A method and system for evaluating crop decline rate based on multi-source remote sensing data

Abstract

The present invention proposes a method for evaluating the decline rate of crops based on multi-source remote sensing data, including: calculating the canopy green vegetation coverage fCover for farmland images during the growth period; constructing an fCover time series dynamic model to calculate the canopy green vegetation coverage fCover value corresponding to T, where T represents the cumulative value of the daily average temperature after the daily temperature exceeds the starting temperature T0 starting from the seedling emergence date; and calculating the crop decline rate based on the canopy green vegetation coverage fCover value. The present invention also correspondingly proposes a system for evaluating the decline rate of crops based on multi-source remote sensing data. The method and system of the present invention can quantitatively evaluate the decline rate of each piece of farmland, greatly reducing manual operations, quickly calculating the decline rate of each piece of farmland, and improving efficiency and accuracy. Moreover, the method of the present invention is less affected by the noise of the observation data and has high robustness.

Description

Technical Field

[0001] The present invention relates to the technical field of crop decline assessment, and more specifically, to a method and system for assessing crop decline rate based on multi-source remote sensing data. Background Art

[0002] The decline of crops is an important part of the growth and development process of crops, and the decline rate describes the speed at which crops develop from the peak of their growth period to withering, which is closely related to crop yields. For example, under stress conditions, a low crop decline rate means that they have a stronger ability to adapt to the environment and a higher cumulative yield. The decline rate of crops is the result of the combined effects of multiple factors such as variety, soil, water, and heat. Real-time, accurate, and efficient monitoring of the decline rate of crops is one of the important bases for screening stress-resistant crop varieties and evaluating crop growth. At the same time, combining the decline rate of crops with the time to reach the peak of the growth period can also effectively estimate the harvest time and plan the harvesting process, providing very valuable reference data for farmland management.

[0003] There are usually two ways to monitor crop decline rate. One is for experienced experts to observe crops in the field, determine whether they have begun to decline and estimate the decline rate. However, this method is highly subjective, and it is difficult to inherit and quantitatively describe experience. In addition, it is difficult to achieve multi-frequency monitoring. In addition, the area of ​​farmland in my country is usually large. Due to the influence of soil, tillage and other management methods, the growth of crops in a piece of farmland may vary greatly. This method is difficult to cover the entire farmland, and it is difficult to capture the differences within the farmland. The second method is to monitor parameters closely related to the crop decline rate, such as leaf area index and chlorophyll content, and calculate their slope to represent the decline rate. This method can be achieved through multi-frequency data collected by sensors mounted on satellites, drones and ground observation platforms. However, this method has the following problems: (1) It requires high accuracy of remote sensing data, that is, data noise will bring great uncertainty to the calculation of the slope. However, remote sensing data is affected by the collection environment, pre-processing, post-processing and other steps, and has great uncertainty itself, which in turn affects the calculation accuracy of the fading rate; (2) It is extremely dependent on the time of the peak growth of crops, and the time discontinuity of remote sensing data itself leads to the lack of data on the exact time of the peak growth, which will greatly affect the accuracy of the fading rate. Summary of the invention

[0004] In response to the problems in the background technology, the present invention proposes a method for evaluating the decline rate of crops based on multi-source remote sensing data, including: S1. Calculating the canopy green vegetation coverage fCover for farmland images during the growth period; S2. Constructing an fCover time series dynamic model to calculate the canopy green vegetation coverage fCover value corresponding to T, where T represents the cumulative value of the daily average temperature after the daily temperature exceeds the starting temperature (T0) starting from the seedling emergence date; S3. Calculating the crop decline rate based on the canopy green vegetation coverage fCover value calculated in S2.

[0005] The present invention also proposes a system for evaluating the decline rate of crops based on multi-source remote sensing data, including a computer executable program. When the program is executed, the following steps can be implemented: S1. Calculating the canopy green vegetation coverage fCover for farmland images within the growth period; S2. Constructing an fCover time series dynamic model to calculate the canopy green vegetation coverage fCover value corresponding to T, where T represents the cumulative value of the daily average temperature after the daily temperature exceeds the starting temperature (T0) starting from the seedling emergence date; S3. Calculating the crop decline rate based on the canopy green vegetation coverage fCover value calculated in S2.

