Crop lodging parameter inversion method, system, terminal device and storage medium

Abstract

This invention discloses a method, system, terminal device, and storage medium for inverting crop lodging parameters. The method includes: acquiring satellite optical radar remote sensing information and UAV optical radar remote sensing information; extracting features from the satellite optical radar remote sensing information and UAV optical radar remote sensing information to obtain remote sensing image feature information; fusing multi-source heterogeneous information from the remote sensing image feature information to construct a remote sensing image dataset; inputting the remote sensing image dataset into a deep neural network model for model training to obtain a crop lodging parameter inversion model; acquiring crop lodging image data; and inputting the crop lodging image data into the crop lodging parameter inversion model to obtain crop lodging parameters. This invention obtains rich crop lodging information by fusing satellite optical radar remote sensing information and UAV optical radar remote sensing information, and then inverts the lodging parameters through a parameter inversion model to obtain high-precision crop lodging parameters.

Description

Technical Field

[0001] This invention relates to the field of agricultural remote sensing, and in particular to a method, system, terminal equipment, and storage medium for inverting crop lodging parameters. Background Technology

[0002] Lodging in crops such as corn, rice, and wheat is primarily determined by three factors: genetics, agronomic management, and climate conditions. Among these, lodging caused by extreme weather events severely impacts yield, quality, and cost. Crop lodging is defined as the permanent displacement of a crop stem from its vertical position due to stem bending and / or root displacement. Characteristic parameters include: lodging type (root / stem lodging), lodging angle, lodging area, and distribution of lodging regions. Timely, rapid, accurate, and quantitative monitoring and assessment of key information regarding crop lodging are crucial for disaster mitigation and prevention.

[0003] Changes in the physical characteristics (plant height, canopy structure, leaf area index, etc.) and physiological characteristics (pigment content, photosynthesis, transpiration, etc.) of crops after lodging disasters form the basis for lodging parameter inversion. Remote sensing is the primary technical means for monitoring and assessing crop lodging disasters. Existing agricultural remote sensing technologies mainly include optical remote sensing and radar remote sensing. Among them, optical remote sensing mainly utilizes the differences in the physiological characteristics of crops after lodging, while radar remote sensing mainly utilizes the differences in the physical characteristics of crops after lodging. Satellite optical remote sensing (multispectral, hyperspectral), satellite radar remote sensing (synthetic aperture radar, SAR), and UAV optical remote sensing (visible light, multispectral, and hyperspectral) are the three main existing remote sensing technologies. Satellite optical remote sensing and satellite radar remote sensing have been widely used for periodic quantitative monitoring of agricultural information across spatial-temporal-spectral scales. However, due to limitations in revisit cycles and resolution, satellite remote sensing data to some extent ignores the temporal-spatial heterogeneity of agricultural information and is usually only suitable for large-scale macroscopic observations before and after disasters. UAV optical remote sensing can acquire high temporal-spatial resolution remote sensing images of local areas, which can largely compensate for the shortcomings of satellite remote sensing in temporal-spatial resolution. With the development of miniature synthetic aperture radar technology, UAV SAR remote sensing technology is gradually moving towards practical applications. The complementary enhancement of satellite optical radar remote sensing and UAV optical radar remote sensing constitutes a four-in-one multi-source heterogeneous fusion scheme, which will become a key technical means for monitoring and assessing crop lodging disasters.

[0004] Therefore, existing technologies still need to be improved and enhanced. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method, system, terminal equipment and storage medium for inverting crop lodging parameters, in view of the above-mentioned defects of the prior art, so as to solve the problem of low accuracy of inverted parameters caused by incomplete monitoring of crop lodging parameters in the prior art.

[0006] In a first aspect, the present invention provides a method for inverting crop lodging parameters, wherein the method includes:

[0007] Acquire satellite optical radar remote sensing information and UAV optical radar remote sensing information, and extract features from the satellite optical radar remote sensing information and UAV optical radar remote sensing information to obtain remote sensing image feature information;

[0008] Multi-source heterogeneous information fusion is performed on the feature information of the remote sensing images to construct a remote sensing image dataset;

[0009] The remote sensing image dataset is input into a deep neural network model for model training to obtain a crop lodging parameter inversion model.

[0010] Acquire crop lodging image data, and input the crop lodging image data into the crop lodging parameter inversion model to perform lodging parameter inversion and obtain crop lodging parameters.

[0011] In one implementation, the acquisition of satellite optical radar remote sensing information and UAV optical radar remote sensing information, and the extraction of features from the satellite optical radar remote sensing information and UAV optical radar remote sensing information to obtain remote sensing image feature information, specifically includes:

[0012] Acquire satellite optical radar remote sensing information and UAV optical radar remote sensing information;

[0013] The satellite optical radar remote sensing information and the UAV optical radar remote sensing information are preprocessed to obtain satellite remote sensing images and UAV remote sensing images.

[0014] Feature extraction is performed on the satellite remote sensing image to obtain satellite remote sensing image feature information, which includes satellite vegetation index, satellite backscattering coefficient and satellite interferometric coherence coefficient;

[0015] Feature extraction is performed on the UAV remote sensing image to obtain UAV remote sensing image feature information, which includes UAV vegetation index, UAV backscattering coefficient and UAV interferometric coherence coefficient;

[0016] Remote sensing image feature information is obtained based on the satellite remote sensing image feature information and the UAV remote sensing image feature information.

[0017] In one implementation, the step of fusing multi-source heterogeneous information from the remote sensing image feature information and constructing a remote sensing image dataset specifically includes:

[0018] The satellite remote sensing image feature information and the UAV remote sensing image feature information are fused to obtain a hybrid dataset;

[0019] Obtain prior information on crop lodging, and perform post-processing on the mixed dataset based on the prior information on crop lodging to obtain a high-quality dataset. The post-processing includes orthorectification, cropping, cleaning, and labeling.

[0020] The high-quality dataset is grouped according to a preset ratio to obtain a remote sensing image training dataset, a remote sensing image verification dataset, and a remote sensing image test dataset.

[0021] In one implementation, the step of inputting the remote sensing image dataset into a deep neural network model for model training to obtain a crop lodging parameter inversion model specifically includes:

[0022] The remote sensing image training dataset and the remote sensing image verification dataset are input into the deep neural network model for model training;

[0023] The remote sensing image test dataset was used to evaluate the performance of the trained deep neural network model, and the model performance evaluation results were obtained.

[0024] Based on the performance evaluation results of the model, the parameters of the deep neural network model are adjusted to obtain the crop lodging parameter inversion model.

