A method and system for making fertilization decisions for crops

Abstract

The present invention relates to the field of agriculture, and particularly to a method for making fertilization decisions for crops. It acquires the spectral images of the crops taken, determines the weights according to the central wavelengths of the bands. If the pixel value in the spectral image of a band is equal to the preset value, the pixel value is adjusted to 1, otherwise it is adjusted to 0. The weights and the adjusted spectral image are multiplied bit by bit to obtain the adjusted spectral image of the band, and the spectral images of all bands are accumulated to obtain the feature map corresponding to the preset value; based on the feature map corresponding to the preset value and the central wavelength of the band, the spectral image of each band is corrected; the vegetation index is obtained by using the corrected spectral image, and the fertilization amount map of the crops is obtained according to the vegetation index; during the fertilization process, the fertilization amount is adjusted in real time according to the fertilization amount map. By correcting the spectrum, the present invention reduces the influence of atmospheric scattering, etc. on the spectrum and improves the accuracy of calculating the fertilization amount.

Description

Technical Field

[0001] The present invention relates to the field of agriculture, and particularly to a method and system for making fertilization decisions for crops. Background Art

[0002] Rational fertilization can increase crop yields, improve quality, and at the same time reduce environmental pollution and waste of resources. However, traditional fertilization methods mostly rely on farmers' experience or general fertilization amount standards, and often it is difficult to accurately meet the specific needs of different plots, climate conditions or crop growth stages. Hyperspectral imaging technology, with its high resolution, can capture the demand of crop leaves for nutrients such as nitrogen, phosphorus, and potassium, and is a key tool for evaluating the nutritional status of crops. Through spectral images taken by drones or satellites, the health status of crops can be monitored in real time and dynamically, and then reasonable fertilization decisions can be made according to specific nutritional requirements.

[0003] However, hyperspectral images face many problems in practical applications, and among them, atmospheric influence is an important source of error. Factors such as aerosols, clouds, and water vapor in the atmosphere will cause absorption and scattering of the spectral signals reflected from the ground surface, resulting in a difference between the spectra received by the sensor and the actual reflectance of the crops. Uncorrected hyperspectral images will affect the accuracy of fertilization decisions, may lead to misjudgment of the nutritional status of crops, and thus cause problems of over-fertilization or under-fertilization. Summary of the Invention

[0004] In a first aspect, the present invention provides a method for making fertilization decisions for crops, and the method includes the following steps:

[0005] Obtain the spectral map of the crops taken, determine the weight according to the central wavelength of the band. If the pixel value in the spectral map of the band is equal to the preset value, adjust the pixel value to 1, otherwise adjust it to 0, multiply the weight and the adjusted spectral map bit by bit to obtain the adjusted spectral map of the band, and accumulate the spectral maps of all bands to obtain the feature map corresponding to the preset value;

[0006] Based on the feature map corresponding to the preset value and the central wavelength of the band, correct the spectral map of each band;

[0007] Use the corrected spectral map to obtain the vegetation index, and obtain the fertilization amount map of the crops according to the vegetation index; during the fertilization process, adjust the fertilization amount in real time according to the fertilization amount map.

[0008] Combined with the first aspect, in some implementation manners, the correcting the spectral map of each band based on the feature map corresponding to the preset value and the central wavelength of the band is specifically:

[0009] Determine the division interval of the band based on the central wavelength of the band, and determine the corresponding selection number of the division interval. Sort the preset values in ascending order, and start from the first one after sorting to determine the preset number of preset values. Add the feature maps corresponding to the preset values to obtain the feature map of the band;

[0010] Obtain the adjustment factor of the original spectral map of the band using the feature map of the band, and subtract the adjustment factor from each pixel value in the original spectral map of the band to obtain the corrected spectral map; if the pixel value minus the adjustment factor is negative, set the pixel value to 0.

[0011] Combined with the first aspect, in some implementation manners, the obtaining the adjustment factor of the original spectral map of the band using the feature map of the band is specifically:

[0012] Determine the position of the maximum value in the feature map of the band, and use the pixel value at this position in the original spectral map of the band as the adjustment factor; or, determine the positions in the feature map of the band that are not 0, and use the average value of the pixel values at these positions in the original spectral map of the band as the adjustment factor; or, perform element summation after multiplying the feature map of the band and the original spectral map of the band bit by bit, perform element summation on the feature map of the band, and use the ratio of the previous summation result to the latter summation result as the adjustment factor.

