Regional crop relative height prediction method and device, electronic equipment and storage medium

Abstract

The invention provides a regional crop relative height prediction method and device, electronic equipment and a storage medium, and the method comprises the steps: determining a cultivated land region based on vector data of an initial region and remote sensing cultivated land mask data corresponding to the initial region; acquiring laser sampling data corresponding to the cultivated land area; based on the historical phenological period of crops in the cultivated land area and the quality attribute information of the laser sampling data, performing quality filtering processing on the laser sampling data corresponding to the cultivated land area to obtain first laser sampling data; screening target laser sampling data based on the maximum geographic positioning deviation of the first laser sampling data; determining a crop relative height bidirectional difference label corresponding to the target laser sampling data; training a target estimation model for predicting the relative height of the crop based on the two-way difference label of the relative height of the crop; and reasoning a to-be-predicted region image by using the target estimation model to obtain a region crop relative height corresponding to the to-be-predicted region image.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method, device, electronic equipment and storage medium for predicting relative height of regional crops. Background Art

[0002] During the growth stage of crops, workers often need to rely on the relative height of crops to manage, classify and predict yields.

[0003] In related technologies, data obtained by intelligent sensor terminals are often used to directly estimate the relative height of crops. However, due to factors such as the large amount of data, spatiotemporal interference, or positioning deviation in the acquired data, inaccurate estimates of the relative height of crops may occur. Summary of the invention

[0004] The present invention provides a method, device, electronic device and storage medium for predicting the relative height of regional crops, which are used to solve the defect in the prior art that the relative height of crops is inaccurately estimated due to inaccurate sampling data, realize the deep fusion application of optical and radar data, and improve the accuracy of the estimation of the relative height of crops.

[0005] The present invention provides a method for predicting relative height of regional crops, comprising the following steps: Determine the cultivated land area based on the vector data of the initial area and the remote sensing cultivated land mask data corresponding to the initial area; the initial area includes the cultivated land area and the non-cultivated land area; Acquiring laser sampling data corresponding to the cultivated land area; Based on the historical phenological period of the crops in the cultivated land area and the quality attribute information of the laser sampling data, quality filtering processing is performed on the laser sampling data corresponding to the cultivated land area to obtain first laser sampling data; the historical phenological period is used to characterize the time period when the relative height value of the crops is stable; Based on the maximum geographic positioning deviation of the first laser sampling data, screening the target laser sampling data; Determine a crop relative height bidirectional difference label corresponding to the target laser sampling data; Based on the bidirectional difference labels of crop relative height, a target estimation model for predicting crop relative height is trained; and the target estimation model is used to infer the image of the area to be predicted to obtain the relative height of the crops in the area corresponding to the image of the area to be predicted.

[0006] According to a method for predicting relative height of regional crops provided by the present invention, target laser sampling data is screened based on the maximum geographic positioning deviation of first laser sampling data, including: determining a scanning radius of the first laser sampling data; determining a buffer radius for quality filtering of the first laser sampling data based on the scanning radius of the first laser sampling data and the maximum geographic positioning deviation of the first laser sampling data; and screening the target laser sampling data based on the buffer radius.

[0007] According to a method for predicting regional crop relative height provided by the present invention, a target estimation model for predicting crop relative height is trained based on crop relative height bidirectional difference labels, including: screening the target remote sensing image corresponding to the cultivated land area; calculating remote sensing spectral features based on the target remote sensing image, and setting the remote sensing spectral features as sample features corresponding to each crop relative height bidirectional difference label; the sampling points of the target laser sampling data correspond to the pixels of the target remote sensing image one by one in time and space; the crop relative height bidirectional difference labels and the sample features are input into an initial estimation model to obtain an estimated bidirectional difference label; based on the crop relative height bidirectional difference labels and the estimated bidirectional difference labels, loss information is determined; based on the loss information, the initial estimation model is continuously optimized to obtain the target estimation model.

[0008] According to a method for predicting the relative height of regional crops provided by the present invention, based on the historical phenological period of the crops in the cultivated land area and the quality attribute information of the laser sampling data, quality filtering processing is performed on the laser sampling data corresponding to the cultivated land area to obtain the first laser sampling data, including: based on the historical phenological period of the crops in the cultivated land area, a first filtering processing is performed on the laser sampling data to obtain the second laser sampling data; based on the laser sampling data of the full-power beam in the second laser sampling data and the corresponding quality attribute information, the second laser sampling data is subjected to a second filtering processing to obtain the third laser sampling data; based on the relative height of the crops corresponding to the third laser sampling data and the terrain information corresponding to the third laser sampling data, the third laser sampling data is subjected to a third filtering processing to obtain the first laser sampling data.

[0009] According to a method for predicting relative height of regional crops provided by the present invention, based on the laser sampling data of the full-power beam in the second laser sampling data and the corresponding quality attribute information, the second laser sampling data is filtered for a second time to obtain third laser sampling data, including: retaining the laser sampling data of the full-power beam in the second laser sampling data to obtain fourth laser sampling data; based on the quality attribute information corresponding to the fourth laser sampling data, filtering the fourth laser sampling data to obtain the third laser sampling data; the quality attribute information includes at least one of the following: detection mode, quality identification, waveform.

[0010] According to a method for predicting regional crop relative height provided by the present invention, the screening of the target remote sensing image corresponding to the cultivated land area includes: obtaining the remote sensing image corresponding to the initial area; preprocessing the remote sensing image to obtain the target remote sensing image; the preprocessing includes at least one of the following: spatial mask processing, temporal filtering processing, resampling processing, spatial interpolation processing and mean smoothing processing.

[0011] According to a method for predicting regional crop relative height provided by the present invention, the remote sensing spectral characteristics are calculated based on the target remote sensing image, including: determining different reflectivity bands in the target remote sensing image; determining the remote sensing spectral characteristics based on texture features in the target remote sensing image, terrain features in the target remote sensing image and different reflectivity bands in the target remote sensing image.