[0006] The method of the present invention can quantitatively evaluate the fading rate of each farmland, greatly reducing manual operations, and can quickly calculate the fading rate of each farmland, and has a high consistency with the data observed by experienced experts, thereby improving efficiency and accuracy. In addition, the method of the present invention is less affected by the noise of the observed data and has high robustness.

[0007] Compared with the prior art, the innovations of the present invention include:

[0008] 1) Based on multiple observations during the growing season, coupled dynamic simulation and canopy radiation transfer model, crop growth curves are constructed to accurately estimate the time when crops reach their growth peak;

[0009] 2) The Area Under the Curve (AUC) method is used to describe the crop fading rate. This method is less affected by the noise of the observation data, has strong robustness, and can efficiently and accurately estimate the crop fading rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to make the present invention more easily understood, the present invention will be described in more detail by referring to the specific embodiments shown in the accompanying drawings. These drawings only depict typical embodiments of the present invention and should not be considered as limiting the scope of protection of the present invention.

[0011] Figure 1 is a flow chart of the method of the present invention.

[0012] Figure 2This is an example of the fCover timing dynamic model of the method of the present invention.

[0013] Figure 3 This is a schematic diagram of fading rate calculation according to the method of the present invention. DETAILED DESCRIPTION

[0014] The following describes the implementation modes of the present invention with reference to the accompanying drawings so that those skilled in the art can better understand the present invention and implement it. However, the enumerated embodiments are not intended to limit the present invention. In the absence of conflict, the following embodiments and the technical features therein may be combined with each other, wherein the same components are represented by the same figure marks.

[0015] like Figure 1 and Figure 2 As shown, the method of the present invention includes S1-S3.

[0016] S1. Calculate the canopy green vegetation coverage fCover for farmland images during the growth period.

[0017] The most intuitive manifestation of crop decline is the yellowing of leaves, so the present invention selects canopy green vegetation coverage fCover as the key parameter for calculating the decline rate. The farmland image is obtained by the following method:

[0018] 1) Collect RGB and multispectral images of farmland. You can use drones equipped with RGB and multispectral cameras to fly over farmland and collect data.

[0019] 2) Preprocess and stitch the RGB image and the multispectral image to generate an orthophoto. In addition, radiometric calibration is required for the multispectral image. Preferably, the commercial drone data processing software Agisoft is used for processing.

[0020] 3) According to the farmland boundaries, cut out the farmland images with corresponding numbers from the orthophoto.

[0021] 4) Downscale and resample the RGB image to the resolution of the multispectral image. Since the resolutions of RGB images and multispectral images are different, the RGB image needs to be downscaled and resampled to the resolution of the multispectral image. There are many resampling methods, such as nearest neighbor interpolation, bilinear interpolation, and cubic convolution interpolation. After testing, the present invention preferably uses bilinear interpolation to perform the resampling operation. After resampling, the spatial resolution of the RGB image and the multispectral image of each farmland is consistent.

[0022] There are two main methods for calculating canopy green vegetation coverage fCover. One is to use machine learning technology to perform image segmentation and distinguish the proportion of green and non-green parts. The proportion of green parts is the canopy green vegetation coverage fCover. This calculation method is more suitable for the early growth of crops. When crops grow lushly, the soil is difficult to segment. When crops begin to decline, the decline usually starts from the bottom of the crop, which cannot be observed from the images taken by drones, affecting the calculation accuracy of canopy green vegetation coverage fCover. Another method is to combine the canopy radiation transmission model and machine learning technology to invert the canopy green vegetation coverage fCover based on multispectral reflectivity. This method is often used to invert canopy green vegetation coverage fCover from satellite remote sensing data. The advantage of this method is that the canopy reflectivity is very sensitive to the changes in the color of the leaves inside the canopy, which helps to invert high-precision fCover throughout the growth period. However, this method collects the overall reflectivity of the canopy and is very sensitive to the soil background and crop fruit components of the canopy. In the application of satellite remote sensing, the research object is very large, so a variety of soil reflectance curves are usually input to build a simulation database. However, due to the high spatial resolution of drone images, this method is not applicable to drone images, so it needs to be improved.

[0023] The present invention proposes a method for combining the above two methods to perform canopy green vegetation coverage fCover inversion.

[0024] 1) For the RGB images of farmland collected from historical flight data, label the green crops, soil and other parts. Preferably, use LabelMe online open source software for labeling.