[0013] Combined with the first aspect, in some implementation manners, the obtaining the optimal fertilization amount map of the crop according to the vegetation index is specifically:

[0014] Obtain the correlations between different vegetation indices and different growth stages and required nutrient elements of the crop, select at least one vegetation index with the largest correlation to construct a fertilization amount model, and input the vegetation index and the crop growth stage into the fertilization amount model to obtain the fertilization amount map.

[0015] Combined with the first aspect, in some implementation manners, the selecting at least one vegetation index with the largest correlation to construct a fertilization amount model is specifically:

[0016] Construct a fitting function, where the input of the fitting function is the at least one vegetation index and the crop growth stage, and the output of the fitting function is the fertilization amount of the nutrient component, and use the fitting function as the fertilization amount model.

[0017] Combined with the first aspect, in some implementation manners, the selecting at least one vegetation index with the largest correlation to construct a fertilization amount model is specifically:

[0018] Construct training samples, where the features of the training samples at least include at least one index with the greatest relevance and the crop growth stage. The training samples also include the fertilization amount corresponding to the nutrient components. Use the training samples to train an ensemble learning model, and use the trained ensemble learning model as the fertilization amount model.

[0019] In a second aspect, the present invention provides a crop fertilization decision-making system, which includes the following modules:

[0020] A spectral acquisition module, configured to obtain a spectral map of the photographed crop, determine weights according to the central wavelength of the band. If the pixel value in the spectral map of the band is equal to the preset value, adjust the pixel value to 1, otherwise adjust it to 0, multiply the weights and the adjusted spectral map bit by bit to obtain the adjusted spectral map of the band, and accumulate the spectral maps of all bands to obtain the feature map corresponding to the preset value;

[0021] A spectral correction module, configured to correct the spectral map of each band based on the feature map corresponding to the preset value and the central wavelength of the band;

[0022] A fertilization control module, configured to obtain a vegetation index using the corrected spectral map, obtain a fertilization amount map of the crop according to the vegetation index; and adjust the fertilization amount in real time according to the fertilization amount map during the fertilization process.

[0023] In combination with the second aspect, in some implementation manners, the correcting the spectral map of each band based on the feature map corresponding to the preset value and the central wavelength of the band is specifically:

[0024] Determine the division interval of the band based on the central wavelength of the band, and determine the selection number corresponding to the division interval. Sort the preset values in ascending order, and start from the first one after sorting to determine the preset number of preset values. Add the feature maps corresponding to the preset values to obtain the feature map of the band;

[0025] Obtain the adjustment factor of the original spectral map of the band using the feature map of the band, subtract the adjustment factor from each pixel value in the original spectral map of the band to obtain the corrected spectral map; if the pixel value minus the adjustment factor is negative, set the pixel value to 0.

[0026] In combination with the second aspect, in some implementation manners, the obtaining the adjustment factor of the original spectral map of the band using the feature map of the band is specifically:

[0027] Determine the position of the maximum value in the feature map of the band, and use the pixel value at this position in the original spectral map of the band as the adjustment factor; or, determine the positions in the feature map of the band that are not 0, and use the average value of the pixel values at these positions in the original spectral map of the band as the adjustment factor; or, perform element summation after multiplying the feature map of the band and the original spectral map of the band bit by bit, perform element summation on the feature map of the band, and use the ratio of the previous summation result to the latter summation result as the adjustment factor.

[0028] Combined with the second aspect, in some implementation manners, the obtaining the optimal fertilization amount map of the crop according to the vegetation index specifically includes:

[0029] Obtain the correlations between different vegetation indices and different growth stages and required nutrient elements of the crop, select at least one vegetation index with the largest correlation to construct a fertilization amount model, and input the vegetation index and the crop growth stage into the fertilization amount model to obtain a fertilization amount map.

[0030] Combined with the second aspect, in some implementation manners, the selecting at least one vegetation index with the largest correlation to construct a fertilization amount model specifically includes:

[0031] Construct a fitting function, where the input of the fitting function is the at least one vegetation index and the crop growth stage, and the output of the fitting function is the fertilization amount of the nutrient component, and use the fitting function as the fertilization amount model.

[0032] Combined with the second aspect, in some implementation manners, the selecting at least one vegetation index with the largest correlation to construct a fertilization amount model specifically includes:

[0033] Construct a training sample, where the features of the training sample at least include the at least one index with the largest correlation and the crop growth stage, and the training sample further includes the fertilization amount corresponding to the nutrient component. Use the training sample to train an ensemble learning model, and use the trained ensemble learning model as the fertilization amount model.

[0034] In a third aspect, the present application provides a computer-readable storage medium, characterized in that instructions are stored in the computer-readable storage medium, and when the instructions run on an electronic device, the method described in the first aspect is executed.