[0025] In one implementation, the step of inputting the crop lodging image data into the crop lodging parameter inversion model to perform lodging parameter inversion and obtain crop lodging parameters specifically includes:

[0026] The crop lodging image data is subjected to convolution and pooling processing to obtain a low-resolution feature map;

[0027] The low-resolution feature map is divided into several feature image blocks of the same size;

[0028] The feature image block is input into the linear embedding module of the crop lodging parameter inversion model to obtain a feature vector map;

[0029] The feature vector map is input into the Transformer encoder of the crop lodging parameter inversion model for encoding processing. The Transformer encoder includes a classification head based on a fully connected neural network (FC), a detection head based on RPN to automatically generate regions of interest (ROI), and a dense prediction segmentation head based on FCN.

[0030] The classification head based on the fully connected neural network (FC) outputs the type and severity of crop lodging.

[0031] The detection head based on RPN automatically generates the region of interest (ROI) and outputs the detection results of crop lodging areas;

[0032] The FCN-based dense prediction segmentation head outputs the crop lodging area segmentation results.

[0033] In one implementation, the method further includes:

[0034] Acquire historical crop lodging parameters, current crop remote sensing image data, and current meteorological data;

[0035] The historical crop lodging parameters, current crop remote sensing image data, and current meteorological data are input into the lodging risk assessment model to predict lodging risk and obtain crop lodging prediction results.

[0036] In one implementation, the lodging risk assessment model is based on an RNN neural network architecture.

[0037] Secondly, embodiments of the present invention also provide a crop lodging parameter inversion system, wherein the system includes:

[0038] The information acquisition module is used to acquire satellite optical radar remote sensing information and UAV optical radar remote sensing information, and to extract features from the satellite optical radar remote sensing information and UAV optical radar remote sensing information to obtain remote sensing image feature information.

[0039] The dataset construction module is used to perform multi-source heterogeneous information fusion on the remote sensing image feature information and construct a remote sensing image dataset.

[0040] The model training module is used to input the remote sensing image dataset into the deep neural network model for model training to obtain the crop lodging parameter inversion model;

[0041] The parameter inversion module is used to acquire crop lodging image data and input the crop lodging image data into the crop lodging parameter inversion model to perform lodging parameter inversion and obtain crop lodging parameters.

[0042] In one implementation, the information acquisition module includes:

[0043] The remote sensing information acquisition unit is used to acquire satellite optical radar remote sensing information and UAV optical radar remote sensing information;

[0044] The data preprocessing unit is used to preprocess the satellite optical radar remote sensing information and the UAV optical radar remote sensing information to obtain satellite remote sensing images and UAV remote sensing images.

[0045] The satellite image feature extraction unit is used to extract features from the satellite remote sensing image to obtain satellite remote sensing image feature information, which includes satellite vegetation index, satellite backscattering coefficient and satellite interferometric coherence coefficient.

[0046] The UAV image feature extraction unit is used to extract features from the UAV remote sensing image to obtain UAV remote sensing image feature information, which includes UAV vegetation index, UAV backscattering coefficient and UAV interference coherence coefficient.

[0047] The remote sensing image feature acquisition unit is used to obtain remote sensing image feature information based on the satellite remote sensing image feature information and the UAV remote sensing image feature information.

[0048] In one implementation, the database construction module includes:

[0049] The information fusion unit is used to fuse the feature information of the satellite remote sensing image and the feature information of the UAV remote sensing image to obtain a hybrid dataset;

[0050] The data post-processing unit is used to acquire prior information on crop lodging and to perform data post-processing on the mixed dataset based on the prior information on crop lodging to obtain a high-quality dataset. The data post-processing includes orthorectification, cropping, cleaning, and labeling.

[0051] The data segmentation unit is used to group the high-quality dataset according to a preset ratio to obtain a remote sensing image training dataset, a remote sensing image verification dataset, and a remote sensing image test dataset.

[0052] In one implementation, the model training module includes:

[0053] The model training unit is used to input the remote sensing image training dataset and the remote sensing image verification dataset into the deep neural network model for model training;

[0054] The performance evaluation unit is used to evaluate the performance of the trained deep neural network model using the remote sensing image test dataset, and obtain the model performance evaluation result.

[0055] The parameter adjustment unit is used to adjust the parameters of the deep neural network model based on the model performance evaluation results to obtain the crop lodging parameter inversion model.

[0056] In one implementation, the parameter inversion module includes:

[0057] The convolution processing unit is used to perform convolution and pooling processing on the crop lodging image data to obtain a low-resolution feature map.

[0058] An image segmentation unit is used to segment the low-resolution feature map into several feature image blocks of the same size;

[0059] The image embedding unit is used to input the feature image block into the linear embedding module of the crop lodging parameter inversion model to obtain a feature vector map;

[0060] The image encoding unit is used to input the feature vector map into the Transformer encoder of the crop lodging parameter inversion model for encoding processing. The Transformer encoder includes a classification head based on a fully connected neural network (FC), a detection head based on RPN to automatically generate regions of interest (ROI), and a dense prediction segmentation head based on FCN.

[0061] The first output unit is used to output the type and severity of crop lodging by the classification head based on the fully connected neural network (FC).

[0062] The second output unit is used to output the detection results of crop lodging areas from the detection head that automatically generates the region of interest (ROI) based on RPN.

[0063] The third output unit is used to output the crop lodging area segmentation results from the FCN-based dense prediction segmentation head.

[0064] In one implementation, the system further includes:

[0065] The data acquisition module is used to acquire historical crop lodging parameters, current crop remote sensing image data, and current meteorological data.

[0066] The lodging prediction module is used to input the historical crop lodging parameters, the current crop remote sensing image data, and the current meteorological data into the lodging risk assessment model to predict the lodging risk and obtain the crop lodging prediction results.

[0067] The lodging risk assessment model is based on an RNN neural network architecture.

[0068] Thirdly, embodiments of the present invention also provide a terminal device, wherein the terminal device includes a memory, a processor, and a crop lodging parameter inversion program stored in the memory and executable on the processor. When the processor executes the crop lodging parameter inversion program, it implements the steps of the crop lodging parameter inversion method of any of the above schemes.

[0069] Fourthly, embodiments of the present invention also provide a computer-readable storage medium, wherein the computer-readable storage medium stores a crop lodging parameter inversion program, and when the crop lodging parameter inversion program is executed by a processor, it implements the steps of the crop lodging parameter inversion method described in any of the above schemes.

