| Plant height is a key index to dynamically measure crop health and overall growth status,which is widely used to estimate biological yield and final grain yield of crop.The traditional method of artificial height measurement is inefficient and subjective,and it is difficult to collect plant height data in a large area.With the rapid development of remote sensing technology in the field of agriculture,crop height acquisition with high precision,high frequency,and high efficiency has made it possible.UAV combined with high-resolution digital camera has become the most extensive way to obtain crop height information,due to its low cost and high flexibility.To obtain high-precision crop height estimation by low-altitude unmanned aerial vehicle remote sensing,various types of spatial auxiliary information are needed,such as digital terrain model(DTM),digital surface model(DSM),and ground control point(GCP)to improve image accuracy and registration.Different fields such as scientific research or agricultural production have different requirements for accuracy and cost of plant height estimation,and environmental conditions in the field also have limits for data acquisition.In view of the above problem,this experiment used unmanned aerial vehicle(UAV)carrying digital camera and multispectral camera to collect image.The accuracy of plant height estimation based on spatial auxiliary data was compared under 4 different complete conditions,one of which is complete condition and 3 of which is incomplete condition.In order to improve the accuracy of plant height estimation,the corresponding improvement scheme was put forward according to the missing data.At the same time,5 kinds of visible light and 13 kinds of five-band vegetation indices were calculated,and the plant height was estimated indirectly through the establishment of exponential regression model.The spectral index and DSM data were combined to establish multiple linear regression modeling to improve the estimation effect of plant height.Finally,the experiment comprehensively summarizes and evaluates the accuracy and cost of crop plant height estimation under different data completeness,and makes corresponding decisions,which can balance the cost while satisfying the accuracy in scientific research and practical agricultural production.The main conclusions of this study are as follows:(1)The low altitude UAV remote sensing platform was equipped with high-definition digital camera to obtain plant height of crops.Under the condition of complete data,the plant height estimation of rapeseed can achieve great prediction accuracy.The average R~2 of 3 images was 0881,with the root mean squared error(RMSE)of 0.030 m,and the relative error(RE)of 7.1%,which has become the most widely used method to obtain plant height.Although the data are complete,plant height estimates are generally underestimated due to the poor performance of structure from motion(Sf M)method of crops with small reflective surfaces,which can be significantly improved by increasing point cloud density.(2)In the absence of DTM data,the accuracy of plant height estimation was decreased due to the influence of terrain,with an average R~2 of 0.734,RMSE and RE of0.041 m and 9.9%,respectively.In this experiment,three improvement methods were proposed,two of which were to extract the elevation data of discrete points of soil as the DTM,and the other was to construct a complete and continuous soil surface by interpolation after extracting the soil points.Due to the closed canopies of rapeseed in this experiment,it was the best effect to extract the discrete elevation values of bare soil around each plot as the DTM.However,when there was more bare soil,it is best to extract soil points for interpolation to obtain complete DTM.(3)In the absence of ground control point data,it was difficult to georeference multi-phase images,which resulted in the decrease of accuracy in plant height estimation.The average R~2 was 0.622,RMSE and RE were 0.045 m and 13.9%,respectively.The experiment proposed that the ground control points with prominent features,fixed positions and evenly distributed in the whole study can be directly selected from the UAV images.The spatial distribution and number of ground control points will affect the quality of image geometric correction,and the uniform distribution of horizontal and vertical directions is necessary.The extraction of individual crop traits is more sensitive to the lack of ground control point data.(4)In the absence of both DTM and ground control point data,the accuracy was reduced,with an average R~2 of 0.689,RMSE and RE of 0.082 m and 20.5%,respectively.DTM flatness directly determines the selection of spatial auxiliary data.In the absence of ground control points,when the local topography is flat,the plant height estimation results are better without DTM.And DTM is necessary when the local terrain is rough.(5)The spectral index was used to establish the exponential regression model,and the optimal accuracy of R~2 was 0.736 and RMSE was 0.053 m.Because the spectral index could not be affected by the absence of DTM,this study used the spectral index and DSM to establish the multiple linear regression model,and the optimal R~2 was improved to 0.881 and RMSE decreased to 0.036 m.Although the accuracy is improved compared with the absence of both DTM and GCP data.However,spectral index method is saturated at the later stage of crop growth,and the empirical model established is not universal.(6)Low-altitude UAV remote sensing technology provides an efficient,convenient,and low-cost means for crop height extraction in the field.In order to obtain the required spatial auxiliary information of plant height,this paper comprehensively summarized the methods of plant height estimation under various data conditions,and evaluated its accuracy and cost by combining experimental data and expert experience,which can provide an effective method reference for scientific research and practical agriculture production. |