[0035] In a fourth aspect, the present application provides a computer program product, characterized in that the computer program product includes computer instructions, and when executed by an electronic device, the electronic device executes the method described in the first aspect.

[0036] Regarding the problem that uncorrected spectral data may be affected by the atmosphere, resulting in inaccurate calculation of vegetation indices and further affecting the scientific nature of fertilization decisions, according to the different influences of atmospheric scattering on different bands, the present invention determines the positions of low pixel values in the spectral map from multiple bands, and then further corrects the spectral map of each band in combination with the central wavelength of the band, avoiding inaccurate selection of dark pixels and using different adjustment factors for different bands in a targeted manner, achieving more accurate correction and ensuring the accuracy of fertilization amount calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a flowchart of Embodiment 1;

[0038] Figure 2 It is a schematic diagram of adjusting the spectral map according to a preset value;

[0039] Figure 3 It is a schematic diagram of weighting the adjusted spectral map;

[0040] Figure 4 It is a schematic diagram of obtaining a feature map corresponding to a preset value. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the accompanying drawings required for description in the embodiments will be briefly introduced below.

[0042] The terms "including" and "having" and any variations thereof mentioned in the description of the present application are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes other steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products or devices.

[0043] It should be noted that in the embodiments of the present application, words such as "exemplary" or "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the embodiments of the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of words such as "exemplary" or "for example" is intended to present relevant concepts in a specific manner.

[0044] In the description of the present application, unless otherwise specified, "a plurality of" means two or more. The "and / or" herein is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone.

[0045] Applying fertilizers at the right time, in the right amount, and in the right place helps improve the growth rate and health of crops, thereby directly enhancing the yield and quality of crops. Traditional fertilization methods may rely on experience and often cannot precisely control the amount and location of fertilizer application. Applying too little fertilizer can lead to a decrease in crop yield, while applying too much fertilizer may result in fertilizer waste and may also cause environmental problems such as groundwater pollution and eutrophication of rivers and lakes.

[0046] Different substances have different absorption and reflection characteristics of light. Therefore, by analyzing the spectrum, physical and chemical characteristic information of substances can be obtained. When light irradiates the surface of plants, part of the light is reflected back to the sensor, and crops at different growth stages and with different development statuses reflect light differently.

[0047] In the first embodiment of the present invention, a method for making fertilization decisions for crops is provided, which includes the following steps:

[0048] Step 1: Obtain the spectral image of the crops taken. Determine the weight according to the central wavelength of the band. If the pixel value in the spectral image of the band is equal to the preset value, adjust the pixel value to 1; otherwise, adjust it to 0. Multiply the weight and the adjusted spectral image bit by bit to obtain the adjusted spectral image of the band, and accumulate the spectral images of all bands to obtain the feature map corresponding to the preset value;

[0049] The spectral image of the crops is taken by a multispectral sensor or a hyperspectral sensor carried by a drone. In another embodiment, a remote sensing satellite is used to obtain the spectral image of the crops. Preferably, the spectral image is a hyperspectral image.

[0050] There are multiple channels in the spectral image, and each channel corresponds to a band. For example, the near-infrared band corresponds to a channel in the spectral image. In a remote sensing image, the optical signal received by the sensor is first converted into an analog electrical signal. Due to the influence of atmospheric scattering, etc., the data collected in the spectral image is not entirely the reflection of the plants. Moreover, different wavelengths have different sensitivities to atmospheric scattering. The long waveband is less affected by atmospheric scattering, which means that most of the optical signal received by the sensor is the signal reflected by the crops, rather than atmospheric scattering; while the short waveband is much more affected by atmospheric scattering than the long waveband.

[0051] In an alternative embodiment, the weight is determined according to the central wavelength of the band. Specifically, the shorter the central wavelength of the band, the greater the weight. In this way, if the pixel value of a short waveband is 0 or close to 0, it indicates that the influence of atmospheric scattering at this pixel is indeed very low, and the credibility is higher than that of the long waveband. In a more specific embodiment, the weight is determined in the form of a negative exponential function. The input of the negative exponential function is the central wavelength of the band, and the output is the weight.