[0025] 2) Use the labeled data set as the training data set for model training and prediction. Preferably, the YoLov5 deep learning network is used for training and prediction. The main reason for using this network is its high precision and high efficiency performance. All labeled images are divided into two parts. One part (e.g. 75%) of the images is used as the training data set and input into the YoLov5 network for training. The remaining part (e.g. 25%) of the images is used to detect the accuracy of model training.

[0026] 4) Apply the trained model (e.g. YoLov5 network) to the RGB images of each farmland collected by the drone to extract the corresponding soil pixels.

[0027] 5) Record the positions of these soil pixels on the RGB image, find the corresponding soil pixels on the multispectral image, and calculate the mean reflectance of each band of these soil pixels as the soil background reflectance of the farmland. In addition, if the soil pixels on the multispectral image cannot be found due to dense crop growth, the soil reflectance of the nearest date is used as the soil background reflectance of the farmland on that date.

[0028] 6) Use the canopy radiation transfer model to establish a training data set. There are currently a variety of canopy radiation transfer models, and the PROSAIL model is preferred because of its high accuracy, high efficiency and ease of use, and has been extensively verified in the study of satellite remote sensing parameter inversion. The training data set can be constructed in the manner of Table 1, but the present invention uses the actual reflectivity of the soil in each farmland obtained in step 5) or 6) as the soil reflectivity and inputs it into the canopy radiation transfer model, thereby simulating a large number of reflectivity data sets.

[0029] Table 1

[0030]

[0031]

[0032] 7) Input the canopy green vegetation coverage fCover, multi-band reflectivity data, solar zenith angle, and solar azimuth information of the trained reflectivity data set into the machine learning model for training. The input data includes multi-band reflectivity data, solar zenith angle, and solar azimuth, and the output data is the canopy green vegetation coverage fCover. There are currently a large number of machine learning methods, including lookup tables, neural networks, Gaussian processes, etc. The Gaussian process model is preferred because the model can return uncertainty coefficients, has good training robustness, does not require a large amount of training data sets (does not require massive data sets), and has high efficiency.

[0033] 8) After the model training is completed, the measured reflectance data, solar zenith angle and solar azimuth of each farmland are input into the trained machine learning model to invert the canopy green vegetation coverage fCover and its uncertainty index of each farmland.

[0034] S2. Construct a time series dynamic model of fCover and calculate the canopy green vegetation coverage fCover value corresponding to T, where T represents the cumulative value of the daily average temperature after the daily temperature exceeds the starting temperature (T0) starting from the seedling emergence date.

[0035] S21. Based on the changing trend of canopy green vegetation coverage fCover during the growing season (whole growth period), a time series dynamic model of fCover is constructed.

[0036] In one embodiment, as shown in formula (1), a temporal dynamic model of canopy green vegetation coverage fCover is constructed.

[0037]

[0038] T represents the cumulative value of the daily average temperature after the daily temperature exceeds the starting temperature (T0) from the seedling emergence date. For crops such as wheat and corn, T0 is 8°C. m Represents that the green vegetation coverage of crop canopy fCover reaches the growth peak fCover max T, T s represents T when the crop completely declines. a and b are the coefficients to be estimated.

[0039] S22. Based on formula (1), for the canopy green vegetation coverage fCover obtained in the growing season (full growth period) in S1, use the least squares fitting method to estimate the coefficient to be estimated in formula (1): fCover max , a, b, T m and T s .

[0040] Preferably, in order to ensure the accuracy of the fitting, the following initial conditions are set:

[0041] (1)0≤fCover max ≤1

[0042] (2)T0<T M <T s

[0043] In order to better fit the model, at least 5 canopy green vegetation coverage fCover observation data are required, which has been considered and satisfied in the early stage of experimental design.

[0044] S23. Repeat step S22 in each farmland until 5 coefficients to be estimated for each farmland are obtained: fCover max , a, b, T m and T s .

[0045] S24. According to formula (1), calculate the canopy green vegetation coverage fCover value corresponding to each T. Figure 2 A fCover timing dynamic model is shown.

[0046] S3. Calculate the crop decline rate based on the canopy green vegetation coverage fCover value calculated in S2.

[0047] S31. Set the growth peak of each farmland to fCover max Normalization.

[0048] Due to the difference in the growth of crops in each farmland, the growth peak fCover max The values ​​may be different. In order to ensure the consistency of the fading rate calculation of different farmlands, firstly, the growth peak fCover of each farmland is calculated. max Normalized to 1, as shown in formula (2), all growth peaks fCover are normalized.

