Potato leaf area index inversion calculation method based on Gaofen-7 random forest algorithm
By using the random forest algorithm of the Gaofen 7 satellite in the potato growth area, the saturation phenomenon, regionality and timeliness in the traditional LAI inversion method are solved, and LAI estimation with higher accuracy and stability is achieved, and for agricultural production, it is used to achieve higher accuracy and stability. Provides more reliable technical support.
Patent Information
- Application Number
- CN202411459076.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-18
- Publication Date
- 2025-02-11
AI Technical Summary
In the prior art, there are saturation phenomena, regionality and timeliness of the potato leaf area index (LAI), making it difficult to achieve high-precision and stable remote sensing inversion.
The Gaofen 7 (GF-7) random forest algorithm was used to obtain and process the data in the study area, and a vegetation index with high correlation with LAI was selected to construct a random forest model for LAI inversion.
It improves the estimation accuracy and stability of potato LAI, overcomes the limitations of saturation, regionality and timeliness in traditional methods, and provides more reliable agricultural production technical support.
Smart Images
Figure CN119418202A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of potato leaf area index calculation methods, and in particular to a potato leaf area index calculation method using a high-resolution random forest algorithm to invert the potato leaf area index. Background Art
[0002] Potato is an important agricultural product in the world, and the monitoring of its growth status plays a decisive role in agricultural production efficiency. Leaf area index (LAI) is a key parameter reflecting the growth status of potato, which is crucial for evaluating crop growth, predicting yield and managing farmland resources. Traditionally, the measurement of LAI relies on ground sampling and manual measurement, but this method has many limitations, such as low temporal and spatial resolution, high cost and difficulty in large-scale application. With the rapid development of remote sensing technology, high-resolution satellite image data has provided new opportunities for remote sensing inversion of LAI. These satellite data have a wide spatial coverage and high temporal and spatial resolution, making them an ideal choice for monitoring large areas of farmland. In the quantitative remote sensing research of LAI, both the inversion methods and data sources have been significantly improved. Traditional empirical models and physical models are two commonly used inversion methods. The physical model has strong mechanistic rationality and a wide range of applications, but its parameters are complex and difficult to obtain; while the empirical model based on vegetation index, although weak in mechanistic rationality, can monitor crop growth in real time and non-destructively, and is easy to combine with remote sensing observation data. In terms of data sources, from multispectral data to hyperspectral data, lidar data, high spatial resolution data and drone remote sensing data, more options are provided for LAI inversion.
[0003] In recent years, the univariate regression model based on a single vegetation index has been widely used in LAI estimation, but it has problems such as saturation, regionality and timeliness. In order to overcome these limitations, machine learning regression algorithms have been introduced into LAI remote sensing estimation. These algorithms can fit flexible models, learn the relationship between input variables and output variables, and have good nonlinear fitting and generalization capabilities. Existing studies have shown that machine learning regression algorithms have achieved remarkable results in LAI estimation. For example, algorithms such as BP neural network, support vector machine regression and random forest have shown high accuracy and stability. However, although machine learning regression algorithms have achieved good results in LAI remote sensing estimation, there are relatively few studies on random forest algorithms for estimating potato LAI. As a powerful machine learning algorithm, random forest algorithm has excellent classification and regression performance, and is expected to further improve the estimation accuracy and stability of potato LAI. Therefore, in the future, research on random forest algorithms in potato LAI remote sensing estimation should be strengthened to provide more reliable technical support for agricultural production. Summary of the invention
[0004] In order to solve the problems existing in the prior art, the present invention provides a method for calculating potato leaf area index by inverting the random forest algorithm of Gaofen-7, which solves the problems of saturation, regionality and timeliness limitation existing in the prior art.
[0005] The calculation method of potato leaf area index inversion using the Gaofen-7 random forest algorithm includes the following steps:
[0006] Step S1: Research the research area and methods;
[0007] Step S2: Obtain analysis results based on the research area and method.
[0008] Preferably, step S1 includes the following sub-steps:
[0009] Sub-step S11: studying the general situation of the study area;
[0010] Sub-step S12: acquiring data and processing data;
[0011] Sub-step S13: Obtaining a research method based on acquiring and processing data.
[0012] Preferably, step S12 includes the following sub-steps:
[0013] Sub-step S121: obtaining LAI measured data;
[0014] Sub-step S122: Acquire remote sensing image data.
