An accurate inversion method for field corn biomass based on UAV remote sensing

Abstract

The invention provides a method for estimating field corn biomass based on unmanned aerial vehicle data. The method comprises the following steps: data preprocessing, pixel segmentation of vegetation and soil, calculation of vegetation coverage, construction of a crop elevation model, extraction of elevation of vegetation areas, construction of a biomass estimation model based on pixel integration, and construction of a biomass prediction model based on partial pixel integration after defining and substituting an optimized n value. The method proposes for the first time the idea of ​​predicting biomass by integrating image pixels. In terms of vegetation coverage calculation, the method solves the problem of segmenting corn and soil at maturity by setting a threshold value for height. Compared with the existing results of predicting biomass by corn height, predicting biomass by vegetation coverage, and predicting biomass by combining height with vegetation coverage, the method uses a simpler process to obtain more accurate results, and the model is easy to generalize.

Description

Technical Field

[0001] The invention belongs to the technical field of agricultural remote sensing, and in particular relates to an accurate inversion method for field corn biomass based on unmanned aerial vehicle remote sensing. Background Art

[0002] Currently, researchers estimate the biomass of field corn through UAV data, mainly by establishing a regression model with various parameters extracted from UAV images (such as vegetation index, height, spectral parameters, color, texture, etc.) and measured samples. However, estimating biomass through vegetation index is greatly affected by environmental conditions, and the estimated results are often affected by the spectrum. In addition, the empirical model of biomass constructed by combining vegetation index, height, texture and other characteristics lacks universality. The AGB estimation model established through height and canopy cover (or leaf area) does not take into account the situation where there is no space under the leaves, resulting in inaccurate prediction results.

[0003] Especially in the calculation of vegetation coverage, researchers mainly segment by setting thresholds for RGB color or vegetation index, but it is difficult to segment yellow and dry vegetation through these methods. Summary of the invention

[0004] Technical solution: In order to solve the above technical problems, in terms of biomass estimation calculation, the present invention proposes a method for biomass prediction based on partial pixel integration of drone images to solve the problems existing in the existing methods. Specifically, a method for estimating field corn biomass based on drone data is provided, which includes the following steps:

[0005] Step 1: Data preprocessing;

[0006] The corn plants in a certain area are collected and weighed for fresh weight. After drying, the plants are weighed again for dry weight. The fresh biomass and dry biomass are calculated based on the planting density and used as the actual measurement values. The images collected by the drone are spliced ​​to obtain the digital surface model (DSM).

[0007] Step 2: Separation of vegetation and soil;

[0008] By using the difference in vegetation and soil heights and setting a height threshold, the vegetation and soil in the digital surface model DSM obtained in step 1 are segmented according to the pixel value size;

[0009] Step 3: Calculation of vegetation coverage;

[0010] For the segmentation result in step 2, the pixel ratio of vegetation to the entire plot is calculated to obtain the vegetation coverage;

[0011] Step 4: Construction of crop elevation model;

[0012] Perform raster conversion operation on the image of the soil area segmented in step 2. According to these soil pixel points, the digital elevation model DEM of this plot is interpolated by the inverse distance weighted method IDW. The crop elevation model CHM can be obtained by subtracting DSM from DEM. That is, the height PH of corn plants in each plot is calculated by CHM.

[0013] Step 5: Extract elevation of vegetation area;

[0014] Extract the vegetation area segmented in step 2 by mask on CHM, so as to obtain the vegetation area image containing elevation value;

[0015] Step 6: Construction of biomass estimation model based on pixel integration;

[0016] By reading the vegetation area image containing elevation values ​​and integrating the pixel values ​​of the image, a biomass estimation model can be constructed;

[0017] Step 7: Define the prediction value n and construct a biomass prediction model for partial pixel integration;

[0018] The prediction value n was defined, and the optimal n value was calculated by linear fitting between the pixel values ​​of the DSM image generated by splicing the real measurement value with the UAV image using a computer iterative method. The first n percent of the pixels in the image elevation value were integrated, and AGB=n%∫hdx was used as the optimized biomass prediction model, where n% was the ratio of the vertical projection of the corn trunk to the whole corn plant.

[0019] Step 8: Output predicted biomass;

[0020] The digital surface model DSM collected in real time is input into the prediction model optimized in step seven to obtain the corresponding output value as the prediction value.