[0070] Beneficial Effects: Compared with existing technologies, this invention provides a method for inverting crop lodging parameters. First, it acquires satellite optical radar remote sensing information and UAV optical radar remote sensing information, and extracts features from these two sources to obtain remote sensing image feature information. Then, it fuses the remote sensing image feature information using multi-source heterogeneous information to construct a remote sensing image dataset. Further, it inputs the remote sensing image dataset into a deep neural network model for training, obtaining a crop lodging parameter inversion model. Finally, it acquires crop lodging image data and inputs this data into the lodging parameter inversion model to invert lodging parameters, obtaining the crop lodging parameters. This invention, by fusing satellite optical radar and UAV optical radar remote sensing information, comprehensively acquires remote sensing data of crops, trains it to obtain a crop lodging parameter inversion model, and finally acquires crop lodging image data and inputs it into the model for lodging parameter inversion, thus obtaining higher-precision crop lodging parameters. Attached Figure Description

[0071] Figure 1 A flowchart illustrating a specific implementation of the crop lodging parameter inversion method provided in this invention.

[0072] Figure 2 The flowchart below shows the overall scheme of the crop lodging parameter inversion method provided in this embodiment of the invention.

[0073] Figure 3 A flowchart of satellite remote sensing data processing for the crop lodging parameter inversion method provided in this embodiment of the invention.

[0074] Figure 4 This is a schematic diagram of an UAV optical remote sensing flight mission planning method for the crop lodging parameter inversion method provided in an embodiment of the present invention.

[0075] Figure 5 This is a schematic diagram illustrating the data set construction for the crop lodging parameter inversion method provided in this embodiment of the invention.

[0076] Figure 6 This is a framework diagram of the crop lodging parameter inversion model provided in an embodiment of the present invention.

[0077] Figure 7A functional schematic diagram of the system provided in an embodiment of the present invention.

[0078] Figure 8 A schematic diagram of a terminal device provided in an embodiment of the present invention. Detailed Implementation

[0079] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0080] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0081] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0082] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0083] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrases "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0084] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0085] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0086] Because current technologies for monitoring crop lodging parameters rely on satellite optical remote sensing, satellite radar remote sensing, or UAV optical remote sensing, the data obtained is not comprehensive enough, resulting in low accuracy of the lodging parameters derived from them.

[0087] To address the problems of existing technologies, this embodiment provides a method for inverting crop lodging parameters. This method comprehensively acquires remote sensing data of crops, trains a crop lodging parameter inversion model, and then inverts the lodging parameters to obtain higher accuracy. In specific implementation, firstly, satellite optical radar remote sensing information and UAV optical radar remote sensing information are acquired, and features are extracted from these information to obtain remote sensing image feature information. Then, multi-source heterogeneous information fusion is performed on the remote sensing image feature information to construct a remote sensing image dataset. Further, the remote sensing image dataset is input into a deep neural network model for model training to obtain the crop lodging parameter inversion model. Finally, crop lodging image data is acquired and input into the crop lodging parameter inversion model to invert the lodging parameters, thus obtaining the crop lodging parameters. This invention obtains richer crop lodging information by fusing satellite optical radar remote sensing information and UAV optical radar remote sensing information, and then trains a crop lodging parameter inversion model. Finally, it acquires crop lodging image data and inputs it into the crop lodging parameter inversion model to obtain crop lodging parameters, thus obtaining high-precision crop lodging parameters.

[0088] For example, firstly, satellite optical radar remote sensing information and UAV optical radar remote sensing information are acquired through satellites and UAVs, and features are extracted from the acquired remote sensing information to obtain remote sensing image feature information; then, all the extracted remote sensing image feature information is fused using multi-source heterogeneous information to construct a remote sensing image dataset; further, a deep neural network model is trained using the acquired data to obtain a crop lodging parameter inversion model; finally, after acquiring crop lodging image data, it can be input into the trained crop lodging parameter inversion model to invert lodging parameters, ultimately obtaining the crop lodging parameters. This invention obtains richer information on crop lodging by fusing satellite optical radar remote sensing information and UAV optical radar remote sensing information. This information is then used to train a crop lodging parameter inversion model. Finally, crop lodging image data is acquired and input into the model to retrieve the lodging parameters, resulting in high-precision lodging parameters. The complementary enhancement of satellite and UAV remote sensing allows for the simultaneous acquisition of high temporal-spatial-spectral resolution images of large-scale and key areas, ultimately leading to high-precision crop lodging parameters.

[0089] Exemplary methods

[0090] This embodiment provides a method for inverting crop lodging parameters, which can be applied to terminal devices. Specifically, as follows... Figure 1 As shown, the method includes:

[0091] Step S100: Acquire satellite optical radar remote sensing information and UAV optical radar remote sensing information, and extract features from the satellite optical radar remote sensing information and UAV optical radar remote sensing information to obtain remote sensing image feature information.

[0092] Optical remote sensing is the most mature remote sensing technology used in agriculture, primarily focusing on the physiological characteristics of crops. However, in the early stages of crop lodging, changes in physiological characteristics are relatively weak, leading to insignificant differences in the spectral responses of healthy and damaged plants in optical remote sensing images. Furthermore, commonly used spectral vegetation indices tend to saturate when crop biomass is high. Radar remote sensing, on the other hand, is sensitive to changes in crop physical parameters, and radar vegetation indices have higher saturation points compared to spectral vegetation indices. Moreover, optical remote sensing is a passive technology, where lighting, cloud, and fog conditions directly determine image quality, while radar remote sensing is an active technology, unaffected by lighting, cloud, fog, or rain conditions, allowing for all-weather, all-time Earth observation. Therefore, complementary enhancement of optical and radar remote sensing can acquire multiple types of remote sensing images characterizing the differences in physical and physiological parameters of lodged crops, which is beneficial for improving the accuracy of parameter inversion at different lodging stages.

[0093] Satellite remote sensing is suitable for large-scale monitoring, but its resolution is low (meter-level) and the revisit period is usually several days, making it difficult to acquire high-resolution remote sensing images immediately after a disaster. On the other hand, UAVs have high mobility and can carry high-resolution (centimeter-level) sensors to immediately collect remote sensing images of the observation area after a disaster. However, they are usually only suitable for local and small-scale remote sensing operations. If satellite remote sensing and UAV remote sensing are complemented and enhanced, high temporal-spatial-spectral resolution remote sensing images of large-scale and key areas can be acquired simultaneously, which is beneficial to further improve the accuracy of parameter inversion at different lodging stages of crops.

[0094] In one implementation, acquiring satellite optical radar remote sensing information and UAV optical radar remote sensing information, and extracting features from the satellite optical radar remote sensing information and UAV optical radar remote sensing information to obtain remote sensing image feature information, specifically includes: acquiring satellite optical radar remote sensing information and UAV optical radar remote sensing information; performing data preprocessing on the satellite optical radar remote sensing information and the UAV optical radar remote sensing information to obtain satellite remote sensing images and UAV remote sensing images; extracting features from the satellite remote sensing images to obtain satellite remote sensing image feature information, which includes satellite vegetation index, satellite backscattering coefficient, and satellite interferometric coherence coefficient; extracting features from the UAV remote sensing images to obtain UAV remote sensing image feature information, which includes UAV vegetation index, UAV backscattering coefficient, and UAV interferometric coherence coefficient; and obtaining remote sensing image feature information based on the satellite remote sensing image feature information and the UAV remote sensing image feature information. Therefore, simultaneously acquiring satellite optical radar and UAV optical radar remote sensing information is beneficial for improving the accuracy of parameter inversion at different lodging stages of crops.