[0015] Preferably, sub-step S13 includes the following sub-steps:
[0016] Sub-step S131: Selecting vegetation index;
[0017] Sub-step S132: obtaining a random forest algorithm;
[0018] Sub-step S133: analyzing correlation;
[0019] Sub-step S134: Obtaining estimation model accuracy verification.
[0020] Preferably, step S2 includes the following sub-steps:
[0021] Sub-step S21: obtaining LAI correlation analysis according to vegetation index;
[0022] Sub-step S22: Perform LAI inversion on the GF-7 image based on LAI correlation analysis.
[0023] The beneficial effects of the method for calculating potato leaf area index by using the high-resolution random forest algorithm of the present invention are as follows:
[0024] In the future, research on the application of random forest algorithm in potato LAI remote sensing estimation should be strengthened to provide more reliable technical support for agricultural production. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 This is the location of the study area and the distribution map of sampling points of the present invention.
[0026] FIG. 2( a ) is a GNDVI graph and a LAI correlation graph of the present invention.
[0027] FIG. 2( b ) is a correlation diagram between RVI and LAI of the present invention.
[0028] FIG. 2( c ) is a correlation diagram between NLI and LAI of the present invention.
[0029] Figure 3(a) is a fitting index diagram of the 2021 LAI measured data and remote sensing inversion data of the present invention.
[0030] FIG3( b ) is the fitting index of the 2022 LAI measured data and the remote sensing inversion data of the present invention.
[0031] Figure 3(c) is a fitting index diagram of the LAI measured data and remote sensing inversion data from 2021 to 2022 of the present invention.
[0032] Figure 3 (d) Fitting index diagram of LAI measured data in 2021 and remote sensing inversion data of the best band (original band + GNDVI).
[0033] Figure 3 (e) Fitting index diagram of LAI measured data in 2022 and remote sensing inversion data in the best band (original band + GNDVI).
[0034] Figure 3 (f) Fitting index diagram of LAI measured data in 2021-2022 and remote sensing inversion data in the best band (original band + GNDVI).
[0035] Figure 3 (g) Fitting index diagram of LAI measured data in 2021 and remote sensing inversion data of the best band (original band + GRVI).
[0036] Figure 3 (h) Fitting index diagram of LAI measured data in 2022 and remote sensing inversion data in the best band (original band + GRVI).
[0037] Figure 3 (i) Fitting index diagram of LAI measured data and remote sensing inversion data of the best band (original band + GRVI) in 2021-2022.
[0038] Figure 3 (j) Fitting index diagram of LAI measured data in 2021 and remote sensing inversion data of the best band (original band + NLI).
[0039] Figure 3 (k) Fitting index diagram of LAI measured data in 2022 and remote sensing inversion data of the best band (original band + NLI).
[0040] Figure 3 (l) Fitting index diagram of LAI measured data and remote sensing inversion data of the best band (original band + NLI) from 2021 to 2022.
[0041] Figure 3 (m) Fitting index diagram of LAI measured data in 2021 and optimal band remote sensing inversion data.
[0042] Fig. 3 (n) Fitting index diagram of LAI measured data in 2022 and remote sensing inversion data in the best band.
[0043] Figure 3 (o) Fitting index diagram of LAI measured data and optimal band remote sensing inversion data in 2021-2022. Specific embodiments
[0044] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.
[0045] 1 Research area and methods
[0046] 1.1 Overview of the study area
[0047] like Figure 1 As shown, the study area is located in the north of Hale Town, Wuchuan County, Hohhot City, Inner Mongolia Autonomous Region, with a longitude range of 41°14'49.83"-41°15'0.78"E and a latitude range of 111°40'43.18"-111°41'4.90"N. It is located at the northern foot of Yinshan Mountain and has a temperate continental climate with a large temperature difference between day and night; the average temperature over the years is 2.5℃, the average frost-free period is 90-110 days per year, the average annual sunshine hours are 2958 hours, and the annual total radiation is 336.53 kcal / cm2; the average annual wind speed is 2.5 m / s, and the average annual precipitation is 250-360 mm.