[0021] As an improvement, in step 2, the plot is segmented into vegetation and soil according to the pixel value of the digital surface model image, and the threshold segmentation is performed by height, using the preset threshold ε h , elevation value h≥ε h The pixel with is a vegetation pixel, h<ε h The pixels are soil pixels.

[0022] As an improvement, in step six, the specific method is to arrange the pixel values ​​from large to small, draw an elevation change map of the vegetation area of ​​the plot, integrate the image, AGB = ∫hdx, and obtain the prediction result.

[0023] As an improvement, in step 7, the for loop method is used to ensure that the linear regression fitting result of the estimated biomass value and the actual measured value after substituting the n value for each plot is the best, that is, R2 is closest to 1, thus finally determining the optimal n value for each specific plot, where R 2 The calculation formula is:

[0024]

[0025] SSE is the residual sum of squares, and SST is the total sum of squares.

[0026] Beneficial effect: The new method for predicting field corn biomass based on UAV remote sensing images proposed in the present invention is the first to propose the idea of ​​predicting biomass by integrating image pixels. In terms of vegetation coverage calculation, the segmentation problem is solved by setting a threshold for height segmentation. Compared with the existing results of predicting biomass of corn height, predicting biomass by vegetation coverage, and predicting biomass by combining height with vegetation coverage, the accuracy can be significantly improved, and the modeling process is simple and the model is easy to generalize.

[0027] At the same time, in terms of data requirements, this method only needs DSM images to realize the prediction of biomass; in terms of the implementation process, this method is simple and easy to use, and does not require the establishment of complex models for solution.

[0028] In terms of practicality, this method is highly versatile and is not only applicable to this experimental scenario, but also has reference significance for other scenarios. In terms of prediction results, the accuracy of this method has been significantly improved compared with other methods, and the implementation process is simpler. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Flow chart of the method of the present invention.

[0030] Figure 2 is the predicted biomass under different methods in the present invention and the comparative examples, Figure 2 a-1 is the fresh biomass predicted by pH, Figure 2 a-2 is the dry biomass predicted by pH, Figure 2 b-1 is the fresh biomass predicted by FVC, Figure 2 b-2 is the dry biomass predicted by FVC, Figure 2 c-1 is the fresh biomass predicted by PH×FVC, Figure 2 c-2 is the dry biomass predicted by PH×FVC, Figure 2 d-1 is the pixel integral predicted fresh biomass, Figure 2 d-2 is the pixel integral predicted dry biomass, Figure 2 The fresh biomass is predicted by the pixel integration of f-1. Figure 2 The dry biomass is predicted by integrating pixels in part f-2.

[0031] Figure 3The flight data of the drone in step 1 of embodiment 1 of the present invention, Figure 3 a is the digital orthophoto DOM, Figure 3 b is the digital surface model DSM.

[0032] Figure 4 This is a schematic diagram of the segmentation result obtained in step 2 in Example 1 of the present invention.

[0033] Figure 5 It is the FVC calculation result obtained in step 3 in Example 1 of the present invention.

[0034] Figure 6 This is a schematic diagram of the process of constructing a crop elevation model in step 4 of Example 1 of the present invention. Figure 6 a is DEM, Figure 6 b is CHM, Figure 6 c is the height estimation result.

[0035] Figure 7 This is a schematic diagram of a vegetation area containing elevation values ​​in Example 1 of the present invention.

[0036] Figure 8 The biomass estimation based on pixel integration in Example 1 of the present invention is: Figure 8 a is the elevation change map of the vegetation area, Figure 8 b is the biomass prediction result.

[0037] Fig. 9 is the biomass estimation result based on partial pixel integration in Example 1 of the present invention, Fig. 9 a is fresh biomass, Fig. 9 b is dry biomass. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the present invention will be described clearly and completely below so that those skilled in the art can better understand the advantages and features of the present invention, thereby making a clearer definition of the protection scope of the present invention. The embodiments described in the present invention are only a part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present invention.

[0039] The present invention provides a new method for predicting field corn biomass based on UAV remote sensing images. The method is compared with the results of predicting biomass by corn height, predicting biomass by vegetation coverage, and predicting biomass by height combined with vegetation coverage. The method can significantly improve the accuracy, and the modeling process is simple and the model is easy to generalize.