[0012] The present invention also provides a regional crop relative height prediction device, comprising the following modules: A first determination module is used to determine the cultivated land area based on the vector data of the initial area and the remote sensing cultivated land mask data corresponding to the initial area; the initial area includes the cultivated land area and the non-cultivated land area; An acquisition module, used for acquiring laser sampling data corresponding to the cultivated land area; A filtering module, configured to perform quality filtering on the laser sampling data corresponding to the cultivated land area based on the historical phenological period of the crops in the cultivated land area and the quality attribute information of the laser sampling data, so as to obtain first laser sampling data; the historical phenological period is used to characterize a time period in which the relative height value of the crops is stable; A second determination module, configured to screen target laser sampling data based on a maximum geographic positioning deviation of the first laser sampling data; A third determination module is used to determine a bidirectional difference label of a relative height of a crop corresponding to the target laser sampling data; The prediction module is used to train a target estimation model for predicting the relative height of crops based on the two-way difference labels of the relative height of crops; and use the target estimation model to infer the image of the area to be predicted to obtain the relative height of the area crops corresponding to the image of the area to be predicted.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any of the above-mentioned methods for predicting relative height of regional crops is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for predicting the relative height of regional crops as described in any one of the above is implemented.

[0015] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method for predicting relative height of regional crops as described above is implemented.

[0016] The method, device, electronic device and storage medium for predicting regional crop relative height provided by the present invention first determine the cultivated land area through vector data of the initial area and remote sensing cultivated land mask data corresponding to the initial area; then, use the historical phenological period of crops in the cultivated land area to perform spatiotemporal filtering on the laser sampling data of the cultivated land area to obtain first laser sampling data, thereby eliminating irrelevant data or inaccurate data and enhancing the operational efficiency of the model; and, use the geographic positioning deviation of the first laser sampling data to determine the target laser sampling data and improve the accuracy of the data; finally, use the highly accurate laser sampling data to produce a two-way difference label for the relative height of the crop, thereby enhancing the ability to characterize the relative height of the crop canopy and bottom layer, improving the accuracy and reliability of the label, and further improving the accuracy of the estimation of the relative height of the crop. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 This is one of the flow charts of the method for predicting relative height of regional crops provided by the present invention.

[0019] Figure 2 This is the second flow chart of the method for predicting relative height of regional crops provided by the present invention.

[0020] Figure 3 It is a schematic diagram of the prediction results of the target estimation model provided by the present invention.

[0021] Figure 4 It is a schematic diagram of estimating the relative height of crops in a certain area provided by the present invention.

[0022] Figure 5 It is a structural schematic diagram of the regional crop relative height prediction device provided by the present invention.

[0023] Figure 6 It is a schematic diagram of the physical structure of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] Combine the following Figure 1-Figure 4 The method for predicting the relative height of regional crops of the present invention is applicable to data processing and relative height estimation of any crop. The execution subject of the method can be an electronic device, or a device for predicting the relative height of regional crops set in the electronic device. The device for predicting the relative height of regional crops can be implemented by software, hardware, or a combination of the two.

[0026] Figure 1 FIG. 1 is one of the flow charts of the method for predicting relative height of regional crops provided by the present invention. Figure 1 As shown, the method includes the following: Step 101: Determine a cultivated land area based on vector data of an initial area and remote sensing cultivated land mask data corresponding to the initial area; the initial area includes a cultivated land area and a non-cultivated land area.

[0027] Here, the initial region may be any suitable region including a cultivated land region and a non-cultivated land region, such as a province region, a city region, etc. That is, when it is necessary to detect crops in a province, it is necessary to first determine the cultivated land area in the province region.

[0028] Here, the vector data of the initial area is essentially the geographic space data of the initial area, which records a series of coordinate points in a rectangular coordinate system to describe the position and shape of the geographic entity.

[0029] Here, the remote sensing image can be an image acquired by the Sentinel-2 satellite, which is an earth observation satellite in the Copernicus program of the European Space Agency. It consists of two polar-orbiting satellites, A and B. The revisit period of the two satellites can reach 5 days. The satellite can cover 13 bands, including visible light, near infrared, red edge, and short-wave infrared.

[0030] Here, the remote sensing cultivated land mask data is essentially a binary image with the same size and shape as the initial area, 0 represents a non-cultivated land area, and 1 or 255 represents a cultivated land area. The remote sensing cultivated land mask data can be obtained from the European Space Agency's global 10-meter global cultivated land grid dataset, or can be obtained through manual annotation.

[0031] Specifically, the remote sensing image corresponding to the initial area is multiplied by the pixel points of the remote sensing cultivated land mask data corresponding to the initial area to obtain a binary image of 0 and non-0, and the non-0 area image is determined as the cultivated land area image. Then, the actual geographical location of the cultivated land area image in the initial area is determined according to the vector data of the initial area and the cultivated land area image to obtain the cultivated land area.

[0032] In addition, the remote sensing intelligent processing cloud platform can be used to automatically process the vector data of the initial area and the remote sensing cultivated land mask data to determine the cultivated land area.

[0033] Step 102: Acquire laser sampling data corresponding to the cultivated land area.

[0034] Here, the laser sampling data can be the data of circular sampling points with a preset radius, or rectangular sampling points. Among them, the laser sampling data can be obtained by sampling the Global Ecosystem Dynamics Survey Satellite, which is a geodetic-grade light detection and ranging (lidar) system consisting of three lasers that produce eight parallel observation tracks. Each laser emits 242 pulses per second and illuminates a 25-meter-sized point (i.e., footprint) on the surface. Three-dimensional structure measurements are performed through these footprints. Each footprint is 60 meters apart in the direction of the track, and the lateral distance between each track is about 600 meters.

[0035] Step 103: Based on the historical phenological period of the crops in the cultivated land area and the quality attribute information of the laser sampling data, quality filtering is performed on the laser sampling data corresponding to the cultivated land area to obtain first laser sampling data; the historical phenological period is used to characterize the time period when the relative height value of the crops is stable.

[0036] Here, the historical phenological period of crops refers to the time corresponding to different stages of the crop's growth process, such as the sowing stage, flowering stage, jointing stage, relatively high stability stage, etc.

[0037] The first laser sampling data may be obtained only through time filtering, or may be obtained through time filtering and other filtering processes, and the present invention does not limit this.