[0052] In the atmospheric correction of the spectrum using dark pixels, the selection of dark pixels is directly related to the correction effect. In many cases, the darkest pixel in the spectrum is directly selected. However, the spectral characteristics of different bands vary greatly. Some pixels may appear as the darkest pixels in one band, but not in other bands. Moreover, if there is only one dark pixel point in a spectrum, this point may be caused by noise, etc. Directly selecting the darkest pixel in a certain band as the dark pixel for all bands may lead to incorrect correction. In the present invention, multiple preset values are set in advance. In one embodiment, the preset values are all values less than a certain value. Specifically, if a certain value is 5, the preset values are 0, 1, 2, 3, 4. In another embodiment, the multiple preset values are a series of consecutive integers starting from 0. In order to perform atmospheric correction on the spectral map, the number of preset values is within a certain number, such as 5 or 6, etc.

[0053] There are multiple preset values, and for each preset value, a feature map will be obtained. Specifically, for the spectral map of each band, each pixel value (pixel point) of the spectral map is compared with a specific preset value. If the pixel value is equal to the preset value, the pixel value is replaced with 1; if it is not equal to the preset value, it is replaced with 0, as Figure 2 shown. For the spectral image of each band, after adjusting the pixel values to 1 or 0 first, then multiply them bit by bit with the corresponding weights of the band, and then adjust the contribution values of each pixel according to the weights, so as to generate a spectral map after adjustment of one band, as Figure 3 shown. Accumulate the adjusted spectral maps of each band, add them pixel by pixel, fuse the information of multiple bands together, and finally generate a feature map, which is the feature map corresponding to the preset value, as Figure 4 shown.

[0054] For example, when the preset value is 0, the pixel values in the spectral map of band 1 that are equal to the preset value, that is, 0, are adjusted to 1, and the pixel values in the spectral map of band 1 that are not equal to the preset value, that is, 0, are adjusted to 0. Then multiply according to the weight of band 1 and the adjusted spectral map of band 1. The same operations are performed for other bands, such as band 2 and band 3. Finally, the feature map obtained by adding the spectral maps after the above operations of all bands bit by bit is the feature map of the preset value 0. The larger the value on the feature map, the greater the possibility that this pixel is 0 in all bands, that is, the greater the possibility of being a dark pixel. Similarly, the feature maps of preset values 1, 2, etc. can be obtained.

[0055] Step 2, correct the spectral map of each band based on the feature map corresponding to the preset value and the central wavelength of the band;

[0056] Due to the influence of sensor noise, etc., some pixels are very low, but they are not really the places with low reflectivity. Moreover, different bands are affected by atmospheric scattering differently due to different wavelengths. For example, the short-wave band is more severely affected by atmospheric scattering, and the pixel values of its corresponding spectral map are generally larger than the true values. Different correction methods should be used for different bands.

[0057] In an optional embodiment, correcting the spectral map of each band based on the feature map corresponding to the preset value and the central wavelength of the band specifically includes:

[0058] Determine the division interval of the band based on the central wavelength of the band, and determine the corresponding selection number of the division interval. Sort the preset values in ascending order, and start from the first one after sorting to determine the preset number of preset values. Add the feature maps corresponding to the preset values to obtain the feature map of the band;

[0059] Obtain the adjustment factor of the original spectral map of the band by using the feature map of the band, and subtract the adjustment factor from each pixel value in the original spectral map of the band to obtain the corrected spectral map; if the pixel value minus the adjustment factor is negative, set the pixel value to 0.

[0060] Since the short-wave band is more affected by atmospheric scattering, its value is higher than the true value. For example, if the true value is 1, the pixel value of the short-wave spectral map may be 2 or 3. This requires observing the situation of larger pixels. In one embodiment, the shorter the central wavelength of the band, the larger the corresponding selection number of the division interval. For example, the central wavelength of band 1 is divided into the first division interval, and its corresponding selection number is 2. If the central wavelength of band 2 is shorter than that of band 1, it is divided into the second division interval, and its corresponding selection number is 3. However, the maximum value of the selection number does not exceed the number of preset values.

[0061] Arrange all the given preset values in ascending order. Assume the preset values are 0, 1, 2, 3. After sorting, they are 0, 1, 2, 3. Further assume that the selection number of band 1 is 2, then the feature maps with preset values 0 and 1 will be selected, and these two feature maps are added bit by bit to obtain the feature map of band 1. The feature maps of band 2, band 3, etc. can be determined in the same way.