[0095] In specific implementation, such as Figure 2As shown, satellite optical radar remote sensing information includes both satellite optical remote sensing information and satellite radar remote sensing information, while UAV optical radar remote sensing information includes both UAV optical remote sensing information and UAV radar remote sensing information. The acquired satellite optical radar remote sensing information provides large-scale, high-spectral-resolution crop remote sensing data, while the acquired UAV optical radar remote sensing information provides localized, small-area, high-spatial-temporal-resolution crop remote sensing data. Therefore, simultaneously acquiring both satellite optical radar and UAV optical radar remote sensing information and extracting features for parameter inversion is beneficial for improving the accuracy of parameter inversion at different lodging stages of crops. Examples of satellites that can provide optical remote sensing information include: ESA's Sentinel-2A / B, the US Landsat series, and China's GF-1 and GF-2; examples of satellites that can provide radar remote sensing information include: ESA's Sentinel-1A / B, Canada's RADARSAT-2, Japan's ALOS-2, and China's GF-3. Although the performance, data format, and revisit cycle of each remote sensing satellite differ, data preprocessing can be performed using mature commercial software, such as ENVI. This application uses the ESA Sentinel-1 / 2 as an example to obtain satellite optical radar remote sensing imagery. This software can obtain high-resolution images, the data is publicly available and freely accessible, and it is widely used in agriculture; the data processed by it is reliable and stable. First, multiple Sentinel-1 SLC and GRD images of the same target area and Sentinel-2 MSI products were downloaded from the ESA website. The obtained satellite optical radar remote sensing information was then preprocessed. The Sentinel-1 / 2 preprocessing workflow is as follows: Figure 3As shown. For Sentinel-1SLC products: Synthetic Aperture Radar data is analyzed using InSAR pairs of complex images from heavy-orbit SAR. For example, two adjacent SAR images before and after crop lodging can be combined into a complex image pair, and interferometric measurements can be performed to obtain interferometric fringes, thereby extracting the canopy deformation in the target area caused by crop lodging. During data preprocessing, taking the ESA SNAP software as an example, the general processing steps of SNAP software are: a) importing complex image pairs, b) applying orbital files, c) TOPS registration, d) interferogram formation, e) TOPS Deburst, f) multilooking, g) Goldstein Phase Filtering, h) unwrapping, i) phase to displacement, j) range Doppler terrain correction, and k) subset of region of interest. For Sentinel-1GRD products: Synthetic Aperture Radar (SAR) data, excluding phase information, allows extraction of radar intensity images and backscattering coefficients for the observed area. The typical SNAP software processing steps are: a) Data import, b) Apply Orbit File, c) Thermal Noise Removal, d) Speckle Filter, e) Radiometric and Geometric Correction, f) Subset cropping, and g) Coregistrapping. For Sentinel-2MSI products: Multispectral data, totaling 13 bands, with a maximum resolution of 10m. The typical SNAP software processing steps are: a) Data import, b) Resampling (resamples all bands to 10m resolution), c) Subset cropping, and d) Collocation. After processing by the above software, satellite remote sensing images can be obtained. Then, the corresponding image features can be extracted to obtain satellite remote sensing image feature information, which includes satellite vegetation index, satellite backscattering coefficient, and satellite interferometric coherence coefficient, etc.Satellite optical remote sensing features mainly consist of various vegetation indices, such as: Normalized Difference Vegetation Index (NDVI): NDVI = (NIR - Red) / (NIR + Red), Normalized Difference Red Edge Vegetation Index (NDRE): NDRE = (NIR - Red Edge) / (NIR + Red Edge), Green Normalized Difference Vegetation Index (GNDVI): GNDVI = (NIR - Green) / (NIR + Green), Ratio Vegetation Index (RVI): RVI = NIR / Red, Difference Environmental Vegetation Index (DVI): DVI = NIR - Red, and Soil-Regulated Vegetation Index (SAVI): SAVI = (NIR - Red) * (1 + L) / (NIR + Red + L). Here, NIR represents the reflectance in the near-infrared band, Red represents the reflectance in the red band, Red Edge represents the reflectance in the red edge band, Green represents the reflectance in the green band, and L is a constant typically used to adjust the denominator to avoid division by zero and may help stabilize the index calculation. By calculating various vegetation indices, the growth status, coverage, biomass, and health condition of vegetation can be quantified. Satellite radar remote sensing features mainly include: backscattering coefficients under different polarization conditions. and in Represents the backscattering coefficient under HH polarization, and similarly Represents the backscattering coefficient under HV polarization; calculate the ratio of backscattering coefficients: etc.; Radar Vegetation Index (RVI): Calculate the interference coefficient γ under different polarization conditions HH γ HV γ VH γ VV ,

[0096]

[0097] s * s is the complex conjugate of s, <...> is the sliding window spatial average, and s1 and s2 are two SAR images of the same observation scene at different time phases. The interferometric coherence coefficient γ ranges from 0 to 1 (1 indicates complete coherence). For UAV optical radar remote sensing information, UAV-RGB and UAV-MSI are used to provide UAV optical remote sensing information, while UAV-Mini-SAR is used to provide UAV radar remote sensing information. The data acquisition and preprocessing process is as follows: For UAV optical remote sensing (RGB / MSI), taking a multi-rotor UAV as an example, the UAV is used as the flight carrier, carrying an RGB camera and a multispectral camera simultaneously or separately. Combining camera parameters, flight altitude, and heading / lateral overlap requirements, the flight path and flight speed are planned. The specific method is as follows: Figure 4As shown. Whether the flight trajectory and flight speed settings are reasonable and accurate will directly affect the image quality. α represents the horizontal field of view of the camera, H represents the flight altitude, PH represents the height resolution of the camera image, Pw represents the width resolution of the camera image, Rsampling represents the sampling rate of the camera, FPw represents the width of the ground projection image, FPH represents the height of the ground projection image, OLPC represents the overlap between two adjacent images in the flight direction, OLPT represents the overlap between two adjacent images in the side direction, DBC represents the distance between two adjacent images in the flight direction, DBT represents the distance between two adjacent images in the side direction, and Vmax represents the maximum flight speed. The preprocessing of UAV optical remote sensing images can be completed on the workstation with the help of professional software (such as Pix4DMapper). The main tasks include: orthorectification stitching, quality assessment, vegetation index generation (the same as the extraction of vegetation index from satellite optical remote sensing), etc. The calculation method of UAV optical remote sensing flight trajectory parameters is as follows (2):

[0098]

[0099] For UAV radar remote sensing, taking multi-rotor UAVs as an example, the UAV is used as the flight platform, carrying Mini-SAR sensors of different bands simultaneously or independently. After planning the flight path, the observation area is imaged. Extractable features include backscattering coefficient, polarization, and interferometric coherence. The data processing method is the same as that for satellite radar remote sensing.