[0040] See Figure 1FIG. 1 is a flowchart of a novel method for predicting biomass based on drones proposed by the present invention, comprising the following steps:

[0041] Step 1: Data preprocessing

[0042] The images acquired by the drone were stitched and processed in DJITerra software to obtain digital orthophotos DOM and digital surface model DSM images. The DOM and DSM images were then cropped according to the plots to obtain the DOM and DSM corresponding to each plot. In terms of real data collection, three plants were selected from each plot by destructive sampling method, and their fresh weight was weighed. The dry weight was weighed after drying. The true values ​​of fresh biomass and dry biomass were obtained by density calculation, and the predicted results were compared with the true results.

[0043] In the present invention, the entire experimental area to be predicted is viewed through the digital orthophoto DOM.

[0044] Step 2: Separation of vegetation and soil

[0045] Since the vegetation has turned yellow in September, it is difficult to distinguish soil and vegetation by color or vegetation index. Considering that the vegetation has reached a certain height at this time, the height difference between vegetation and soil is used to separate vegetation and soil by setting a height threshold.

[0046] In specific operations, the plots are segmented into vegetation and soil according to the pixel values ​​(i.e., elevation values) of the DSM image. When performing threshold segmentation by height, the preset threshold ε is used. h , elevation value h≥ε h The pixel with is a vegetation pixel, h<ε h The pixels are soil pixels.

[0047] Step 3: Calculation of vegetation coverage

[0048] For the segmentation result in step 2, the pixel ratio of vegetation to the entire plot is calculated, namely, the fractional vegetation coverage (FVC).

[0049] Step 4: Construction of crop elevation model

[0050] Perform raster conversion operation on the soil segmented in step 2. Based on these soil pixel points, the digital elevation model (DEM) of this plot is interpolated by the inverse distance weighted method (IDW). The crop elevation model (CHM) can be obtained by subtracting DSM from DEM, that is, CHM = DSM-DEM. The corn plant height (PH) of each plot is calculated by CHM.

[0051] Step 5: Extract elevation of vegetation areas

[0052] The vegetation area segmented in step 2 is extracted on the CHM according to the mask, thereby obtaining a vegetation area image containing elevation values.

[0053] Step 6: Construction of biomass estimation model based on pixel integration

[0054] By reading the vegetation area image containing elevation values ​​and integrating the pixel values ​​of the image, a biomass estimation model can be constructed. The specific implementation method is to arrange the pixel values ​​from large to small, draw an elevation change map of the vegetation area of ​​the plot, integrate the image, AGB = ∫hdx, and the prediction result can be obtained.

[0055] Step 7: Define the prediction value n and construct a biomass prediction model for partial pixel integration;

[0056] In actual conditions, the main trunk of corn is solid, and the trunk is usually the highest part of the entire plant, but the area below the corn leaves is empty. Simply integrating all areas of the vegetation will make the estimated value of the biomass too large. Therefore, the method is improved on this basis. A value n is substituted into the biomass estimation model of pixel integration in step six, and the pixels before the n percent of the image elevation value are integrated, AGB = n% ∫hdx, where n% is the ratio of the corn trunk to the vertical projection of the entire corn plant. The cyclic method is used to ensure that the estimated value of the biomass after substituting the n value for each plot is the best fit to the actual measured value, thereby ultimately determining the optimal n value for each specific plot.

[0057] Step 8: Output predicted biomass;

[0058] The digital surface model DSM collected in real time is input into the prediction model optimized in step seven to obtain the corresponding output value as the prediction value.

[0059] In step seven of the present invention, the accuracy of predicting biomass using partial pixel integration is the highest and the effect is the best, which is introduced and illustrated by specific data below.

[0060] Comparative Example

[0061] As a comparative example of the present invention, the results of predicting fresh biomass and dry biomass by linear fitting based on PH, FVC, PH×FVC, and pixel integral are shown. PH is the vegetation height, and FVC is the vegetation coverage, specifically: predicting biomass by height alone, predicting biomass by vegetation coverage alone, predicting biomass by combining height and vegetation coverage, and predicting biomass by pixel integral.

[0062] See Figure 2 a-1, Figure 2 a-2, Figure 2 b-1, Figure 2 b-2, Figure 2 c-1, Figure 2 c-2, Figure 2 d-1, Figure 2 d-2 shows the predicted fresh biomass and dry biomass based on PH, FVC, PH×FVC and pixel integration method in the comparative example.