[0048] 1.2 Data acquisition and processing
[0049] 1.2.1 LAI measured data
[0050] On July 23, 2021 and July 23, 2022, ground data collection was carried out in the study area. In the experimental area, a handheld GPS device was used to collect ground control points and record the coordinate information of the sample plots. In addition, the LAI-2200 plant canopy analyzer was used to measure the LAI value. To ensure the reliability of the data, three measurements were performed in different directions at each measuring point.
[0051] 1.2.2 Remote sensing image data
[0052] The Gaofen-7 (GF-7) satellite is my country's first civilian sub-meter-level high-resolution optical transmission stereo mapping satellite, including panchromatic and multispectral (near infrared, red, green and blue bands), with a panchromatic resolution of 0.8 meters and a multispectral resolution of 3.2 meters. This institute uses the GF-7 remote sensing satellite, which was acquired on July 27, 2021 and July 20, 2022. The acquired images were atmospherically corrected, orthorectified and fused using ENVI5.6 software.
[0053] 1.3 Research Methods
[0054] 1.3.1 Vegetation index selection
[0055] The change of LAI is associated with the change of optical properties of leaves, and has a strong correlation with the visible near-infrared bands of vegetation spectra. The vegetation index constructed using these characteristic bands can be used to estimate the leaf area index and monitor vegetation growth. When using multi-band spectral data to invert LAI, the inversion method is complex to calculate and is easily interfered by factors such as water vapor absorption in remote sensing images. The vegetation index is easy to calculate and can replace multi-band spectral data as a model parameter to reduce interference factors and highlight target information. The present invention selects 21 vegetation indices, selects vegetation indices with a higher correlation with LAI for the best band combination, and the calculation formulas of each vegetation index are shown in Table 1.
[0056] Table 1 Vegetation index and calculation formula
[0057]
[0058] Note: nir , R red , R blue , R green They are the reflectances of the near-infrared band, red band, blue band, and green band of the GF-7 image.
[0059] 1.3.2 Random Forest Algorithm
[0060] The Random Forest (RF) algorithm was proposed by Breiman in 2001. It is an integrated machine learning method based on multiple classification and regression trees. It constructs multiple decision trees for classification or regression prediction. The core of random forest lies in "integration" and "randomness", and improves the accuracy of prediction by combining multiple decision trees. Random forest belongs to the Bagging (Bootstrap Aggregating) method in ensemble learning. It constructs multiple models (decision trees) and integrates their prediction results in the hope of obtaining better prediction performance than a single model. Random forest introduces randomness when constructing each decision tree, which is reflected in two aspects: first, in the selection of training data, different training sets are generated for each tree through bootstrap sampling; second, in feature selection, each tree only randomly selects a part of all features as candidate features when splitting a node.
[0061] The working principle of random forest is to generate multiple training subsets from the original training set using sampling with replacement. For each training subset, a decision tree is constructed, and a portion of features are randomly selected from all features when each node is split. Each tree grows as large as possible without pruning. For classification problems, the prediction results of all trees for new samples are used to determine the final result through majority voting; for regression problems, the average of the prediction results of multiple trees is taken. After experimental comparison, the number of random forest regression trees in this study is set to 300, and the number of random features for splitting nodes is set to 2.
[0062] 1.3.3 Correlation Analysis
[0063] The correlation coefficient is a statistical indicator used to reflect the closeness of the correlation between variables. It is a quantity that studies the degree of linear correlation between variables. It is usually expressed as r and the calculation formula is as follows:
[0064]
[0065] Where N is the number of samples, x and y are variables, and are the average values of variables x and y, respectively. The value range of the correlation coefficient is between -1 and 1. A positive correlation coefficient indicates that the two sets of data are positively correlated; a negative correlation coefficient indicates that the two sets of data are negatively correlated. The larger the absolute value of the correlation coefficient (the closer it is to 1 or equal to 1), the higher the correlation between the two sets of data.
[0066] 1.3.4 Estimation Model Accuracy Verification
[0067] The coefficient of determination is a measure used to characterize the degree of fit of the sample data of the prediction model, denoted as R2, with a value range of 0 to 1, and the calculation formula is as follows:
[0068]
[0069] Where N is the number of samples, i is the sample number, and yi is the measured value of the i-th sample. is the predicted value of the i-th sample, is the average of all sample measured values. Calculating the R2 of different models can be used to evaluate the prediction accuracy of each model. The closer the R2 value is to 1, the better the explanation of the dependent variable, the higher the model fitting effect, and the denser the distribution of the predicted value points near the regression line; otherwise, it means that the prediction results of the model are of poor reference value.