[0100] Step S200: Perform multi-source heterogeneous information fusion on the remote sensing image feature information and construct a remote sensing image dataset.

[0101] In one implementation, the multi-source heterogeneous information fusion of the remote sensing image feature information and the construction of a remote sensing image dataset specifically includes: fusing the satellite remote sensing image feature information and the UAV remote sensing image feature information to obtain a hybrid dataset; acquiring prior information on crop lodging, and performing data post-processing on the hybrid dataset based on the prior information on crop lodging to obtain a high-quality dataset, wherein the data post-processing includes orthorectification, cropping, cleaning, and labeling; and grouping the high-quality dataset according to a preset ratio to obtain a remote sensing image training dataset, a remote sensing image verification dataset, and a remote sensing image test dataset. Data fusion mainly includes two dimensions: first, the fusion of optical remote sensing data and radar remote sensing data; and second, the fusion of satellite data and UAV data. For the first dimension, optical remote sensing mainly extracts spectral reflectance features that reflect crop physiological parameters (such as chlorophyll content); while radar remote sensing mainly extracts radar scattering features that reflect crop physical parameters (such as plant height, canopy structure, etc.). The fusion of the two can provide a more comprehensive characterization of lodged crops. For the second dimension: satellite remote sensing images have a spatial resolution of meters and a temporal resolution of several days, while UAV remote sensing images can reach a resolution of centimeters or even millimeters and have a higher temporal resolution (acquiring data on demand). The fusion of the two mainly involves image registration technology and super-resolution technology.

[0102] In specific implementation, such as Figure 5 As shown, after obtaining the satellite remote sensing image feature information and the UAV remote sensing image feature information, they are fused to obtain a hybrid dataset. After acquiring the hybrid dataset with long-term SAR, MSI, and RGB data, and high spatial-temporal-spectral resolution, prior information on crop lodging is obtained. Then, the hybrid dataset is post-processed based on this prior information, including orthorectification, cropping, cleaning, labeling, data augmentation, noise reduction, variation correction, and generation. The post-processed dataset is a high-quality dataset. Finally, the high-quality dataset is grouped according to a preset ratio, such as 6:2:2 in this application, into training, validation, and test datasets. The prior knowledge includes human experience and human observation data of the study area. The prior information on crop lodging includes: crop variety, process management (water, fertilizer, pesticide application) information, and field measurement data after lodging. Combining UAVs and satellite remote sensing can obtain SAR, MSI, and RGB images with high temporal-spatial-spectral resolution at different spatial scales. This allows for the extraction of rich feature information such as radar scattering coefficient, polarization, coherence, radar vegetation index, spectral reflectance, spectral vegetation index, color, and texture. By fusing the above multi-source heterogeneous information, the timeliness, accuracy, reliability, and generalization of crop lodging parameter inversion and risk prediction can be effectively improved.

[0103] Step S300: Input the remote sensing image dataset into a deep neural network model for model training to obtain a crop lodging parameter inversion model.

[0104] Remote sensing image interpretation is another significant challenge in the field of agricultural remote sensing. The mainstream approach involves extracting significant lodging characteristics from radar optical remote sensing images, constructing a correlation model between radar / spectral vegetation indices and crop biophysical parameters, and indirectly retrieving the lodging area and severity. However, the accuracy and robustness of this retrieving model still need improvement. With the rapid development of artificial intelligence, artificial neural networks and deep learning algorithms are increasingly being applied to target classification, recognition, and segmentation in remote sensing images. However, due to disciplinary barriers, deep learning algorithms are currently mainly used for UAV remote sensing image interpretation, and the neural network models and deep learning algorithms applied are significantly lagging behind cutting-edge technologies in computer vision.

[0105] CNNs are widely used in downstream tasks of computer vision, such as image / video classification, detection, segmentation, and tracking. CNNs extract features by sharing convolutional kernels, reducing the number of network parameters and improving model efficiency. The success of CNNs relies on their two inherent inductive biases: translation invariance and local correlation. However, the limited receptive field of CNN convolutional kernels causes them to focus more on spatial local information and struggle to capture global data relationships. Transformers are self-attention-based encoder-decoder architectures, initially applied in natural language processing. In recent years, some studies (such as ViT, DETR, SETR, and SwinTransformer) have successfully extended them to computer vision. The significant advantage of visual Transformers lies in their multi-head self-attention mechanism, which can capture long-range dependencies of global image information. However, because visual Transformers lack the inductive biases of CNNs, they typically require extensive training data to match CNN performance. Furthermore, the computational complexity of visual Transformer models is related to the square of the input image size, making them unsuitable for processing high-resolution images. Therefore, this application proposes a hybrid architecture of CNN and Transformer for classifying lodging types, classifying lodging severity, and identifying lodging regions. This approach maximizes the inheritance of the local and global feature modeling capabilities of CNN and Transformer while reducing data dependency and computational complexity. The network structure is as follows: Figure 6 As shown.

[0106] In one implementation, the step of inputting the remote sensing image dataset into a deep neural network model for model training to obtain a crop lodging parameter inversion model specifically includes: inputting the remote sensing image training dataset and the remote sensing image validation dataset into the deep neural network model for model training; using the remote sensing image test dataset to evaluate the model performance of the trained deep neural network model and obtain the model performance evaluation result; and adjusting the parameters of the deep neural network model based on the model performance evaluation result to obtain the crop lodging parameter inversion model. The remote sensing image dataset is used to train the deep neural network model to obtain a model that can accurately invert crop lodging parameters.

[0107] In practical implementation, remote sensing image training and validation datasets are input into the deep neural network model for training. Then, a remote sensing image test dataset is used to evaluate the performance of the trained model. Based on the evaluation results, the model parameters are adjusted to obtain better-performing parameters, which are then set as the model parameters. This results in a crop lodging parameter inversion model for crop lodging parameter inversion. Performance evaluation metrics for the deep neural network model include average accuracy, computational complexity, and inference time. "Parameter tuning" mainly refers to adjusting the network parameters and hyperparameters of the deep neural network. Network parameters include: neural network depth, activation function, regularization, etc.; hyperparameters include: learning rate, number of iterations, optimizer, etc. The purpose of model parameter tuning is to validate the model performance on the test dataset after training and validation datasets, and analyze whether overfitting or underfitting exists. After adjusting the model parameters based on the evaluation results, the model is retrained and then evaluated again to achieve comparable performance on the test and validation datasets. The parameter-tuned deep neural network exhibits optimal performance and is used as the crop lodging parameter inversion model.

