Rice leaf area index estimation method and system fused with space-time angle data

Abstract

The invention provides a rice leaf area index estimation method and system fused with space-time angle data, and the method comprises the steps: providing a multi-head attention depth regression model, so as to fuse the data of different space-time angle scales, and achieve the high-precision estimation of a rice leaf area index in multiple periods. On the basis of analyzing the change of the normalized difference red edge index of the rice canopy along with the observation height, time and angle, the optimal observation condition suitable for estimating the rice leaf area index is determined, and a frame based on multi-head attention regression fusion space-time angle scale data is provided; and high-precision estimation of the multi-period rice leaf area index is realized.

Description

Technical Field

[0001] The present invention relates to the field of agricultural remote sensing technology, and in particular to a rice leaf area index estimation method and system integrating spatiotemporal angle data. Background Art

[0002] Rice growth parameters are important indicators for evaluating rice growth, conducting precision agricultural management, and increasing crop yields. Leaf Area Index (LAI) quantifies the leaf area per unit ground area and is an important parameter for controlling biological and physical processes related to surface vegetation. It is closely related to biophysical processes such as photosynthesis and transpiration. Timely and accurate acquisition of rice LAI data can provide a decision-making basis for field management and is of great significance for precision agriculture and crop trait screening.

[0003] Traditional manual lossy measurement methods are time-consuming and labor-intensive and cannot serve large-scale crop growth monitoring. The method based on physical model inversion cannot be applied on a large scale due to the complex operation process and difficult parameter adjustment. UAV multispectral remote sensing, with its advantages of unified spectrum and high-throughput observation, provides a feasible solution for the efficient extraction of crop growth parameters. In recent years, due to the development and integration of sensors, automatic control, computers and aviation technologies, spectral data obtained by remote sensing technology can realize the rapid and non-destructive extraction of rice growth information, with the characteristics of multiple spatiotemporal angular scales. Rice observations at different spatial scales can understand the spatial distribution of rice growth from the microscopic and macroscopic levels, and rice observations at different time scales can help provide dynamic and continuous growth information. Rice observations at different angles can obtain more comprehensive canopy information. Monitoring rice growth based on multiple spatiotemporal angular scales is crucial to improving the accuracy of LAI estimation and revealing the interaction between rice canopy and light.

[0004] In existing UAV remote sensing monitoring research, although most of them describe the flight strategies of UAVs (altitude, time, etc.), they often ignore the impact of different UAV flight strategies on the estimation of crop growth parameters. Most scholars are blind when formulating actual UAV flight strategies. Studies by different scholars have shown that the UAV flight strategies suitable for estimating different growth parameters of crops are not the same, and they are all result-oriented, without in-depth analysis of the reasons and changing trends of different image features obtained at different heights or times. In addition, existing research mainly focuses on discussing the impact of spatial texture changes on the estimation of growth parameters, ignoring the analysis of spectral changes.

[0005] In addition, data fusion at multiple spatiotemporal angle scales has become an important way to improve the accuracy of crop growth monitoring. For machine learning, not all variables are equally important, and redundant variable inputs will only reduce model performance. Since data at different spatiotemporal angles have different sensitivities to crop LAI, choosing the right fusion mode is crucial to model accuracy when fusion of multiple data. Summary of the invention

[0006] The present invention provides a rice leaf area index estimation method and system integrating spatiotemporal angle data, so as to solve the defects existing in the prior art.

[0007] In a first aspect, the present invention provides a rice leaf area index estimation method integrating spatiotemporal angle data, comprising: Collect multi-spectral UAV images at multiple time and space angles in the study area, and pre-process the multi-spectral UAV images at multiple time and space angles to obtain pre-processed UAV multi-spectral images; Calculate the normalized difference red edge index of the rice canopy at multiple time and space angles in the multispectral preprocessed image of the UAV based on the reflectivity of the multispectral image in the corresponding band; Obtaining a typical growth period of rice, calculating the estimation accuracy of a linear regression model of rice leaf area index in a single period based on the normalized difference red edge index, comparing the estimation accuracy, and determining that the best observation condition is when the rice leaf area index estimation accuracy is the highest; The estimation accuracy of the multi-period rice leaf area index linear regression model under various time and space angle combination conditions is compared to determine the highest estimation accuracy of the multi-period rice leaf area index under the optimal observation conditions; A multi-head attention deep regression model is constructed, and the normalized difference red edge index obtained under the optimal observation conditions is input into the multi-head attention deep regression model to obtain a high-precision estimate of the rice leaf area index in multiple periods, and the optimal rice leaf area index estimation result is obtained after comparison.