[0038] It should be understood that the present invention requires that the remote sensing data be data in which the relative height value of the crop is in a stable stage.

[0039] Step 104: Filter target laser sampling data based on the maximum geographic positioning deviation of the first laser sampling data.

[0040] Here, geolocation deviation refers to the deviation of the position of the laser sampling point, which is generally up to 10 meters.

[0041] The target laser sampling data may be determined by any suitable method, for example, the first laser sampling data whose geographic positioning deviation is within a preset range is determined as the target laser sampling data; for another example, the geographic location deviations of the first laser sampling data are sorted from small to large, and the first n first laser sampling data are determined as the target laser sampling data, where n is an integer greater than 1.

[0042] Step 105: Determine a crop relative height bidirectional difference label corresponding to the target laser sampling data; the crop relative height bidirectional difference label is used to train a target estimation model for estimating the crop relative height.

[0043] It should be noted that the two-way difference of crop relative height refers to the difference between the positive height and negative height of the crop. Here, the positive height and negative height can be determined based on the percentage of the height reached by the laser radar reflection signal. Generally, the higher the energy return percentage, the higher the relative height of the crop; the lower the percentage, the lower the relative height of the crop. For example, when the energy return percentage of the reflected signal of the radar wave is 98, the determined relative height of the crop is named RH98.

[0044] Here, you can set a percentage threshold, where the percentage threshold can be any number between 1 and 100. The height below the percentage threshold is considered negative, and the height above the percentage threshold is considered positive. For example, if the percentage threshold is 50%, the height below 50% is considered negative, and the height above 50% is considered positive.

[0045] Step 106: Based on the crop relative height bidirectional difference label, a target estimation model for predicting the crop relative height is trained; and the target estimation model is used to infer the image of the area to be predicted to obtain the relative height of the area crops corresponding to the image of the area to be predicted.

[0046] In an embodiment of the present invention, first, the cultivated land area is determined by the vector data of the initial area and the remote sensing cultivated land mask data corresponding to the initial area; then, the laser sampling data of the cultivated land area is temporally and spatially filtered using the historical phenological period of the crops in the cultivated land area to obtain the first laser sampling data, thereby eliminating irrelevant data or inaccurate data and enhancing the operational efficiency of the model; and, the target laser sampling data is determined using the geographic positioning deviation of the first laser sampling data, thereby improving the accuracy of the data; finally, the high-accuracy laser sampling data is used to produce a two-way difference label for the relative height of the crop, thereby enhancing the characterization capability of the relative height of the crop canopy and the bottom layer, improving the accuracy and reliability of the label, and further improving the accuracy of the estimation of the relative height of the crop.

[0047] In an embodiment of the present invention, the step of screening target laser sampling data based on the maximum geographic positioning deviation of the first laser sampling data includes: Determining a scanning radius of the first laser sampling data; The scanning radius of the first laser sampling data refers to the radius of the sampling point pixel, which is generally 12.5-15 meters.

[0048] Determining a buffer radius for quality filtering of the first laser sampling data based on a scanning radius of the first laser sampling data and a maximum geographic positioning deviation of the first laser sampling data; It should be noted that the buffer radius may be the sum of the radius of the first laser sampling data and the maximum geographic positioning deviation, or may be obtained by weighted summation of the radius of the first laser sampling data and the geographic positioning deviation, which is not limited in the present invention.

[0049] The target laser sampling data is filtered based on the buffer radius.

[0050] The target laser sampling data may be determined by any suitable method, for example, the first laser sampling data having a buffer radius smaller than a preset buffer radius is determined as the target laser sampling data; for another example, the buffer radius of the first laser sampling data is sorted from small to large, and the first n first laser sampling data are determined as the target laser sampling data, where n is an integer greater than 1.

[0051] Specifically, the buffer radius The calculation of is as follows: (1); Among them, Rbuff is the buffer radius, Rgedi is the scanning radius of the radar laser sampling data, and Rgeo is the maximum geographic positioning deviation of the radar laser sampling data. Generally, the length of the radar laser sampling data is about 12.5-15 meters, and the maximum geographic positioning deviation is 10 meters.

[0052] In the embodiment of the present invention, the buffer radius is determined according to the radius of the first laser sampling data and the geographic positioning deviation, and the first laser sampling data corresponding to the buffer radius that meets the conditions is determined as the target laser sampling data. In this way, the sampling range is determined for the laser sampling data, and the data beyond the sampling range is eliminated, so as to avoid the sampling data in the non-cultivated land area, thereby improving the accuracy of the data.

[0053] In an embodiment of the present invention, training a target estimation model for predicting the relative height of crops based on the bidirectional difference labels of the relative height of crops includes: Screening the target remote sensing image corresponding to the cultivated land area; The target remote sensing image refers to the remote sensing image that only contains cultivated land areas.

[0054] Based on the target remote sensing image, the remote sensing spectral feature is calculated, and the remote sensing spectral feature is set as the sample feature corresponding to each of the crop relative height bidirectional difference labels; the sampling points of the target laser sampling data correspond to the pixels of the target remote sensing image one by one in time and space; Here, the spectral characteristics may include at least one of the following: crop greenness, canopy moisture content, vegetation nutrient content, vegetation growth and withering rate, soil background, surface texture and harmonic coefficient, etc.

[0055] Inputting the crop relative height bidirectional difference label and the sample feature into an initial estimation model to obtain an estimated bidirectional difference label; The initial estimation model can be any model in a machine learning model or a neural network model, such as a random forest model. Here, the initial estimation model can be a trained model obtained from a server, or can be trained using a relevant feature data set in a sample folder after the model is built.

[0056] Determining loss information based on the crop relative height bidirectional difference label and the estimated bidirectional difference label; Here, the loss information can be determined based on the loss function, and is used to characterize the difference between the crop relative height bidirectional difference label and the estimated bidirectional difference label. Generally, the greater the loss information, the greater the difference between the crop relative height bidirectional difference label and the estimated bidirectional difference label, and the worse the model effect; conversely, the smaller the difference between the crop relative height bidirectional difference label and the estimated bidirectional difference label, the better the model effect. The loss function can be a cross entropy loss function, a square loss function, etc.