[0049]

[0050] Where fCover norm is the normalized fCover value.

[0051] S32. Select from T m Start to T s All fCover norm (T), For each farmland, the area under the AUC curve was calculated.

[0052] Specifically, the AUC function in the sklearn package in the Python program can be called to calculate the area between the curve and the x- and y-axes (Area Under Curve, AUC) for each piece of farmland.

[0053] S33. Calculate the fading rate R of each farmland according to formula (3) senescenCe .like Figure 3 shown.

[0054] R senescence =1-AUC (3)

[0055] The method of the present invention was run in a plurality of sunflower and corn fields, and the results were reliable, especially in small breeding plots, and had a high correlation with the decline rate of each variety calculated by genetic methods.

[0056] The present invention also proposes a system for evaluating the rate of crop decline based on multi-source remote sensing data, which includes an executable program. When the executable program is executed, the steps of the method of the present invention as described above are implemented.

[0057] The embodiments described above are only preferred specific implementations of the present invention. This specification uses the phrases "in one embodiment", "in another embodiment", "in yet another embodiment" or "in other embodiments" to refer to one or more of the same or different embodiments according to the present disclosure. Common changes and substitutions made by those skilled in the art within the scope of the technical solution of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for evaluating crop decline rate based on multi-source remote sensing data, characterized in that: include: S1. Calculate the canopy green vegetation coverage fCover for farmland images in the growth period; S2. Based on the changing trend of canopy green vegetation coverage fCover during the growing season, a time series dynamic model of fCover is constructed; the system of the time series dynamic model of canopy green vegetation coverage fCover is estimated using the least squares fitting method; the canopy green vegetation coverage fCover value corresponding to T is calculated, where T represents the cumulative value of the daily average temperature after the daily temperature exceeds the starting temperature T0 starting from the seedling date; wherein the time series dynamic model of canopy green vegetation coverage fCover is: Among them, T m Represents that the green vegetation coverage of crop canopy fCover reaches the growth peak fCover max T, T s T represents the complete decline of crops, a and b are coefficients; S3. Calculate the crop decline rate based on the canopy green vegetation coverage fCover value calculated in S2: max Normalization; Select from T m Start to T s All normalized fCover values ​​fCover norm (T), for each piece of farmland, calculate the area under the AUC curve AUC; calculate the fading rate R of each piece of farmland according to the following formula senescence : R senescence =1-AUC.

2. The method according to claim 1, characterized in that In step S1, the farmland image is obtained by the following method: 1) Collect RGB images and multispectral images of farmland; 2) Preprocess and stitch RGB images and multispectral images to generate orthophotos, and perform radiometric calibration on multispectral images; 3) According to the boundaries of the farmland, cut out the farmland images with corresponding numbers from the orthophoto; 4) Downscale and resample the RGB image to the resolution of the multispectral image.

3. The method according to claim 1, characterized in that In step S1, the steps of calculating the canopy green vegetation coverage fCover include: 1) For the RGB images of farmland, mark the green crops and soil; 2) Use the labeled data set as the training data set to perform model training and prediction; 3) Apply the trained model to the RGB image of each farmland to extract the corresponding soil pixels; 4) Record the position of the soil pixel on the RGB image, find the corresponding soil pixel on the multispectral image, and calculate the mean reflectance of each band of the soil pixel as the soil background reflectance of the farmland; 5) Use the canopy radiation transfer model to establish a training data set and perform Gaussian process model training. Use the trained Gaussian process model to invert the canopy green vegetation coverage fCover of each farmland.

4. The method according to claim 3, characterized in that In step 4), if no soil pixel on the multispectral image is found, the soil reflectance of the nearest date is used as the soil background reflectance of the farmland.

5. The method according to claim 3, characterized in that: Step 5) includes: Gaussian process model training is performed using multi-band reflectance data, solar zenith angle, and solar azimuth angle.

6. The method according to claim 1, characterized in that In step S2, the initial condition of the fitting is: 0≤fCover max ≤1, T0<T m <T s .

7. A system for evaluating crop decline rate based on multi-source remote sensing data, characterized in that: The invention comprises a computer executable program, and when the program is executed, the method according to any one of claims 1 to 6 can be implemented.

Images