[0063] Figure 2 Predict fresh biomass for PH, predict dry biomass for PH, predict fresh biomass for FVC, predict dry biomass for FVC, predict fresh biomass for PH×FVC, predict dry biomass for PH×FVC, predict fresh biomass for PH×FVC, predict dry biomass for pixel integration, Figure 2 The fresh biomass is predicted by the pixel integration of f-1. Figure 2 The dry biomass is predicted by integrating pixels in part f-2.

[0064] Example 1

[0065] See Figure 2 f-1, Figure 2 f-2 shows the result of the present invention predicting fresh biomass and dry biomass by comparing and fitting based on partial pixel integration.

[0066] The R values ​​of different fitting methods for the five methods of estimating biomass are shown below. 2 For comparison, the results are summarized in Table 1 below:

[0067] Table 1 Comparison of biomass prediction results

[0068]

[0069] Table 1 shows that by comparing the results of different models in predicting fresh biomass and dry biomass, it is found that the use of partial pixel integration to predict biomass has the highest accuracy and the best effect.

[0070] In order to better introduce the technical content of the present invention, it is described below through specific embodiments.

[0071] Example 1

[0072] Step 1: Data preprocessing

[0073] On September 10, 2023, data collection was carried out in the experimental field in Dafeng County, Yancheng City, Jiangsu Province. The images taken by the drone were stitched and processed in the DJI Terra software to obtain the digital orthophoto (DOM), that is, the experimental area, such as Figure 3 As shown in a; the digital surface model (DSM) is obtained, and then the DSM is clipped according to the plots in ArcGIS Pro to obtain the DSM corresponding to each plot, as shown in Figure 3 As shown in b.

[0074] Three corn plants were selected from each plot using the destructive sampling method. Their fresh weight was measured and their dry weight was measured again after drying. The true values ​​of fresh biomass and dry biomass were calculated based on the density of the corn plants.

[0075] Step 2: Separation of vegetation and soil

[0076] In ArcGIS Pro, the DSM image is reclassified and the plots are segmented into vegetation and soil according to the pixel values ​​(i.e., elevation values) of the DSM image. When threshold segmentation is performed by height, the preset threshold ε is used. h , elevation value h≥ε h The pixel with is a vegetation pixel, h<ε h The pixels are soil pixels, and the segmentation results are as follows Figure 4 shown.

[0077] Step 3: Calculation of vegetation coverage

[0078] For the segmentation result in step 2, the pixel ratio of vegetation to the entire plot is calculated, that is, the fractional vegetation coverage (FVC), and the calculation result is visualized, as shown in the figure below: Figure 5 shown.

[0079] Step 4: Construction of crop elevation model

[0080] In ArcGIS Pro, the raster-to-point operation is performed on the soil part segmented in step 2, and then the digital elevation model (DEM) of this plot is obtained by interpolation based on these soil pixels using the inverse distance weighted method (IDW). Figure 6 As shown in a; by subtracting DSM from DEM, the crop elevation model (CHM) can be obtained, that is, CHM = DSM-DEM, as shown in Figure 6 b; The corn plant height (PH) of each plot was calculated by CHM, as shown in Figure 6 As shown in c.

[0081] Step 5: Extract elevation of vegetation areas

[0082] The vegetation area segmented in step 2 is extracted on the CHM according to the mask, so as to obtain the vegetation area image containing the elevation value. The extraction result is as follows: Figure 7 shown.

[0083] Step 6: Construction of biomass estimation model based on pixel integration

[0084] By reading the vegetation area image containing elevation values ​​and integrating the pixel values ​​of the image, a biomass estimation model can be constructed. The specific implementation method is to arrange the pixel values ​​from large to small and draw an elevation change map of the vegetation area of ​​the plot. The result is as follows: Figure 8As shown in a, by integrating the elevation change map, AGB = ∫hdx, the prediction result can be obtained, which can be visualized as follows: Figure 8 As shown in b.

[0085] Step 7: Define the predicted value n and construct a biomass prediction model for partial pixel integration

[0086] In actual conditions, the main trunk of corn is solid, and the trunk is usually the highest part of the entire plant, but the area below the corn leaves is empty. Simply integrating all areas of the vegetation will make the estimated value of the biomass too large. Therefore, on this basis, the present invention improves the method, substitutes a numerical value n, and integrates the first n percent of the pixels of the image elevation value, AGB=n%∫hdx, where n% is the ratio of the trunk of corn to the vertical projection of the entire corn plant, and constructs a partial biomass prediction model. Through a cyclic method, it is ensured that the estimated value of the biomass after substituting the n value for each plot is the best to fit the actual measured value, thereby finally determining the optimal n value for each specific plot.