[0057] Based on the loss information, the initial estimation model is continuously optimized to obtain the target estimation model.

[0058] It should be understood that the initial estimation model essentially updates the weight parameters of the computational model to be trained layer by layer through the back propagation algorithm based on the loss information, and uses an optimization algorithm (such as gradient descent method, Adam optimizer) to iteratively update the computational model to be trained until the loss information is minimized or the maximum number of iterations is reached.

[0059] In the embodiment of the present invention, the laser sampling data purification strategy is used to purify the laser sampling data points in time and space, and a bidirectional crop relative height difference label is developed as the prediction label of the random forest, and the spectral feature is established as the sample feature. The initial estimation model is trained by the bidirectional difference label and spectral feature of the crop relative height, and the parameters are continuously learned and updated from the data to improve the generalization ability and robustness of the model, thereby improving the accuracy of the crop relative height estimation. In addition, compared with the unidirectional height, the error of the unidirectional height cannot be adjusted, while the difference between the positive and negative directions can relatively offset the error, ensuring the accuracy and reliability of the data.

[0060] In an embodiment of the present invention, the quality filtering process is performed on the laser sampling data corresponding to the cultivated land area based on the historical phenological period of the crops in the cultivated land area and the quality attribute information of the laser sampling data to obtain the first laser sampling data, including: Based on the historical phenological period of crops in the cultivated land area, the laser sampling data is subjected to a first filtering process to obtain second laser sampling data; The first filtering process can be understood as time filtering, and the laser sampling data collected during a period of time when the crop is relatively highly stable is used as the second laser sampling data.

[0061] Based on the laser sampling data of the full-power beam in the second laser sampling data and the corresponding quality attribute information, performing a second filtering process on the second laser sampling data to obtain third laser sampling data; It is important to understand that radar lasers have two modes during the sampling process, namely full coverage beam and full power beam. Full power beam means that the radar antenna concentrates all or almost all of the transmission power or receiving sensitivity in a specific beam direction when transmitting or receiving. Full coverage beam describes the coverage range of the radar beam in space, which means that the radar beam can cover all or most of the space of the target area or area of ​​interest.

[0062] The method for determining the third laser sampling data may be any appropriate method, for example, retaining the laser sampling data of the full-power beam, or removing the laser sampling data of the full-power beam.

[0063] Based on the relative height of the crop corresponding to the third laser sampling data and the terrain information corresponding to the third laser sampling data, the third laser sampling data is filtered for the third time to obtain the first laser sampling data.

[0064] Here, the crop relative height and terrain information of the third laser sampling data can be calculated using a digital terrain model.

[0065] Specifically, laser sampling data whose crop relative height and terrain information both meet preset conditions can be determined as the first laser sampling data, or laser sampling data whose crop relative height or terrain information either meets the preset conditions can be determined as the first laser sampling data, and the present invention is not limited to this.

[0066] Exemplarily, the method for filtering the relative height of crops may include: determining the lowest high-level mode (elev_lowestmode) of the radar satellite and comparing it with the SRTM digital terrain high-level model (SRTM DEM) and the TanDEM digital terrain high-level model (TanDEM-X DEM) to obtain the difference in relative heights of the two crops, and filtering according to the difference in relative heights of the crops. Generally, if the difference in both heights exceeds 50 meters or any one of the heights exceeds 50 meters, it indicates a ground detection error, and therefore, the corresponding laser sampling data is excluded.

[0067] Exemplarily, the method for filtering terrain information may include: setting a preset height, and when the terrain information of the crop field is greater than the preset terrain information, excluding the corresponding laser sampling data. Generally, the preset terrain information is less than or equal to 6 degrees.

[0068] In the embodiment of the present invention, the laser sampling data is first filtered by the historical phenological period of the crop to obtain the second laser sampling data of the crop with relatively high stability; then, based on the full power beam, the third laser sampling data is determined from the second laser sampling data, and the third laser sampling data is filtered according to the height and slope to obtain the first laser sampling data. In this way, by progressively filtering the laser sampling data multiple times, pure data is obtained, which improves the accuracy and reliability of the data.

[0069] In an embodiment of the present invention, the second laser sampling data is filtered for the second time based on the laser sampling data of the full-power beam in the second laser sampling data and the corresponding quality attribute information to obtain the third laser sampling data, including: retaining the laser sampling data of the full-power beam in the second laser sampling data to obtain fourth laser sampling data; It should be understood that the full coverage beam of the Global Ecosystem Dynamics Survey Radar Satellite is almost ineffective in identifying and separating relative height differences of crops, while the full power beam performs well. Therefore, it is necessary to retain the laser sampling data of the full power beam in the second laser sampling data.

[0070] Based on the quality attribute information corresponding to the fourth laser sampling data, the fourth laser sampling data is filtered to obtain the third laser sampling data; the quality attribute information includes: detection mode, quality mark, waveform.

[0071] It should be noted that the detection mode (num_detectedmodes) is used to characterize the geographic accuracy of the laser sampling data. If num_detectedmodes is 0, it means that the geographic positioning accuracy of the laser sampling data is poor and the laser sampling data needs to be excluded.

[0072] The quality flag (quality_flag) is used to characterize the signal quality of the current sampling data point. If quality_flag is 0, it means that the signal quality of the current sampling data point is poor and the laser sampling data needs to be excluded.

[0073] Here, the waveform may be waveform integrity. Generally, waveform integrity is represented by the search end time (search_end) and the sample count (rx_sample_count). If the search end time (search_end) matches the sample count (rx_sample_count), it indicates that the waveform is complete. If the search end time (search_end) matches the sample count (rx_sample_count), it indicates that the waveform is incomplete. When the waveform is incomplete, the laser sampling data needs to be excluded. Among them, the search end time (search_end) matches the sample count (rx_sample_count) and can be understood as the number of microwave transmissions is the same as the number of receptions.

[0074] Here, a crop height threshold or range may be set to exclude laser sampling data whose crop height is greater than the crop height threshold or laser sampling data whose crop height is not within the range.

[0075] In an embodiment of the present invention, the laser sampling data of the full-power beam in the second laser sampling data is retained, and the retained laser sampling data is screened according to one or more of the detection mode, quality identification, waveform and corresponding crop height information of the retained laser sampling data, thereby improving the accuracy and reliability of the data.