[0008] According to a rice leaf area index estimation method integrating spatiotemporal angle data provided by the present invention, In a second aspect, the present invention further provides a rice leaf area index estimation system integrating spatiotemporal angle data, comprising: A preprocessing module is used to collect multi-temporal and spatial angle UAV multispectral images of the research area, and preprocess the multi-temporal and spatial angle UAV multispectral images to obtain UAV multispectral preprocessed images; A calculation module, used for calculating the normalized difference red edge index of the rice canopy at multiple time and space angles in the multispectral preprocessed image of the UAV based on the reflectivity of the multispectral image in the corresponding band; An acquisition module is used to acquire a typical growth period of rice, calculate the estimation accuracy of a linear regression model of rice leaf area index in a single period based on the normalized difference red edge index, compare the estimation accuracy, and determine that the best observation condition is when the rice leaf area index estimation accuracy is the highest; A comparison module is used to compare the estimation accuracy of the multi-period rice leaf area index linear regression model under various time and space angle combination conditions, and determine the highest estimation accuracy of the multi-period rice leaf area index under the optimal observation conditions; A construction module is used to construct a multi-head attention deep regression model, and the normalized difference red edge index obtained under the optimal observation conditions is input into the multi-head attention deep regression model to obtain a high-precision estimate of the rice leaf area index in multiple periods, and the optimal rice leaf area index estimation result is obtained after comparison.

[0009] The rice leaf area index estimation method and system for integrating spatiotemporal angular data provided by the present invention determine the optimal observation conditions suitable for rice leaf area index estimation based on the analysis of changes in the normalized difference red edge index of the rice canopy with observation height, time and angle, and propose a framework based on multi-head attention regression to integrate spatiotemporal angular scale data, thereby achieving high-precision estimation of rice leaf area index in multiple periods. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0011] Figure 1 It is a schematic diagram of the flow chart of the rice leaf area index estimation method integrating spatiotemporal angle data provided by the present invention; Figure 2 Schematic diagram of the structure of the rice leaf area index estimation system integrating spatiotemporal angle data provided by the present invention; Figure 3 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0013] In view of the problems existing in the prior art, the present invention proposes a rice leaf area index estimation method that integrates spatiotemporal angular data. Through a multi-head attention deep regression model, data of different spatiotemporal angular scales are integrated to achieve high-precision estimation of rice LAI in multiple periods.

[0014] Figure 1 FIG. 1 is a flow chart of a rice leaf area index estimation method for integrating spatiotemporal angle data provided by an embodiment of the present invention. Figure 1 As shown, including: Step 100: Collect multi-temporal and spatial angle UAV multispectral images of the study area, and pre-process the multi-temporal and spatial angle UAV multispectral images to obtain UAV multispectral pre-processed images; Step 200: Calculate the normalized difference red edge index of the rice canopy at multiple time and space angles in the multispectral preprocessed image of the UAV based on the reflectivity of the multispectral image in the corresponding band; Step 300: obtaining a typical growth period of rice, calculating the estimation accuracy of a linear regression model of rice leaf area index in a single period based on the normalized difference red edge index, comparing the estimation accuracy, and determining that the best observation condition is when the rice leaf area index estimation accuracy is the highest; Step 400: comparing the estimation accuracies of the multi-period rice leaf area index linear regression models under various time-space angle combination conditions, and determining the highest estimation accuracy of the multi-period rice leaf area index under the optimal observation conditions; Step 500: construct a multi-head attention deep regression model, input the normalized difference red edge index obtained under the optimal observation conditions into the multi-head attention deep regression model, obtain a high-precision estimate of the rice leaf area index in multiple periods, and obtain the optimal rice leaf area index estimation result after comparison. The present invention determines the optimal observation conditions suitable for estimating the rice leaf area index by analyzing the changes in the Normalized Difference RedEdge Index (NDRE) of the rice canopy with observation height, time and angle, and proposes a framework based on multi-head attention regression to fuse spatiotemporal angular scale data, thereby achieving high-precision estimation of the rice leaf area index in multiple periods.