[0070] The root mean square error (RMSE) is a measure of the deviation between the predicted value and the measured value. The calculation formula is the square root of the ratio of the square of the deviation between the predicted value and the measured value to the number of observations:
[0071]
[0072] In the formula, N is the number of samples, i is the sample number, yi is the measured value of the i-th sample, is the predicted value of the i-th sample. The smaller the RMS error value, the smaller the deviation between the predicted value and the measured value; the larger the RMS error value, the greater the deviation between the predicted value and the measured value.
[0073] 2 Results and analysis
[0074] 2.1 Correlation analysis between vegetation index and LAI
[0075] Table 2 shows the correlation results between different vegetation indices and measured LAI. The absolute value of the correlation between the potato vegetation index obtained using GF-7 image data and the measured LAI is good, among which the GNDVI index is the highest, at 0.678, followed by GRVI and NLI (as shown in Figure 2(a), Figure 2(b) and Figure 2(c)).
[0076] Table 2 Correlation analysis between vegetation index and LAI
[0077]
[0078] 2.2 GF-7 image LAI inversion
[0079] This study conducted a comprehensive comparison and analysis of the model-inverted LA and the corresponding ground-measured LAI in 2021 and 2022. Figure 3aThe fitting data in 2021 showed that the LAI inverted by the model showed a high regression fit with the measured value, and the regression fit index was 0.525, indicating that the model can predict the change trend of LAI well in most cases. However, the results of this study showed that in some areas and specific time points, the LAI inverted by the model had a certain degree of deviation, and its RMSE was 0.139, indicating that there is still room for improvement in practical applications. Figure 3b , the same regression fitting analysis was conducted again in 2022. The results show that the accuracy of the model inversion has improved, and the regression fit index has risen to 0.7834, indicating that the model has a certain ability to adapt to environmental changes and data updates. At the same time, the RMSE dropped to 0.160, indicating that the model error has been reduced and the inversion results are closer to the measured values. Through the scatter plot analysis, it is found that the scatter points between the model inversion and the measured LAI are more closely distributed around the fitting line, which further verifies the improvement of the model performance. According to Figure 3c A comprehensive regression fitting analysis was performed on the model-inverted LAI and the corresponding ground-measured LAI in 2021 and 2022. The results showed that the accuracy of the model inversion was further improved, the regression fit index increased to 0.812, and the RMSE decreased to 0.150, indicating that the model fits better as the amount of data for fitting increases.
[0080] In order to reduce the phenomenon of different objects with the same spectrum and different objects with the same spectrum, this study conducted correlation analysis based on the above bands to screen the bands that are helpful in identifying the study objects and the original bands for the best band combination. The model was used to invert the LAI in 2021 and 2022, and further fit them with the corresponding ground measured LAI.
[0081] according to Figure 3d The fitting data in 2021 showed that the LAI inverted based on the optimal band combination (original band + GNDVI) model showed a high regression fit with the measured value, and the regression fit index was 0.623, indicating that the LAI inverted based on the optimal band combination had a better fit with the measured LAI value. Figure 3e In 2022, the same regression fitting analysis was performed between the LAI inverted based on the optimal band combination (original band + GNDVI) model and the measured values. The results showed that the accuracy of the model inversion was improved, and the regression fit index rose to 0.829, indicating that the model showed a high regression fit between the LAI inverted based on the optimal band combination model and the measured values. Figure 3fA comprehensive regression fitting analysis was performed on the best band combination (original band + GNDVI) of the model inversion in 2021 and 2022 and the corresponding ground-measured LAI. The results showed that the accuracy of the model inversion was further improved, and the regression fit index rose to 0.850, indicating that the model fits better as the amount of data for fitting increases.
[0082] according to Figure 3g The fitting data in 2021 showed that the LAI inverted based on the optimal band combination (original band + GRVI) model showed a high regression fit with the measured value, and the regression fit index was 0.626, indicating that the LAI inverted based on the optimal band combination had a better fit with the measured LAI value. Figure 3h In 2022, the same regression fitting analysis was performed between the LAI inverted based on the optimal band combination (original band + GRVI) model and the measured values. The results showed that the accuracy of the model inversion was improved, and the regression fit index rose to 0.828, indicating that the model showed a high regression fit between the LAI inverted based on the optimal band combination model and the measured values. Figure 3i A comprehensive regression fitting analysis was performed on the best band combination (original band + GRVI) of the model inversion in 2021 and 2022 and the corresponding ground-measured LAI. The results showed that the accuracy of the model inversion was further improved, and the regression fit index rose to 0.850, indicating that the model fits better as the amount of data for fitting increases.