[0076] In an embodiment of the present invention, screening the target remote sensing image corresponding to the cultivated land area includes: Acquire a remote sensing image corresponding to the initial area; The remote sensing image is preprocessed to obtain the target remote sensing image; the preprocessing includes at least one of the following: spatial mask processing, temporal filtering processing, resampling processing, spatial interpolation processing and mean smoothing processing.

[0077] Here, the target remote sensing image refers to a remote sensing image that only contains cultivated land areas and a period of time when crops are relatively highly stable.

[0078] The role of spatial mask processing is to determine the cultivated land area. Specifically, the remote sensing image is multiplied by the corresponding mask image to obtain a binary image of 0 and non-0, where the non-0 area is the cultivated land area, and the mask image is a binary image with the same size and shape as the remote sensing image.

[0079] It should be noted that clouds will also be removed during the process of spatial mask processing of remote sensing images. Therefore, it is necessary to perform spatial interpolation to supplement the remote sensing images after spatial mask processing to ensure the integrity of the image.

[0080] Temporal filtering processing uses the historical phenological period of crops as a time window to obtain remote sensing images of crops in a relatively highly stable period of time.

[0081] Resampling refers to regenerating or constructing another remote sensing image from a remote sensing image obtained after time filtering by changing the sampling frequency or range.

[0082] Spatial interpolation processing refers to converting discrete image data into continuous image data. Here, spatial interpolation processing is performed to supplement the image shortage caused by the previous spatial mask processing. The methods of spatial interpolation processing include but are not limited to inverse distance weighted average interpolation method, Kriging interpolation method, trend surface method, etc.

[0083] Mean smoothing is also called mean filtering, which is used to reduce noise in images. Here, a neighborhood of each pixel in the remote sensing image (usually a rectangular window, such as 3x3, 5x5, etc.) is used to calculate the average value of all pixel values ​​in the neighborhood and assign this average value to the current pixel.

[0084] In the embodiment of the present invention, the accuracy and reliability of the data are improved by performing a series of preprocessing such as spatial mask processing, temporal filtering processing, resampling processing, spatial interpolation processing and mean smoothing processing on the remote sensing image of the initial area.

[0085] In the embodiment of the present invention, the step of calculating the remote sensing spectral feature based on the target remote sensing image includes: Determining different reflectivity bands in the target remote sensing image; The band is actually a specific interval divided in order of wavelength. Here, the bands in the target remote sensing image can include visible light band, near infrared band, red edge band and shortwave infrared band, specifically blue, green, red, red edge 1, red edge 2, red edge 3, near infrared, narrow near infrared, shortwave infrared 1 and shortwave infrared 2 bands.

[0086] Reflectivity refers to the ability of an object's surface to reflect solar radiation or incident radiation within this band, that is, the ratio of the emitted radiation to the incident radiation.

[0087] In addition, the surface reflectivity of the target remote sensing image can be directly obtained through the remote sensing intelligent processing cloud platform.

[0088] Specifically, after preprocessing the remote sensing image, a target remote sensing image is obtained, and then time series extraction is performed on the bands of the target remote sensing image.

[0089] The remote sensing spectral feature is determined based on the texture feature in the target remote sensing image, the terrain feature in the target remote sensing image and different reflectance bands in the target remote sensing image.

[0090] Texture features include contrast, entropy and correlation calculated based on the gray-level co-occurrence matrix.

[0091] Specifically, the contrast The calculation of can refer to the following formula (2): (2); Among them, i and j are the row and column indices of gray-level co-occurrence, P is the probability of occurrence of matrix elements, and ∑ is the summation symbol.

[0092] entropy The calculation of can refer to the following formula (3): (3); Where i and j are the row and column indices of gray-level co-occurrence, P is the probability of occurrence of matrix elements, and log is the logarithmic function, which is the natural logarithm by default.

[0093] Relevance The calculation of can refer to the following formula (4): ) (4); Among them, i and j are the row and column indices of gray level co-occurrence, P is the probability of occurrence of matrix elements, and μ and σ are the mean and variance of the matrix gray levels, respectively.

[0094] Topographic features include the slope and height of remote sensing image pixels.

[0095] Spectral characteristics can include vegetation greenness, canopy moisture content, vegetation nutrient content, vegetation growth and withering rate, soil background, surface texture and harmonic coefficients, etc.

[0096] The canopy moisture content can be calculated based on the normalized difference wetness index. The calculation of can refer to the following formula (5): (5); in, represents the emissivity of the red band, Indicates the emissivity of the red edge 1 band.

[0097] Canopy water content can also include plant water content, which can be obtained based on the surface water index. The calculation of can refer to the following formula (6): (6); in, represents the reflectivity in the near-infrared band, Represents the reflectivity in the shortwave infrared band.

[0098] The content of various nutrients in vegetation can include chlorophyll content and anthocyanin content, wherein the chlorophyll content can be obtained according to the chlorophyll absorption reflectance index. The calculation of can refer to the following formula (7): (7); in, represents the emissivity of the green band,

[0099] In addition, the chlorophyll content can also be obtained based on the green chlorophyll vegetation index. The calculation of can refer to the following formula (8): (8); in, Represents the emissivity in the near-infrared band.

[0100] The anthocyanin content can be obtained based on the anthocyanin reflectance index. The calculation of can refer to the following formula (9): (9); Vegetation greenness can be obtained based on canopy greenness, which can be obtained based on the normalized difference vegetation index. The calculation of can refer to the following formula (10): (10); In addition, canopy greenness can also be obtained based on the enhanced vegetation index. The calculation of the enhanced vegetation index can refer to the following formula (11): (11); in, Represents the emissivity of the blue band.