[0015] Based on the above embodiment, before step 100, the following step is further included: A plurality of fields of equal size are set up in the research area, and a plurality of different hybrid rice varieties are planted in the plurality of fields; A plurality of different nitrogen fertilizer levels are also set in the plurality of fields, and the same field management plan is adopted; Rice is planted in equally spaced holes in each field.

[0016] Specifically, the embodiment of the present invention designs a field experiment, which includes 27 plots (each plot is about 3 meters × 5 meters), and three representative hybrid rice varieties are planted, namely Fengliangyou No. 4 (G1), Changjingyou 582 (G2) and Luoyou 9348 (G3). The experiment sets three nitrogen fertilizer levels (1 / 4N: 2.4 kg / mu, N: 9.6 kg / mu, 2N: 19.2 kg / mu) and 3 repetitions. In addition to nitrogen fertilizer, phosphate fertilizer and potassium fertilizer are applied in the same proportion as nitrogen fertilizer. The nitrogen application level of each experimental plot is the same, and white plastic film is laid between plots with different nitrogen application levels to isolate water and fertilizer. In addition to the nitrogen application level, all plots adopt the same field management plan, and each plot is planted with 378 rice holes (14×27 holes), with a spacing of 20 cm between each hole.

[0017] Based on the above embodiment, collecting multi-spectral drone images of the research area at multiple time and space angles in step 100 includes: During different critical growth periods of rice, drone photography at multiple altitudes, times and angles was carried out under preset weather conditions. The drone was equipped with a multi-channel multi-spectral camera to acquire images. When the drone is performing multi-altitude observation, keep the center point of the observation target unchanged, adjust the flight altitude of the drone vertically, and take photos at equal intervals; When the drone is conducting multi-time observation, keep the drone in the same position and take photos at equal intervals within the preset observation time range; When the drone is performing multi-angle observation, the drone is kept at a preset height and takes photos around and in the center of the observation target; Multi-height observations and multi-angle observations are completed within the preset time period.

[0018] Specifically, during the key periods of rice tillering, jointing, booting, heading, and grain filling (a total of 12 periods), drone observations were carried out under clear, cloudless, and windless conditions, including multi-height, multi-time, and multi-angle observations. The drone data was acquired by a 12-channel multispectral camera mounted on an octorotor drone. The central bands of the 12 channels involved were 490, 520, 550, 570, 670, 680, 700, 720, 800, 850, 900, and 950 nm. When the drone was used for multi-height observation, the center point of the observation target remained unchanged, and the flight altitude of the drone was adjusted vertically (50-250 m, with a shooting interval of 10 m). The higher the altitude, the closer the same position on the edge of the image was to vertical observation. When the multi-time observation was carried out, observations were carried out from 9:00 a.m. to 16:00 p.m. every day on the observation date (each observation interval was about 1 hour). In addition, the drone position was kept basically the same during each observation, and the observations at each time point were made at multiple heights. During multi-angle observations, the drone was kept at an altitude of 200 m, so that the study area was completely located on all sides (east, south, west, and north) and in the center (close to vertical observation) of the image.

[0019] In particular, within a single flight mission, the time for multi-altitude and multi-angle observations is controlled within 3 minutes.

[0020] Based on the above embodiment, the multi-spectral image of the multi-temporal and spatial angle UAV is preprocessed in step 100 to obtain the multi-spectral preprocessed image of the UAV, including: Performing vignetting correction, band registration and distortion correction on the multispectral image of the UAV in sequence; Using several calibration plates with known reflectivity to perform radiation calibration, a reflectivity image is obtained; The spectral information of the reflectance image is statistically analyzed by outlining the region of interest to obtain the multi-spectral pre-processed image of the UAV.

