonous weeds.(4)A classifier of non edible herbage(Stellera chamaejasme)was designed.In this paper,support vector machine(SVM)and random forest(RF)are used to construct classifiers to recognize and classify Stellera chamaejasme in RGB images,multispectral images and fused images respectively.For RGB images,the identification of Stellera chamaejasme by texture features is carried out.The experimental results show that the accuracy of SVM and RF classifier is 72.0%and 87.4%respectively.The reflectance feature of multispectral image is extracted as the model input,and SVM and RF classifier are used to study the recognition of Stellera chamaejasme.The recognition accuracy is 84.0%and 88.0%respectively,which shows that Stellera chamaejasme is quite different from other Herbages in spectral characteristics.The fused image shows the best recognition rate,which is 82.8%for SVM and 93.1%for RF.The recognition results of three kinds of image data show that the recognition accuracy of random forest is due to support vector machine.(5)Biomass estimation of Achnatherum splendens.Taking Achnatherum splendens as an example,a model for estimating Herbage Biomass was established.Firstly,pix4d mapper is used to obtain the point cloud data of pasture grass,and than the height of pasture is estimated based on the point cloud data.The results showed that the accuracy of forage height estimation was high in the flat area,with an error of about 3cm,while in the sloping area,the error of forage height estimation increased to 5-7cm.Height data,spectral data and height data combined with spectral data were used as model inputs to build forage biomass estimation models.The three models were evaluated by measuring the actual forage biomass in the sampling frame.The estimation accuracy was from high to low:Based on the spectral data(R2=92.3%),based on the height data(R2=89.0%)and based on the height data combined with spectral data(R2=88.0%).The combination of identification model and biomass estimation model can be used to estimate the biomass of edible forage.