[0101] Vegetation growth can be obtained based on growth rate, which can be obtained based on relative growth rate index. The calculation of can refer to the following formula (12): (12); The withering rate can be obtained based on the aging rate, which can be obtained based on the vegetation aging reflectance index. The calculation of can refer to the following formula (13): (13); The soil background can be obtained based on the vegetation coverage, and the vegetation coverage can be calculated based on the soil adjusted vegetation index. The calculation of the soil adjusted vegetation index can refer to the following formula (14): (14); Furthermore, the spectral characteristics can also include harmonic dynamics. The harmonic dynamics can be obtained based on the harmonic coefficients. The harmonic coefficients can be LSWI, GCVI, NDVI, MSAVI2 and PSRI. The harmonic dynamics can be calculated using the following formula (15): (15); in, It represents harmonic dynamics, and a, b, c, d, and e are harmonic coefficients.

[0102] In addition, since remote sensing images do not need to be collected every day, the spectral features in the remote sensing images can be synthesized using a 10-day median synthesis.

[0103] In the embodiment of the present invention, spectral features are calculated in combination with a plurality of crop indices to determine sample features, thereby further improving the diversity of sample features, thereby improving the breadth and robustness of the model and improving the accuracy of crop relative height estimation.

[0104] The following is an application scenario of a method for predicting relative height of regional crops provided by an embodiment of the present invention.

[0105] Figure 2 FIG2 is a flow chart of the method for predicting the relative height of regional crops provided by the present invention, and the specific steps include the following: Step S201: Determine the cultivated land area based on the vector data of the initial area and the remote sensing cultivated land mask data corresponding to the initial area.

[0106] Step S202: Based on the historical phenological period of crops in the cultivated land area, time filtering is performed on the laser sampling data corresponding to the cultivated land area to obtain second laser sampling data.

[0107] Step S203: based on the laser sampling data of the full-power beam in the second laser sampling data, the detection mode, the quality identifier, the waveform and the corresponding crop height information, the second laser sampling data is filtered to obtain third laser sampling data.

[0108] Step S204: Based on the height corresponding to the third laser sampling data and the slope corresponding to the third laser sampling data, perform a second filtering process on the third laser sampling data to obtain the first laser sampling data.

[0109] Step S205 : determining target laser sampling data based on the maximum geographic positioning deviation of the first laser sampling data.

[0110] Step S206: Determine the crop relative height bidirectional difference label corresponding to the target laser sampling data.

[0111] Step S207: pre-process the remote sensing image corresponding to the initial area to obtain a target remote sensing image.

[0112] Step S208: determining the spectral characteristics of the target remote sensing image based on the texture characteristics in the target remote sensing image, the terrain characteristics in the target remote sensing image, and the different reflectivity bands in the target remote sensing image.

[0113] Step S209: Based on the crop relative height bidirectional difference labels and spectral features, the initial estimation model is continuously optimized to obtain a target estimation model.

[0114] Step S210: Use the target estimation model to predict the area to be predicted to obtain the relative height of the crops in the area.

[0115] Here, the initial estimation model can be a random forest model. The main parameters of the random forest model (for example, the number of regression trees, the maximum depth of the regression tree, the maximum and minimum number of leaf nodes of the regression tree, etc.) are all optimized before training. Through the random resampling method, n bootstrap sample sets are randomly generated from the training sample set with replacement, and each bootstrap sample set is input into each regression tree in the random forest as all its training data. In the process of constructing the regression number, among all the sub-features of each split point, sub-features less than the total number of sub-features are randomly selected for growth, and each split point is scored by the weighted sum of the impurities of all sub-nodes after the split, and the optimal growth node is determined. For each regression tree, it grows from top to bottom, and in this process, the process parameters of the random forest regression model are continuously adjusted to determine the termination conditions of the model to optimize the model. The optimal parameter settings of the random forest model are: the training test ratio is 7:3, the number of regression trees is 500, and the maximum depth of the regression tree is 50. In addition, the input feature set of the model includes 10 spectral indices, 10 original bands, 3 texture features and 25 harmonic coefficients. The positive label and negative label are RH98 (positive value) and RH10 (negative value) of the global ecosystem dynamics survey radar laser sampling data. Therefore, the two-way difference label for crop relative height estimation can be calculated as follows (16): (16); in, is the bidirectional label of the crop relative height estimation model, RH98 is the positive label of the laser sampling data, and RH10 is the negative label of the laser sampling data.

[0116] In addition, after the target estimation model is trained, the target estimation model can be used to estimate the relative height of crops. During the application process, the acquired remote sensing image is directly input into the target estimation model, and the relative height of crops is inferred by the target estimation model. In this way, the spatial extrapolation of the relative height of crops from point to surface is realized accurately based on the random forest model, which reduces the geographic positioning error of laser sampling data, reduces the uncertainty of crop vertical structure measurement, and realizes high-precision estimation of relative height of crops under the premise of making full use of limited data. In addition, the target estimation model has robust migration capabilities on different crops and time and space scales, which can provide reliable reference for tasks such as farmland management, crop classification and yield prediction at a large regional scale.

[0117] Figure 3 is a schematic diagram of the prediction results of the target estimation model provided by the present invention, such as Figure 3The horizontal axis is the radar sampling height, the vertical axis is the target model estimated height, the solid line is the actual relationship between the true height and the predicted height, and the dotted line is the ideal relationship between the true height and the predicted height. In addition, the mean absolute error (MAE) of the target estimation model is 0.15 meters, the root mean square error (RMSE) is 0.22 meters, and the determination coefficient (R-squared, R 2 ) is 0.85, and the Mean Absolute Percentage Error (MAPE) is 3.42%.

[0118] Figure 4 This is a schematic diagram of the relative height estimation of crops in a certain area provided by the present invention, such as Figure 4 As shown, the areas at different depths have different heights, wherein the relative height of the crops ranges from 3 meters to 5 meters.

[0119] The regional crop relative height prediction device provided by the present invention is described below. The regional crop relative height prediction device described below and the regional crop relative height prediction method described above can be referenced to each other.