[0021] Specifically, the embodiment of the present invention uses a unified preprocessing process for all acquired drone images. Geometric processing includes vignetting correction, band registration and distortion correction, etc., and this process is completed in the PixelWrech2 (Tetracam, Inc., Chatsworth, CA, USA) software. Subsequently, 8 calibration plates with known reflectivity (reflectivities are 3%, 6%, 12%, 24%, 36%, 48%, 56% and 80%) are used for radiation calibration. After obtaining the reflectivity image, the spectral information at the field level is counted by outlining the region of interest (each field excluding the edge part is the field of interest), that is, the spectral mean of all pixels in the field of interest is the reflectivity at the field level.

[0022] Based on the above embodiment, step 200 includes: Obtaining multiple reflectances of the multispectral preprocessed image of the UAV in different bands; According to the difference and sum of the multiple reflectivities, the normalized difference red edge index in the rice canopy at multiple temporal and spatial angles is calculated.

[0023] Specifically, the embodiment of the present invention uses a combination of different bands, and the vegetation index can enhance the sensitivity to specific crop parameters and suppress the influence of other interference factors (for example, leaves and canopy structure). In recent years, vegetation indices based on red edge bands, such as the normalized difference red edge index NDRE and the red edge chlorophyll index CIred edge, have been widely used in crop growth monitoring. Relevant researchers have found that these indices are very effective in estimating plant growth parameters, especially when the vegetation is dense. In the present invention, NDRE is used as an input variable for LAI estimation at different time-space-angular scales, and the specific calculation formula is as follows: NDRE = (R850-R720) / (R850 + R720) In the formula, R720 and R850 represent the reflectance of the multispectral image in the corresponding bands.

[0024] Based on the above embodiment, the typical growth period of rice is obtained in step 300, and the estimation accuracy of the linear regression model of rice leaf area index in a single period based on the normalized difference red edge index is calculated, including: The leaf area index is obtained by using the normalized difference red edge index, and the change trend of the normalized difference red edge index of the rice canopy at different time and space angle scales is obtained by using the leaf area index; Randomly sampling rice plants in each field, and obtaining all leaves of the rice plants; Scanning all the leaves one by one to obtain the leaf area of ​​each rice hole, converting it with the rice planting density to obtain the leaf area index of each field; During the typical growth period of rice, the changing trend of the normalized difference red edge index of the rice canopy obtained at different drone heights, times and angles was analyzed for fields at different observation locations.

[0025] Specifically, the leaf area index LAI data is first obtained. The embodiment of the present invention uses LI3100C (LI-COR, Lincoln, NE, USA) to perform destructive sampling on rice to measure LAI. Before the measurement, three-hole rice plants are randomly sampled in each field, and the green leaves are cut off one by one after being brought back indoors (in the late stage of rice growth, yellow leaves will be removed), and then all leaves are scanned one by one to obtain the leaf area of ​​each hole of rice, and then converted with the rice planting density to finally obtain the LAI of each field.

[0026] Taking the typical growth stages of rice (tillering, jointing, booting, heading, and filling) as an example, for fields in different observation directions, such as forward scattering direction, backscattering direction, and vertical observation direction, the change trend of rice canopy NDRE obtained by different drone flight altitudes, times, and angles is analyzed. It is also necessary to consider the change of NDRE over time at different observation altitudes, as well as the change of NDRE with observation altitude under different time conditions.

[0027] Based on the above embodiment, the step 300 of comparing the estimation accuracy and determining that the rice leaf area index estimation accuracy is the highest is the best observation condition, includes: Under different time conditions, the accuracy of the linear regression estimation model of leaf area index based on the normalized difference red edge index in the typical period of rice was analyzed with different UAV observation altitudes; Under different altitude conditions, the accuracy of the linear regression estimation model of leaf area index based on the normalized difference red edge index in the typical period of rice is analyzed with the change trend of different drone observation times; Compare the leaf area index estimation accuracy under multi-angle observation at different observation times; The coefficient of determination, root mean square error and relative root mean square error were used to calculate the measured value, estimated value and mean value of leaf area index, and the model accuracy statistics were performed; The observation condition corresponding to the highest rice leaf area index estimation accuracy is taken as the optimal observation condition.