[0108] Step S400: Obtain crop lodging image data, and input the crop lodging parameter inversion model to perform lodging parameter inversion and obtain crop lodging parameters.

[0109] In one implementation, the step of inputting the crop lodging image data into the crop lodging parameter inversion model to invert lodging parameters and obtain crop lodging parameters specifically includes: performing convolution and pooling processing on the crop lodging image data to obtain a low-resolution feature map; segmenting the low-resolution feature map into several feature image blocks of the same size; inputting the feature image blocks into the linear embedding module of the crop lodging parameter inversion model to obtain a feature vector map; inputting the feature vector map into the Transformer encoder of the crop lodging parameter inversion model for encoding processing, wherein the Transformer encoder includes a classification head based on a fully connected neural network (FC), a detection head based on an RPN to automatically generate regions of interest (ROIs), and a dense prediction segmentation head based on an FCN; the classification head based on the fully connected neural network (FC) outputs the crop lodging type and lodging severity; the detection head based on the RPN to automatically generate regions of interest (ROIs) outputs the crop lodging region detection result; and the dense prediction segmentation head based on the FCN outputs the crop lodging region segmentation result.

[0110] In practice, the high-resolution input image is first subjected to convolution and pooling operations. This means the crop lodging image data is input into the model and subjected to convolution and pooling processes sequentially to obtain a low-resolution feature map. Then, the patch partition module divides the two-dimensional feature map into multiple feature image blocks of the same size. These blocks are then fed into a linear embedding module and undergo positional encoding before being used as input to the Transformer encoder. Since the Transformer encoder includes a classification head based on a fully connected neural network (FC), a detection head based on an RPN (Relationship of Interest) to automatically generate Regions of Interest (ROIs), and a dense prediction segmentation head based on an FCN (Focused Prediction Network), the three output heads output the lodging parameter inversion results. Specifically, the classification head based on the FC outputs the crop lodging type and severity; the detection head based on the RPN to automatically generate ROIs outputs the crop lodging area detection results; and the dense prediction segmentation head based on the FCN outputs the crop lodging area segmentation results. The encoder structure here mainly borrows the idea of ​​multi-head self-attention (MSA) computation in local windows from the Swin (Shifted Window) Transformer. Its main advantages are: firstly, it ensures a linear relationship between computational complexity and image size, thus significantly reducing computational complexity; secondly, based on window offset, it increases information exchange between windows, facilitating a trade-off between image resolution, computational complexity, and the receptive field of the self-attention space. The computation process of the Swin Transformer Block is as follows:

[0111]

[0112] in,

[0113]

[0114] The feature maps extracted by CNN and Transformer possess both local and global information correlations. Based on the requirements of lodging type, severity classification, region detection, and region segmentation tasks, a classification head based on a fully connected neural network (FC), a detection head based on RPN for automatically generating Regions of Interest (ROIs), and a dense prediction segmentation head based on FCN are designed. These three branches together constitute the basic framework of a crop lodging parameter inversion model based on a hybrid CNN and Transformer architecture. The lodging type classification, lodging severity classification, and lodging region identification and segmentation algorithms based on this hybrid CNN and Transformer architecture can maximize the inheritance of the local and global feature modeling capabilities of CNN and Transformer while reducing data dependence and computational complexity. In practical implementation, some optimization methods (dilated convolution, short cutting, feature pyramids, etc.) can be introduced to further reduce the model training difficulty and improve model performance.

[0115] In one implementation, historical crop lodging parameters, current crop remote sensing image data, and current meteorological data are obtained; these parameters are then input into a lodging risk assessment model to predict lodging risk, resulting in a crop lodging prediction.

[0116] Based on the foregoing, high-resolution, high-quality UAV and satellite SAR and MSI images can be acquired, enabling intelligent inversion of crop lodging parameters after lodging disasters occur. Furthermore, this application proposes a method for assessing and predicting seasonal lodging risks of crops based on long-term time-series UAV and satellite remote sensing data, thereby providing information support for lodging prevention before disasters occur. Addressing the problem of analyzing, assessing, and predicting seasonal lodging of spatially heterogeneous crops, this application proposes to introduce an RNN neural network model from the field of computer vision. It comprehensively utilizes currently observed crop vegetation parameters, soil parameters, environmental parameters, and historical lodging parameters from time series, designing a crop lodging risk assessment and prediction model based on GRU or LSTM. This establishes lodging data associations and dependencies in the time domain, assessing and predicting crop lodging risks in the short term. Both GRU and LSTM can memorize key information based on gating mechanisms and selectively forget useless information. Their characteristic is that they are no longer limited to matching between images at two adjacent time points, but extend to image information at multiple adjacent time points, possessing a wider temporal receptive field, thereby establishing long-term and short-term data associations between previous time-S image information and the current time-S image.

[0117]

[0118] Where, x t This is the input at the current moment; h t-1 It is the network output from the previous moment; W Z and W r These are resetting the gating system and updating the gating network parameters, respectively.

[0119]

[0120] x t It is the input at the current moment; h t-1 It is the network output from the previous moment; c t-1 It is the state parameter of the previous moment; (W) (f) U (f) ), (W (i) U (i) ), (W (o) U (o) These are the network parameters for the forget gate, input gate, and output gate, respectively.

[0121] In practice, historical crop lodging parameters, current remote sensing image data of crops, and current meteorological data are acquired. These parameters are then input into a lodging risk assessment model to predict lodging risk, yielding the predicted lodging results. The lodging risk assessment model is based on an RNN neural network architecture.

[0122] Exemplary System

[0123] like Figure 7 As shown in the illustration, this embodiment of the invention provides a system comprising: an information acquisition module 10, a database construction module 20, a model training module 30, and a parameter inversion module 40. Specifically, the information acquisition module 10 is used to acquire satellite optical radar remote sensing information and UAV optical radar remote sensing information, and to extract features from the satellite optical radar remote sensing information and UAV optical radar remote sensing information to obtain remote sensing image feature information; the database construction module 20 is used to perform multi-source heterogeneous information fusion on the remote sensing image feature information and to construct a remote sensing image dataset; the model training module 30 is used to input the remote sensing image dataset into a deep neural network for model training to obtain a crop lodging parameter inversion model; the parameter inversion module is used to acquire crop lodging image data and input the crop lodging image data into the crop lodging parameter inversion model to perform lodging parameter inversion to obtain crop lodging parameters.