[0120] Figure 5 is a schematic diagram of the structure of the regional crop relative height prediction device provided by the present invention, referring to Figure 5 As shown, the regional crop relative height prediction device 500 includes: a first determination module 501, an acquisition module 502, a filtering module 503, a second determination module 504, a third determination module 505 and a prediction module 506, wherein: The first determination module 501 is used to determine the cultivated land area based on the vector data of the initial area and the remote sensing cultivated land mask data corresponding to the initial area; the initial area includes the cultivated land area and the non-cultivated land area; An acquisition module 502 is used to acquire laser sampling data corresponding to the cultivated land area; The filtering module 503 is used to perform quality filtering on the laser sampling data corresponding to the cultivated land area based on the historical phenological period of the crops in the cultivated land area and the quality attribute information of the laser sampling data to obtain the first laser sampling data; the historical phenological period is used to characterize the time period when the relative height value of the crops is stable; A second determination module 504 is used to screen target laser sampling data based on the maximum geographic positioning deviation of the first laser sampling data; The third determination module 505 is used to determine the crop relative height bidirectional difference label corresponding to the target laser sampling data; The prediction module 506 is used to train a target estimation model for predicting the relative height of crops based on the two-way difference labels of the relative height of crops; and use the target estimation model to infer the image of the area to be predicted to obtain the relative height of crops in the area corresponding to the image of the area to be predicted.

[0121] In an exemplary embodiment, the second determination module 504 is specifically used to: determine the scanning radius of the first laser sampling data; determine the buffer radius of the quality filtering of the first laser sampling data based on the scanning radius of the first laser sampling data and the maximum geographic positioning deviation of the first laser sampling data; and screen the target laser sampling data based on the buffer radius.

[0122] In an exemplary embodiment, the prediction module 506 is specifically used to: screen the target remote sensing image corresponding to the cultivated land area; calculate the remote sensing spectral characteristics based on the target remote sensing image, and set the remote sensing spectral characteristics as the sample characteristics corresponding to each of the crop relative height two-way difference labels; the sampling points of the target laser sampling data correspond one-to-one with the pixels of the target remote sensing image in time and space; input the crop relative height two-way difference labels and the sample characteristics into the initial estimation model to obtain the estimated two-way difference labels; determine the loss information based on the crop relative height two-way difference labels and the estimated two-way difference labels; based on the loss information, continuously optimize the initial estimation model to obtain the target estimation model.

[0123] In an exemplary embodiment, the filtering module 503 is specifically used to: perform a first filtering process on the laser sampling data based on the historical phenological period of the crops in the cultivated land area to obtain second laser sampling data; perform a second filtering process on the second laser sampling data based on the laser sampling data of the full-power beam in the second laser sampling data and the corresponding quality attribute information to obtain third laser sampling data; perform a third filtering process on the third laser sampling data based on the relative height of the crops corresponding to the third laser sampling data and the terrain information corresponding to the third laser sampling data to obtain the first laser sampling data.

[0124] In an exemplary embodiment, the filtering module 503 is further specifically used to: retain the laser sampling data of the full-power beam in the second laser sampling data to obtain the fourth laser sampling data; based on the quality attribute information corresponding to the fourth laser sampling data, filter the fourth laser sampling data to obtain the third laser sampling data; the attribute information includes at least one of the following: detection mode, quality identification, waveform.

[0125] In an exemplary embodiment, the prediction module 506 is further specifically used to: obtain the remote sensing image corresponding to the initial area; preprocess the remote sensing image to obtain the target remote sensing image; the preprocessing includes at least one of the following: spatial mask processing, temporal filtering processing, resampling processing, spatial interpolation processing and mean smoothing processing.

[0126] In an exemplary embodiment, the prediction module 506 is further specifically used to: determine different reflectance bands in the target remote sensing image; determine the remote sensing spectral features based on texture features in the target remote sensing image, terrain features in the target remote sensing image and different reflectance bands in the target remote sensing image.

[0127] Figure 6 is a schematic diagram of the physical structure of the electronic device provided by the present invention, such as Figure 6 As shown, the electronic device may include: a processor (processor) 610 , a communication interface (Communications Interface) 620 , a memory (memory) 630 and a communication bus 640 , wherein the processor 610 , the communication interface 620 , and the memory 630 communicate with each other through the communication bus 640 . The processor 610 can call the logic instructions in the memory 630 to execute the regional crop relative height prediction method, which includes: determining the cultivated land area based on the vector data of the initial area and the remote sensing cultivated land mask data corresponding to the initial area; the initial area includes the cultivated land area and the non-cultivated land area; obtaining the laser sampling data corresponding to the cultivated land area; based on the historical phenological period of the crops in the cultivated land area and the quality attribute information of the laser sampling data, performing quality filtering processing on the laser sampling data corresponding to the cultivated land area to obtain the first laser sampling data; the historical phenological period is used to characterize the time period in which the relative height value of the crop is stable; based on the maximum geographic positioning deviation of the first laser sampling data, screening the target laser sampling data; determining the crop relative height bidirectional difference label corresponding to the target laser sampling data; based on the crop relative height bidirectional difference label, training a target estimation model for predicting the crop relative height; and using the target estimation model to infer the image of the area to be predicted to obtain the regional crop relative height corresponding to the image of the area to be predicted.

[0128] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0129] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the regional crop relative height prediction method provided by the above-mentioned methods, the method including: determining a cultivated land area based on vector data of an initial area and remote sensing cultivated land mask data corresponding to the initial area; the initial area includes a cultivated land area and a non-cultivated land area; obtaining laser sampling data corresponding to the cultivated land area; based on the historical phenological period of crops in the cultivated land area and the quality attribute information of the laser sampling data, performing quality filtering processing on the laser sampling data corresponding to the cultivated land area to obtain first laser sampling data; the historical phenological period is used to characterize the time period in which the relative height value of the crop is in a stable state; based on the maximum geographic positioning deviation of the first laser sampling data, screening target laser sampling data; determining a crop relative height bidirectional difference label corresponding to the target laser sampling data; based on the crop relative height bidirectional difference label, training a target estimation model for predicting the crop relative height; and using the target estimation model to infer the image of the area to be predicted to obtain the regional crop relative height corresponding to the image of the area to be predicted.