[0028] Specifically, first, under different time conditions, the accuracy of the LAI linear regression estimation model based on NDRE in the typical period of rice is analyzed with the change trend of different drone observation heights. Secondly, under different altitude conditions, the accuracy of the LAI linear regression estimation model based on NDRE in the typical period of rice is analyzed with the change trend of different drone observation times. Finally, the LAI estimation accuracy under multi-angle observation at different observation times is compared. Combined with the changes of NDRE with observation height, time and angle in multiple typical periods, the observation conditions suitable for estimating LAI of rice in different periods are determined. The final conclusion is: in each period, the estimation of rice LAI is suitable for using NDRE obtained under strong light conditions, that is, the backscattering direction (the observation direction is the same as the light direction).

[0029] The present invention adopts the coefficient of determination (Coefficient of determination, R 2 ), root mean square error (RMSE) and relative root mean square error (RRMSE) are used to calculate the model accuracy. The calculation formula is as follows:

[0030]

[0031]

[0032] in y , and represent the measured value, estimated value and average value of rice LAI, respectively, and n is the sample size.

[0033] Based on the above embodiment, step 400 includes: The flight strategies of UAVs in different typical periods were compared, and the multiple estimation accuracies of linear regression models of rice leaf area index in different combinations in a single period were compared; The UAV flight strategies include a combination of low altitude in the morning, a combination of high altitude in the morning, a combination of low altitude in the afternoon, a combination of high altitude in the afternoon, and a combination when the estimation accuracy of rice leaf area index and leaf chlorophyll content in a single period is the highest; The highest estimation accuracy of the rice multi-stage leaf area index under the optimal observation conditions is selected from the multiple estimation accuracies.

[0034] Specifically, the multi-period rice LAI estimation model is crucial for building a unified and universal model. In the actual study of rice growth parameter estimation, data is likely to be obtained at the same drone observation time or altitude in different periods. In order to explore the impact of data obtained at different observation altitudes and times in multiple periods on the accuracy of LAI estimation, the estimation accuracy of the rice LAI linear model under different combinations is compared by combining drone flight strategies in different typical periods. Different drone flight strategy combinations include low-altitude combination in the morning, high-altitude combination in the morning, low-altitude combination in the afternoon, high-altitude combination in the afternoon, and drone flight strategy combination corresponding to the highest estimation accuracy of rice LAI and LCC in a single period (called the preferred combination). Through accuracy comparison, it was found that in the preferred combination, the effect of NDRE in rice multi-period LAI estimation can be effectively improved by selecting observation methods in different periods.

[0035] Based on the above embodiment, step 500 includes: The estimation accuracy of observation data at different spatiotemporal angular scales is compared using multiple periods of dense observation data and a multi-head attention regression algorithm. Determine the best spatiotemporal angle normalized difference red edge index data combination in the deep learning model input data as the best strategy, and the remaining data combinations as the opposite best strategies; Inputting the data corresponding to the optimal strategy into the multi-head attention deep regression model, and outputting a high-precision estimate of the rice leaf area index in multiple periods; The normalized difference red edge index obtained by other conditions is obtained as the estimated result of the model input; The high-precision estimated rice leaf area index for multiple periods is compared with the estimated result to obtain the optimal rice leaf area index estimation result.

[0036] Specifically, the embodiment of the present invention is to build a high-precision LAI estimation model suitable for different rice stages, and use 12 periods of dense observation data and the Multihead Attention Regression (MHAR) algorithm to compare the estimation accuracy of observation data at different spatiotemporal angle scales. In the input data of the deep learning model, the best spatiotemporal angle NDRE data combination that meets the LAI estimation is the best strategy (that is, the data in the backscattering direction corresponding to LAI), and the data combination after removing the best strategy is the opposite best strategy. Our research results show that the LAI estimation accuracy of the optimal strategy (test accuracy is R 2 = 0.89, RMSE = 1.29, RRMSE = 16.78%) is significantly higher than the accuracy of the opposing optimal strategy model (test accuracy is R 2 = 0.78, RMSE = 1.82, RRMSE = 23.65%). In addition, when all spatiotemporal angle data are input into the MHAR model, the model accuracy (test accuracy is R 2 = 0.85, RMSE = 1.48, RRMSE = 19.23%) is not the highest, and is even lower than the optimal model accuracy. This shows that in deep learning, the reasonable selection of input variables has an important impact on model accuracy.