[0124] In one implementation, the information acquisition module includes:

[0125] The remote sensing information acquisition unit is used to acquire satellite optical radar remote sensing information and UAV optical radar remote sensing information;

[0126] The data preprocessing unit is used to preprocess the satellite optical radar remote sensing information and the UAV optical radar remote sensing information to obtain satellite remote sensing images and UAV remote sensing images.

[0127] The satellite image feature extraction unit is used to extract features from the satellite remote sensing image to obtain satellite remote sensing image feature information, which includes satellite vegetation index, satellite backscattering coefficient and satellite interferometric coherence coefficient.

[0128] The UAV image feature extraction unit is used to extract features from the UAV remote sensing image to obtain UAV remote sensing image feature information, which includes UAV vegetation index, UAV backscattering coefficient and UAV interference coherence coefficient.

[0129] The remote sensing image feature acquisition unit is used to obtain remote sensing image feature information based on the satellite remote sensing image feature information and the UAV remote sensing image feature information.

[0130] In one implementation, the database construction module includes:

[0131] The information fusion unit is used to fuse the feature information of the satellite remote sensing image and the feature information of the UAV remote sensing image to obtain a hybrid dataset;

[0132] The data post-processing unit is used to acquire prior information on crop lodging and to perform data post-processing on the mixed dataset based on the prior information on crop lodging to obtain a high-quality dataset. The data post-processing includes orthorectification, cropping, cleaning, and labeling.

[0133] The data segmentation unit is used to group the high-quality dataset according to a preset ratio to obtain a remote sensing image training dataset, a remote sensing image verification dataset, and a remote sensing image test dataset.

[0134] In one implementation, the model training module includes:

[0135] The model training unit is used to input the remote sensing image training dataset and the remote sensing image verification dataset into the deep neural network model for model training;

[0136] The performance evaluation unit is used to evaluate the performance of the trained deep neural network model using the remote sensing image test dataset, and obtain the model performance evaluation result.

[0137] The parameter adjustment unit is used to adjust the parameters of the deep neural network model based on the model performance evaluation results to obtain the crop lodging parameter inversion model.

[0138] In one implementation, the parameter inversion module includes:

[0139] The convolution processing unit is used to perform convolution and pooling processing on the crop lodging image data to obtain a low-resolution feature map.

[0140] An image segmentation unit is used to segment the low-resolution feature map into several feature image blocks of the same size;

[0141] The image embedding unit is used to input the feature image block into the linear embedding module of the crop lodging parameter inversion model to obtain a feature vector map;

[0142] The image encoding unit is used to input the feature vector map into the Transformer encoder of the crop lodging parameter inversion model for encoding processing. The Transformer encoder includes a classification head based on a fully connected neural network (FC), a detection head based on RPN to automatically generate regions of interest (ROI), and a dense prediction segmentation head based on FCN.

[0143] The first output unit is used to output the type and severity of crop lodging by the classification head based on the fully connected neural network (FC).

[0144] The second output unit is used to output the detection results of crop lodging areas from the detection head that automatically generates the region of interest (ROI) based on RPN.

[0145] The third output unit is used to output the crop lodging area segmentation results from the FCN-based dense prediction segmentation head.

[0146] In one implementation, the system further includes:

[0147] The data acquisition module is used to acquire historical crop lodging parameters, current crop remote sensing image data, and current meteorological data.

[0148] The lodging prediction module is used to input the historical crop lodging parameters, the current crop remote sensing image data, and the current meteorological data into the lodging risk assessment model to predict the lodging risk and obtain the crop lodging prediction results.

[0149] The lodging risk assessment model is based on an RNN neural network architecture.

[0150] Based on the above embodiments, the present invention also provides a terminal device, the principle block diagram of which is shown in Figure 8. The terminal device may include one or more processors 100. Figure 8 (Only one is shown in the image), memory 101, and computer program 102 stored in memory 101 and executable on one or more processors 100, such as a program for a crop lodging parameter inversion method. When one or more processors 100 execute computer program 102, they can implement the various steps in the method embodiment. Alternatively, when one or more processors 100 execute computer program 102, they can implement the functions of various modules / units in the system embodiment, which is not limited here.

[0151] In one embodiment, the processor 100 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0152] In one embodiment, memory 101 may be an internal storage unit of an electronic device, such as a hard drive or RAM. Memory 101 may also be an external storage device of the electronic device, such as a plug-in hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc. Furthermore, memory 101 may include both internal and external storage units. Memory 101 is used to store computer programs and other programs and data required by the terminal device. Memory 101 can also be used to temporarily store data that has been output or will be output.

[0153] Those skilled in the art will understand that Figure 8 The block diagram shown is merely a partial structural diagram related to the present invention and does not constitute a limitation on the terminal device to which the present invention is applied. The specific terminal device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0154] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the methods described above. Any references to memory, storage, operational databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual operating data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0155] In summary, this invention discloses a method, system, terminal device, and storage medium for inverting crop lodging parameters. The method includes: first, acquiring satellite optical radar remote sensing information and UAV optical radar remote sensing information, and extracting features from the satellite optical radar and UAV optical radar remote sensing information to obtain remote sensing image feature information; then, fusing multi-source heterogeneous information from the remote sensing image feature information to construct a remote sensing image dataset; further, inputting the remote sensing image dataset into a deep neural network model for model training to obtain a crop lodging parameter inversion model; finally, acquiring crop lodging image data and inputting the crop lodging image data into the crop lodging parameter inversion model to invert lodging parameters, thereby obtaining crop lodging parameters. This invention obtains richer crop lodging information by fusing satellite optical radar remote sensing information and UAV optical radar remote sensing information, then trains it to obtain a crop lodging parameter inversion model, and finally acquires crop lodging image data and inverts crop lodging parameters using the crop lodging parameter inversion model to obtain high-precision crop lodging parameters.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for inverting lodging parameters of crops, characterized in that, The method includes: The system acquires satellite optical radar remote sensing information and UAV optical radar remote sensing information, and extracts features from these two sources to obtain remote sensing image feature information. The satellite optical radar remote sensing information is large-scale, high-spectral-resolution crop remote sensing data, while the UAV optical radar remote sensing information is local, small-area, high-spectral-resolution data. The crop remote sensing information has a time resolution; the UAV optical radar remote sensing information includes UAV optical remote sensing information and UAV radar remote sensing information; the satellite optical radar remote sensing information includes satellite optical remote sensing information and satellite radar remote sensing information. Multi-source heterogeneous information fusion is performed on the remote sensing image feature information to construct a remote sensing image dataset. Specifically, this includes: fusing satellite remote sensing image feature information and UAV remote sensing image feature information to obtain a hybrid dataset; acquiring prior information on crop lodging, and performing data post-processing on the hybrid dataset based on the prior information on crop lodging to obtain a high-quality dataset. The data post-processing includes orthorectification, cropping, cleaning, and labeling; and grouping the high-quality dataset according to a preset ratio to obtain a remote sensing image training dataset, a remote sensing image verification dataset, and a remote sensing image test dataset. The remote sensing image dataset is input into a deep neural network model for model training to obtain a crop lodging parameter inversion model. Acquire crop lodging image data and input the crop lodging image data into a crop lodging parameter inversion model to invert lodging parameters and obtain crop lodging parameters. Specifically, this includes: performing convolution and pooling processing on the crop lodging image data to obtain a low-resolution feature map; segmenting the low-resolution feature map into several feature image blocks of the same size; inputting the feature image blocks into the linear embedding module of the crop lodging parameter inversion model to obtain a feature vector map; inputting the feature vector map into the Transformer encoder of the crop lodging parameter inversion model for encoding processing. The Transformer encoder includes a classification head based on a fully connected neural network (FC), a detection head based on an RPN to automatically generate regions of interest (ROIs), and a dense prediction segmentation head based on an FCN. The classification head based on the FC outputs the crop lodging type and lodging severity; the detection head based on the RPN to automatically generate regions of interest (ROIs) outputs the crop lodging region detection result; and the dense prediction segmentation head based on the FCN outputs the crop lodging region segmentation result. The step of extracting features from the satellite optical radar remote sensing information and the UAV optical radar remote sensing information to obtain remote sensing image feature information specifically includes: preprocessing the satellite optical radar remote sensing information and the UAV optical radar remote sensing information to obtain satellite remote sensing images and UAV remote sensing images; extracting features from the satellite remote sensing images to obtain satellite remote sensing image feature information, which includes satellite vegetation index, satellite backscattering coefficient, and satellite interferometric coherence coefficient; extracting features from the UAV remote sensing images to obtain UAV remote sensing image feature information, which includes UAV vegetation index, UAV backscattering coefficient, and UAV interferometric coherence coefficient; and obtaining remote sensing image feature information based on the satellite remote sensing image feature information and the UAV remote sensing image feature information.