[0083] according to Figure 3j The fitting data in 2021 showed that the LAI inverted based on the optimal band combination (original band + NLI) model showed a high regression fit with the measured value, and the regression fit index was 0.603, indicating that the LAI inverted based on the optimal band combination had a better fit with the measured LAI value. Figure 3k In 2022, the same regression fitting analysis was performed between the LAI inverted based on the optimal band combination (original band + NLI) model and the measured values. The results showed that the accuracy of the model inversion was improved, and the regression fit index rose to 0.8283, indicating that the model showed a high regression fit between the LAI inverted based on the optimal band combination model and the measured values. Figure 3l A comprehensive regression fitting analysis was performed on the best band combination (original band + NLI) of the model inversion in 2021 and 2022 and the corresponding ground-measured LAI. The results showed that the accuracy of the model inversion was further improved, and the regression fit index rose to 0.844, indicating that the model fits better as the amount of data for fitting increases.
[0084] according to Figure 3mThe fitting data in 2021 showed that the LAI inverted based on the optimal band combination (original band + GNDVI + GRVI + NLI) model showed a high regression fit with the measured value, and the regression fit index was 0.632, indicating that the LAI inverted based on the optimal band combination has a better fit with the measured LAI value. Figure 3n In 2022, the same regression fitting analysis was performed between the LAI inverted based on the optimal band combination (original band + NLI) model and the measured values. The results showed that the accuracy of the model inversion was improved, and the regression fit index rose to 0.829, indicating that the model showed a high regression fit between the LAI inverted based on the optimal band combination model and the measured values. Figure 3o A comprehensive regression fitting analysis was performed on the best band combination (original band + GNDVI + GRVI + NLI) of the model inversion in 2021 and 2022 with the corresponding ground-measured LAI. The results showed that the accuracy of the model inversion was further improved, and the regression fit index rose to 0.851, indicating that the model fits better as the amount of data for fitting increases.
[0085] In the inter-annual comparison, the present invention found that the stability and consistency of the model inversion in 2022 were enhanced. Environmental factors such as climate conditions and vegetation growth conditions have changed between 2021 and 2022. The model prediction accuracy is within a certain range of data quality and data quantity. When the above conditions are good, the model shows its strong adaptability and robustness. In the future, we will continue to optimize the model algorithm, improve the quality of input data, strengthen the monitoring of environmental variables, etc., in order to further improve the accuracy and reliability of the model inversion LAI.
Claims
1. A method for calculating potato leaf area index by using the Gaofen-7 random forest algorithm, characterized in that: The following steps are involved: Step S1: Research the research area and methods; Step S2: Obtain analysis results based on the research area and method.
2. The method for calculating potato leaf area index by using the high-resolution random forest algorithm according to claim 1, characterized in that: The step S1 comprises the following sub-steps: Sub-step S11: studying the general situation of the study area; Sub-step S12: acquiring data and processing data; Sub-step S13: Obtaining a research method based on acquiring and processing data.
3. The method for calculating potato leaf area index by using the high-resolution random forest algorithm according to claim 1, characterized in that: The step S12 includes the following sub-steps: Sub-step S121: obtaining LAI measured data; Sub-step S122: Acquire remote sensing image data.
4. The method for calculating potato leaf area index by using the high-resolution random forest algorithm according to claim 1, characterized in that: The sub-step S13 includes the following sub-steps: Sub-step S131: Selecting vegetation index; Sub-step S132: obtaining a random forest algorithm; Sub-step S133: analyzing correlation; Sub-step S134: Obtaining estimation model accuracy verification.
5. The method for calculating potato leaf area index by using the high-resolution random forest algorithm according to claim 1, characterized in that: The step S2 comprises the following sub-steps: Sub-step S21: obtaining LAI correlation analysis according to vegetation index; Sub-step S22: Perform LAI inversion on the GF-7 image based on LAI correlation analysis.