[0037] In order to test the performance of the deep learning framework proposed in the present invention, the accuracy was compared with a variety of widely used machine learning and deep learning models, and the input variables of all models were unified as the combined parameters corresponding to the optimal strategy in this study. The results show that compared with shallow machine learning algorithms (support vector regression SVR, random forest regression RFR, partial least squares regression PLSR), the deep learning model has more advantages in estimating multi-stage rice LAI, among which the deep neural network DNN model shows the highest estimation accuracy, but it is still lower than the accuracy of the MHAR model proposed in the present invention. This also highlights the important role of the multi-head attention mechanism in the estimation of multi-stage rice LAI.

[0038] The rice leaf area index estimation system for fusing spatiotemporal angle data provided by the present invention is described below. The rice leaf area index estimation system for fusing spatiotemporal angle data described below and the rice leaf area index estimation method for fusing spatiotemporal angle data described above can correspond to each other.

[0039] Figure 2FIG. 1 is a schematic diagram of the structure of a rice leaf area index estimation system integrating spatiotemporal angle data provided by an embodiment of the present invention. Figure 2 As shown, it includes: a preprocessing module 21, a calculation module 22, an acquisition module 23, a comparison module 24 and a construction module 25, wherein: The preprocessing module 21 is used to collect multi-temporal and spatial angle UAV multispectral images of the study area, and preprocess the multi-temporal and spatial angle UAV multispectral images to obtain UAV multispectral preprocessed images; the calculation module 22 is used to calculate the normalized difference red edge index of the rice canopy at multiple temporal and spatial angles in the UAV multispectral preprocessed images based on the reflectivity of the multispectral images in the corresponding bands; the acquisition module 23 is used to obtain the typical growth period of rice, calculate the estimation accuracy of the linear regression model of the rice leaf area index in a single period based on the normalized difference red edge index, and compare the estimation accuracy to determine the water The best observation condition is when the rice leaf area index estimation accuracy is the highest; the comparison module 24 is used to compare the estimation accuracy of the multi-period rice leaf area index linear regression model under various time and space angle combination conditions, and determine the highest estimation accuracy of the multi-period rice leaf area index under the said best observation conditions; the construction module 25 is used to construct a multi-head attention deep regression model, and input the normalized difference red edge index obtained under the said best observation conditions into the said multi-head attention deep regression model to obtain a high-precision estimation of the multi-period rice leaf area index, and after comparison, the best rice leaf area index estimation result is obtained.

[0040] Figure 3 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 3As shown, the electronic device may include: a processor 310, a communication interface 320, a memory 330 and a communication bus 340, wherein the processor 310, the communication interface 320 and the memory 330 communicate with each other through the communication bus 340. The processor 310 may call the logic instructions in the memory 330 to execute the rice leaf area index estimation method integrating the spatiotemporal angle data, the method comprising: collecting multi-spatial angle drone multispectral images of the study area, preprocessing the multi-spatial angle drone multispectral images to obtain the drone multispectral preprocessed images; calculating the normalized difference red edge index of the rice canopy at multiple spatiotemporal angles in the drone multispectral preprocessed images based on the reflectivity of the multispectral images in the corresponding bands; obtaining the typical growth period of rice, calculating the linear relationship between the single-period rice leaf area index based on the normalized difference red edge index. The estimation accuracy of the regression model is compared with the estimation accuracy to determine that the rice leaf area index estimation accuracy is the highest as the optimal observation condition; the estimation accuracy of the multi-period rice leaf area index linear regression model under various time-space angle combination conditions is compared to determine the highest estimation accuracy of the rice multi-period leaf area index under the optimal observation conditions; a multi-head attention deep regression model is constructed, and the normalized difference red edge index obtained under the optimal observation conditions is input into the multi-head attention deep regression model to obtain a high-precision estimation of the multi-period rice leaf area index, and the optimal rice leaf area index estimation result is obtained after comparison. In addition, the logic instructions in the above-mentioned memory 330 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0041] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0042] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0043] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A rice leaf area index estimation method integrating spatiotemporal angle data, characterized in that: include: Collect multi-spectral UAV images at multiple time and space angles in the study area, and pre-process the multi-spectral UAV images at multiple time and space angles to obtain pre-processed UAV multi-spectral images; Calculate the normalized difference red edge index of the rice canopy at multiple time and space angles in the multispectral preprocessed image of the UAV based on the reflectivity of the multispectral image in the corresponding band; Obtaining a typical growth period of rice, calculating the estimation accuracy of a linear regression model of rice leaf area index in a single period based on the normalized difference red edge index, comparing the estimation accuracy, and determining that the best observation condition is when the rice leaf area index estimation accuracy is the highest; The estimation accuracy of the multi-period rice leaf area index linear regression model under various time and space angle combination conditions is compared to determine the highest estimation accuracy of the multi-period rice leaf area index under the optimal observation conditions; A multi-head attention deep regression model is constructed, and the normalized difference red edge index obtained under the optimal observation conditions is input into the multi-head attention deep regression model to obtain a high-precision estimate of the rice leaf area index in multiple periods, and the optimal rice leaf area index estimation result is obtained after comparison.