2. The method for inverting crop lodging parameters according to claim 1, characterized in that, The step of inputting the remote sensing image dataset into a deep neural network model for model training to obtain a crop lodging parameter inversion model specifically includes: The remote sensing image training dataset and the remote sensing image verification dataset are input into the deep neural network model for model training; The remote sensing image test dataset was used to evaluate the performance of the trained deep neural network model, and the model performance evaluation results were obtained. Based on the performance evaluation results of the model, the parameters of the deep neural network model are adjusted to obtain the crop lodging parameter inversion model.

3. The method for inverting crop lodging parameters according to claim 1, characterized in that, The method further includes: Acquire historical crop lodging parameters, current crop remote sensing image data, and current meteorological data; The historical crop lodging parameters, current crop remote sensing image data, and current meteorological data are input into the lodging risk assessment model to predict lodging risk and obtain crop lodging prediction results.

4. The method for inverting crop lodging parameters according to claim 3, characterized in that, The lodging risk assessment model is based on an RNN neural network architecture.

5. A crop lodging parameter inversion system, characterized in that, The system includes: The information acquisition module is used to acquire satellite optical radar remote sensing information and UAV optical radar remote sensing information, and to extract features from the satellite optical radar remote sensing information and UAV optical radar remote sensing information to obtain remote sensing image feature information; the satellite optical radar remote sensing information is large-scale and high-spectral-resolution crop remote sensing data, and the UAV optical radar remote sensing information is local small-area and high-spectral-resolution data. The crop remote sensing information has a time resolution; the UAV optical radar remote sensing information includes UAV optical remote sensing information and UAV radar remote sensing information; the satellite optical radar remote sensing information includes satellite optical remote sensing information and satellite radar remote sensing information. The dataset construction module is used to perform multi-source heterogeneous information fusion on the remote sensing image feature information and construct a remote sensing image dataset. Specifically, it includes: fusing satellite remote sensing image feature information and UAV remote sensing image feature information to obtain a hybrid dataset; acquiring prior information on crop lodging and performing data post-processing on the hybrid dataset based on the prior information on crop lodging to obtain a high-quality dataset. The data post-processing includes orthorectification, cropping, cleaning, and labeling; and grouping the high-quality dataset according to a preset ratio to obtain a remote sensing image training dataset, a remote sensing image verification dataset, and a remote sensing image test dataset. The model training module is used to input the remote sensing image dataset into the deep neural network model for model training to obtain the crop lodging parameter inversion model; The parameter inversion module is used to acquire crop lodging image data and input the crop lodging image data into the crop lodging parameter inversion model to invert lodging parameters and obtain crop lodging parameters. Specifically, it includes: performing convolution and pooling processing on the crop lodging image data to obtain a low-resolution feature map; segmenting the low-resolution feature map into several feature image blocks of the same size; inputting the feature image blocks into the linear embedding module of the crop lodging parameter inversion model to obtain a feature vector map; and inputting the feature vector map into the Transformer encoder of the crop lodging parameter inversion model for encoding processing. The Transformer encoder includes a classification head based on a fully connected neural network (FC), a detection head based on an RPN to automatically generate regions of interest (ROIs), and a dense prediction segmentation head based on an FCN. The classification head based on the fully connected neural network (FC) outputs the crop lodging type and lodging severity; the detection head based on the RPN to automatically generate regions of interest (ROIs) outputs the crop lodging area detection result; and the dense prediction segmentation head based on the FCN outputs the crop lodging area segmentation result. The information acquisition module includes a data preprocessing unit, a satellite image feature extraction unit, a UAV image feature extraction unit, and a remote sensing image feature acquisition unit. The data preprocessing unit preprocesses the satellite optical radar remote sensing information and the UAV optical radar remote sensing information to obtain satellite remote sensing images and UAV remote sensing images. The satellite image feature extraction unit extracts features from the satellite remote sensing images to obtain satellite remote sensing image feature information, including satellite vegetation index, satellite backscattering coefficient, and satellite interferometric coherence coefficient. The UAV image feature extraction unit extracts features from the UAV remote sensing images to obtain UAV remote sensing image feature information, including UAV vegetation index, UAV backscattering coefficient, and UAV interferometric coherence coefficient. The remote sensing image feature acquisition unit obtains remote sensing image feature information based on the satellite remote sensing image feature information and the UAV remote sensing image feature information.

6. A terminal device, characterized in that, The terminal device includes a memory, a processor, and a crop lodging parameter inversion program stored in the memory and executable on the processor. When the processor executes the crop lodging parameter inversion program, it implements the steps of the crop lodging parameter inversion method as described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, A computer-readable storage medium stores a crop lodging parameter inversion program, which, when executed by a processor, implements the steps of the crop lodging parameter inversion method as described in any one of claims 1-4.

Images