[0130] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented by a processor to execute the regional crop relative height prediction method provided by the above-mentioned methods when the computer program is executed, the method comprising: determining a cultivated land area based on vector data of an initial area and remote sensing cultivated land mask data corresponding to the initial area; the initial area includes a cultivated land area and a non-cultivated land area; obtaining laser sampling data corresponding to the cultivated land area; performing quality filtering processing on the laser sampling data corresponding to the cultivated land area based on the historical phenological period of crops in the cultivated land area and the quality attribute information of the laser sampling data to obtain first laser sampling data; the historical phenological period is used to characterize a time period in which the relative height value of the crop is in a stable state; based on the maximum geographic positioning deviation of the first laser sampling data, screening target laser sampling data; determining a crop relative height bidirectional difference label corresponding to the target laser sampling data; training a target estimation model for predicting the crop relative height based on the crop relative height bidirectional difference label; and using the target estimation model to infer the image of the area to be predicted to obtain the regional crop relative height corresponding to the image of the area to be predicted.

[0131] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0132] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting relative height of regional crops, characterized in that: include: Determine the cultivated land area based on the vector data of the initial area and the remote sensing cultivated land mask data corresponding to the initial area; The initial area includes cultivated land area and non-cultivated land area; Acquiring laser sampling data corresponding to the cultivated land area; Based on the historical phenological period of the crops in the cultivated land area and the quality attribute information of the laser sampling data, quality filtering processing is performed on the laser sampling data corresponding to the cultivated land area to obtain first laser sampling data; the historical phenological period is used to characterize the time period when the relative height value of the crops is stable; Based on the maximum geographic positioning deviation of the first laser sampling data, screening the target laser sampling data; Determine a crop relative height bidirectional difference label corresponding to the target laser sampling data; Based on the crop relative height bidirectional difference label, a target estimation model for predicting the crop relative height is trained; and the target estimation model is used to infer the image of the area to be predicted to obtain the relative height of the area crops corresponding to the image of the area to be predicted.

2. The method for predicting relative height of regional crops according to claim 1, characterized in that: The step of screening the target laser sampling data based on the maximum geographic positioning deviation of the first laser sampling data comprises: Determining a scanning radius of the first laser sampling data; Determining a buffer radius for quality filtering of the first laser sampling data based on a scanning radius of the first laser sampling data and a maximum geographic positioning deviation of the first laser sampling data; The target laser sampling data is filtered based on the buffer radius.

3. The method for predicting relative height of regional crops according to claim 1, characterized in that: The method of training a target estimation model for predicting the relative height of crops based on the bidirectional difference labels of the relative height of crops comprises: Screening the target remote sensing image corresponding to the cultivated land area; Based on the target remote sensing image, the remote sensing spectral feature is calculated, and the remote sensing spectral feature is set as the sample feature corresponding to each of the crop relative height bidirectional difference labels; the sampling points of the target laser sampling data correspond to the pixels of the target remote sensing image one by one in time and space; Inputting the crop relative height bidirectional difference label and the sample feature into an initial estimation model to obtain an estimated bidirectional difference label; Determining loss information based on the crop relative height bidirectional difference label and the estimated bidirectional difference label; Based on the loss information, the initial estimation model is continuously optimized to obtain the target estimation model.

4. The method for predicting relative height of regional crops according to any one of claims 1 to 3, characterized in that: The method of performing quality filtering on the laser sampling data corresponding to the cultivated land area based on the historical phenological period of the crops in the cultivated land area and the quality attribute information of the laser sampling data to obtain the first laser sampling data includes: Based on the historical phenological period of crops in the cultivated land area, the laser sampling data is subjected to a first filtering process to obtain second laser sampling data; Based on the laser sampling data of the full-power beam in the second laser sampling data and the corresponding quality attribute information, performing a second filtering process on the second laser sampling data to obtain third laser sampling data; Based on the relative height of the crop corresponding to the third laser sampling data and the terrain information corresponding to the third laser sampling data, the third laser sampling data is filtered for the third time to obtain the first laser sampling data.

5. The method for predicting relative height of regional crops according to claim 4, characterized in that: The second laser sampling data is filtered for the second time based on the laser sampling data of the full-power beam in the second laser sampling data and the corresponding quality attribute information to obtain the third laser sampling data, including: retaining the laser sampling data of the full-power beam in the second laser sampling data to obtain fourth laser sampling data; Based on the quality attribute information corresponding to the fourth laser sampling data, the fourth laser sampling data is filtered to obtain the third laser sampling data; the quality attribute information includes at least one of the following: detection mode, quality identification, waveform.

6. The method for predicting relative height of regional crops according to claim 3, characterized in that: The screening of the target remote sensing image corresponding to the cultivated land area includes: Acquire a remote sensing image corresponding to the initial area; The remote sensing image is preprocessed to obtain the target remote sensing image; the preprocessing includes at least one of the following: spatial mask processing, temporal filtering processing, resampling processing, spatial interpolation processing and mean smoothing processing.

7. The method for predicting relative height of regional crops according to claim 3, characterized in that: The step of calculating the remote sensing spectral characteristics based on the target remote sensing image includes: Determining different reflectivity bands in the target remote sensing image; The remote sensing spectral feature is determined based on the texture feature in the target remote sensing image, the terrain feature in the target remote sensing image and different reflectance bands in the target remote sensing image.

8. A device for predicting relative height of regional crops, characterized in that: include: A first determination module is used to determine the cultivated land area based on the vector data of the initial area and the remote sensing cultivated land mask data corresponding to the initial area; The initial area includes cultivated land area and non-cultivated land area; An acquisition module, used for acquiring laser sampling data corresponding to the cultivated land area; A filtering module, configured to perform quality filtering on the laser sampling data corresponding to the cultivated land area based on the historical phenological period of the crops in the cultivated land area and the quality attribute information of the laser sampling data, so as to obtain first laser sampling data; the historical phenological period is used to characterize a time period in which the relative height value of the crops is stable; A second determination module, configured to screen target laser sampling data based on a maximum geographic positioning deviation of the first laser sampling data; A third determination module is used to determine a bidirectional difference label of a relative height of a crop corresponding to the target laser sampling data; The prediction module is used to train a target estimation model for predicting the relative height of crops based on the two-way difference labels of the relative height of crops; and use the target estimation model to infer the image of the area to be predicted to obtain the relative height of the area crops corresponding to the image of the area to be predicted.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for predicting the relative height of regional crops as described in any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting the relative height of regional crops as claimed in any one of claims 1 to 7 is implemented.

Images