2. The rice leaf area index estimation method integrating spatiotemporal angle data according to claim 1, characterized in that: Before collecting multi-temporal and spatial angle drone multispectral images of the study area and preprocessing the multi-temporal and spatial angle drone multispectral images to obtain drone multispectral preprocessed images, the method further includes: A plurality of fields of equal size are set up in the research area, and a plurality of different hybrid rice varieties are planted in the plurality of fields; A plurality of different nitrogen fertilizer levels are also set in the plurality of fields, and the same field management plan is adopted; Rice is planted in equally spaced holes in each field.

3. The rice leaf area index estimation method integrating spatiotemporal angle data according to claim 1, characterized in that: Collect multispectral drone images of the study area from multiple temporal and spatial angles, including: During different critical growth periods of rice, drone photography at multiple altitudes, times and angles was carried out under preset weather conditions. The drone was equipped with a multi-channel multi-spectral camera to acquire images. When the drone is performing multi-altitude observation, keep the center point of the observation target unchanged, adjust the flight altitude of the drone vertically, and take photos at equal intervals; When the drone is conducting multi-time observation, keep the drone in the same position and take photos at equal intervals within the preset observation time range; When the drone is performing multi-angle observation, the drone is kept at a preset height and takes photos around and in the center of the observation target; Multi-height observations and multi-angle observations are completed within the preset time period.

4. The rice leaf area index estimation method integrating spatiotemporal angle data according to claim 3, characterized in that: Preprocessing the multi-spectral image of the multi-temporal and spatial angle UAV to obtain the preprocessed multi-spectral image of the UAV includes: Performing vignetting correction, band registration and distortion correction on the multispectral image of the UAV in sequence; Using several calibration plates with known reflectivity to perform radiation calibration, a reflectivity image is obtained; The multi-spectral pre-processed image of the UAV is obtained by outlining the region of interest and counting the spectral information of the reflectance image.

5. The rice leaf area index estimation method integrating spatiotemporal angle data according to claim 1, characterized in that: The normalized difference red edge index of the rice canopy at multiple time and space angles in the multispectral preprocessed image of the UAV is calculated based on the reflectivity of the multispectral image in the corresponding band, including: Obtaining multiple reflectances of the multispectral preprocessed image of the UAV in different bands; According to the difference and sum of the multiple reflectivities, the normalized difference red edge index in the rice canopy at multiple temporal and spatial angles is calculated.

6. The rice leaf area index estimation method integrating spatiotemporal angle data according to claim 1, characterized in that: Obtain the typical growth period of rice and calculate the estimation accuracy of the linear regression model of rice leaf area index in a single period based on the normalized difference red edge index, including: The leaf area index is obtained by using the normalized difference red edge index, and the change trend of the normalized difference red edge index of the rice canopy at different time and space angle scales is obtained by using the leaf area index; Randomly sampling rice plants in each field, and obtaining all leaves of the rice plants; Scanning all the leaves one by one to obtain the leaf area of ​​each rice hole, converting it with the rice planting density to obtain the leaf area index of each field; During the typical growth period of rice, the changing trend of the normalized difference red edge index of the rice canopy obtained at different drone heights, times and angles was analyzed for fields at different observation locations.