[0087] In step 7, the for loop statement is used to iterate the biomass of each plot, and the estimated biomass value is fitted with the actual measured value by linear regression. 2 Closest to 1, thus finally determining the optimal n value for each specific plot.

[0088] Where R 2 The calculation formula is:

[0089]

[0090] SSE is the residual sum of squares, and SST is the total sum of squares.

[0091] Step 8: Output predicted biomass;

[0092] The digital surface model DSM generated by splicing the images collected by the drone in real time is input into the prediction model optimized in step 7 to obtain the corresponding output value as the prediction value. The prediction of fresh biomass and dry biomass in the present invention is visualized, such as Fig. 9 a and Fig. 9 As shown in b.

[0093] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. An accurate inversion method for field corn biomass based on UAV remote sensing, characterized by: The following steps are involved: Step 1: Data preprocessing; The corn plants in a certain area are collected and weighed for fresh weight. After drying, the plants are weighed again for dry weight. The fresh biomass and dry biomass are calculated based on the planting density and used as the actual measurement values. The images collected by the drone are spliced ​​to obtain the digital surface model (DSM). Step 2: Separation of vegetation and soil; By using the difference in vegetation and soil heights and setting a height threshold, the vegetation and soil in the digital surface model DSM obtained in step 1 are segmented according to the pixel value size; Step 3: Calculation of vegetation coverage; For the segmentation result in step 2, the pixel ratio of vegetation to the entire plot is calculated to obtain the vegetation coverage; Step 4: Construction of crop elevation model; Perform raster conversion operation on the image of the soil area segmented in step 2. According to these soil pixel points, the digital elevation model DEM of this plot is interpolated by the inverse distance weighted method IDW. The crop elevation model CHM can be obtained by subtracting DSM from DEM, that is, CHM = DSM-DEM. The height PH of corn plants in each plot is calculated by CHM. Step 5: Extract elevation of vegetation area; Extract the vegetation area segmented in step 2 by mask on CHM, so as to obtain the vegetation area image containing elevation value; Step 6: Construction of biomass estimation model based on pixel integration; By reading the vegetation area image containing elevation values ​​and integrating the pixel values ​​of the image, a biomass estimation model can be constructed; the specific method is to arrange the pixel values ​​from large to small, draw an elevation change map of the vegetation area of ​​the plot, integrate the image, AGB = ∫hdx, and obtain the prediction result; Step 7: Define the prediction value n and construct a biomass prediction model for partial pixel integration; The prediction value n is defined, and the optimal n value is calculated by linear fitting between the real measurement value and the DSM obtained by the drone using a computer loop iteration method. The first n percent of the image elevation value is constructed for integration, AGB = n% ∫hdx, as the optimized biomass prediction model, where n% is the ratio of the vertical projection of the corn trunk to the whole corn plant; Step 8: Output predicted biomass; The digital surface model DSM generated by stitching the images collected by the drone is input into the prediction model optimized in step seven to obtain the corresponding output value as the prediction value.

2. The accurate inversion method of field corn biomass based on UAV remote sensing according to claim 1 is characterized by: In step 1, the method for obtaining the true measurement value of the organism is to select three plants from each plot using the destructive sampling method, weigh their mass, dry them and weigh their mass, and calculate the true value of the fresh biomass and dry biomass based on the planting density; the images obtained by the drone are first spliced ​​to obtain the digital surface model DSM, and then the images are cropped according to the plots to obtain the digital surface model DSM corresponding to each plot.

3. The accurate inversion method of field corn biomass based on UAV remote sensing according to claim 1 is characterized by: In step 2, the plot is segmented into vegetation and soil according to the pixel value of the digital surface model image, and the threshold segmentation is performed by height, using the preset threshold ε h , elevation value h≥ε h The pixel with h is a vegetation pixel, h<ε h The pixels are soil pixels.

4. The accurate inversion method of field corn biomass based on UAV remote sensing according to claim 1 is characterized by: In step 7, the for loop statement is used to iterate the biomass of each plot. When the linear regression fitting result between the estimated biomass value and the actual measured value after substituting the n value for each plot is optimal, the optimal n value for each plot is determined, where R 2 The calculation formula is: SSE is the residual sum of squares, and SST is the total sum of squares.

Images