7. The rice leaf area index estimation method integrating spatiotemporal angle data according to claim 6, characterized in that: By comparing the estimation accuracy, it is determined that the best observation condition is when the rice leaf area index estimation accuracy is the highest, including: Under different time conditions, the accuracy of the linear regression estimation model of leaf area index based on the normalized difference red edge index in the typical period of rice was analyzed with different UAV observation altitudes; Under different altitude conditions, the accuracy of the linear regression estimation model of leaf area index based on the normalized difference red edge index in the typical period of rice is analyzed with the change trend of different drone observation times; Compare the leaf area index estimation accuracy under multi-angle observation at different observation times; The coefficient of determination, root mean square error and relative root mean square error were used to calculate the measured value, estimated value and mean value of leaf area index, and the model accuracy statistics were performed; The observation condition corresponding to the highest rice leaf area index estimation accuracy is taken as the optimal observation condition.

8. The rice leaf area index estimation method integrating spatiotemporal angle data according to claim 1, characterized in that: The estimation accuracy of the multi-period rice leaf area index linear regression model under various time and space angle combination conditions is compared to determine the highest estimation accuracy of the rice multi-period leaf area index under the optimal observation conditions, including: The flight strategies of UAVs in different typical periods were compared, and the multiple estimation accuracies of linear regression models of rice leaf area index in different combinations in a single period were compared; The UAV flight strategies include a combination of low altitude in the morning, a combination of high altitude in the morning, a combination of low altitude in the afternoon, a combination of high altitude in the afternoon, and a combination when the estimation accuracy of rice leaf area index and leaf chlorophyll content in a single period is the highest; The highest estimation accuracy of the rice multi-stage leaf area index under the optimal observation conditions is selected from the multiple estimation accuracies.

9. The rice leaf area index estimation method integrating spatiotemporal angle data according to claim 1, characterized in that: A multi-head attention deep regression model is constructed, and the normalized difference red edge index obtained under the optimal observation conditions is input into the multi-head attention deep regression model to obtain a high-precision estimation of the rice leaf area index in multiple periods, and the optimal rice leaf area index estimation result is obtained after comparison, including: The estimation accuracy of observation data at different spatiotemporal angular scales is compared using multiple periods of dense observation data and a multi-head attention regression algorithm. Determine the best spatiotemporal angle normalized difference red edge index data combination in the deep learning model input data as the best strategy, and the remaining data combinations as the opposite best strategies; Inputting the data corresponding to the optimal strategy into the multi-head attention deep regression model, and outputting a high-precision estimate of the rice leaf area index in multiple periods; The normalized difference red edge index obtained by other conditions is obtained as the estimated result of the model input; The high-precision estimated rice leaf area index for multiple periods is compared with the estimated result to obtain the optimal rice leaf area index estimation result.

10. A rice leaf area index estimation system integrating spatiotemporal angle data, characterized in that: include: A preprocessing module is used to collect multi-temporal and spatial angle UAV multispectral images of the research area, and preprocess the multi-temporal and spatial angle UAV multispectral images to obtain UAV multispectral preprocessed images; A calculation module, used for calculating the normalized difference red edge index of the rice canopy at multiple time and space angles in the multispectral preprocessed image of the UAV based on the reflectivity of the multispectral image in the corresponding band; An acquisition module is used to acquire a typical growth period of rice, calculate the estimation accuracy of a linear regression model of rice leaf area index in a single period based on the normalized difference red edge index, compare the estimation accuracy, and determine that the best observation condition is when the rice leaf area index estimation accuracy is the highest; A comparison module is used to compare the estimation accuracy of the multi-period rice leaf area index linear regression model under various time and space angle combination conditions, and determine the highest estimation accuracy of the multi-period rice leaf area index under the optimal observation conditions; A construction module is used to construct a multi-head attention deep regression model, and the normalized difference red edge index obtained under the optimal observation conditions is input into the multi-head attention deep regression model to obtain a high-precision estimate of the rice leaf area index in multiple periods, and the optimal rice leaf area index estimation result is obtained after comparison.

Images