[0062] Each band has a feature map. The value of a pixel on the feature map represents the degree to which the pixel values at this position in all bands are relatively low. For example, if the value at position (2, 3) on the feature map of band 1 is 3 and the value at position (3, 5) is 0, it means that the pixel values in the spectral maps of other bands at position (2, 3) are also very low. Then, this position as a dark pixel has a better effect than position (3, 5). In one embodiment, the method for obtaining the adjustment factor of the original spectral map of a band using the feature map of the band is specifically as follows:

[0063] Determine the position of the maximum value in the feature map of the band, and use the pixel value at this position in the original spectral map of the band as the adjustment factor; or, determine the positions in the feature map of the band where the values are not 0, and use the average value of the pixel values at these positions in the original spectral map of the band as the adjustment factor; or, multiply the feature map of the band and the original spectral map of the band bit by bit (i.e., multiply the pixel values at the same position), then perform an element-wise summation on the products, perform an element-wise summation on the feature map of the band, and use the ratio of the previous summation result to the latter summation result as the adjustment factor.

[0064] When determining the adjustment factor, there are several methods. An optional embodiment is to find the position with the largest pixel value in the feature map of the band, and then, view the pixel value at the same position in the original spectral map of the band, and use this value as the adjustment factor. For example, if the value at position (2, 3) on the feature map of band 1 is the largest, then use the value at position (2, 3) in the original spectral map of band 1 as the adjustment factor.

[0065] Another optional embodiment is to find all the positions in the feature map of the band where the pixel values are not 0, then find the pixel values at these positions in the original spectral map of the band, and calculate the average value of these values as the adjustment factor. For example, if the values at positions (2, 3) and (9, 12) on the feature map of band 1 are not 0, then use the average value of the values at positions (2, 3) and (9, 12) in the original spectral map of band 1 as the adjustment factor.

[0066] In yet another optional embodiment, multiply the feature map of the band and the original spectral map of the band bit by bit (i.e., multiply the pixel values at the same position), then perform an element-wise summation on the products. Next, perform an element-wise summation on the feature map itself. Finally, calculate the ratio of these two summation results and use the ratio as the adjustment factor. For example, if the values at positions (2, 3) and (9, 12) on the feature map of band 1 are not 0 and are 1 and 5 respectively, and the values at positions (2, 3) and (9, 12) in the original spectral map of band 1 are 7 and 1 respectively, multiply the feature map of the band and the original spectral map of the band bit by bit, and then the element-wise summation result of the products is 12. The element-wise summation result of the feature map itself is 6, and the ratio of the two is 2.

[0067] By using the above method, the adjustment coefficient of each band can be calculated. In this way, each band will have an adjustment coefficient. Subtract the adjustment factor from each pixel value in the original spectral map of the band to obtain the corrected spectral map; if the pixel value minus the adjustment factor is negative, set the pixel value to 0.

[0068] Step 3: Obtain the vegetation index from the corrected spectral map, and obtain the fertilization amount map of the crops according to the vegetation index; adjust the fertilization amount in real time according to the fertilization amount map during the fertilization process.

[0069] The corrected spectral map can reduce the influence of atmospheric scattering, etc. on the spectrum, which can more accurately obtain the vegetation index of the crops, and the calculation of the corresponding fertilization amount is more accurate. After obtaining the vegetation index, use the vegetation index to obtain the optimal fertilization amount. Since each pixel of the spectral map represents a certain range of the ground in the remote sensing image, the fertilization amount within the range of the crops represented by the pixel can be obtained, and the fertilization amount can be adjusted or controlled in real time according to the optimal fertilization amount map during the fertilization process. For example, in the foliar fertilization using a drone, determine the pixel where the drone is located based on the position of the drone, and further control the fertilization amount according to the value of the pixel in the optimal fertilization amount map. The vegetation index includes but is not limited to NDVI (Normalized Difference Vegetation Index), EVI (Enhanced Vegetation Index), SAVI (Soil Adjusted Vegetation Index), etc.

[0070] For different crops and different growth stages, the correlations with the same vegetation index are different. In an optional embodiment, the obtaining the optimal fertilization amount map of the crops according to the vegetation index specifically includes:

[0071] Obtain the correlations between different vegetation indices and different growth stages and required nutrient elements of the crops, select at least one vegetation index with the largest correlation to construct a fertilization amount model, and input the vegetation index and the crop growth stage into the fertilization amount model to obtain the fertilization amount map.

[0072] By analyzing the relationship between each vegetation index and the nutrient elements required by the crops, such as nitrogen, phosphorus, and potassium, find out which vegetation indices can best reflect the nutrient status of the crops. Preferably, use linear regression or an ensemble learning model to construct a relationship model between the vegetation index and the nutrient requirements of the crops. Input the selected vegetation index and the current growth stage of the crop into the fertilization amount model to calculate the fertilization requirements of each pixel point.

[0073] In an optional embodiment, the selecting at least one vegetation index with the largest correlation to construct a fertilization amount model specifically includes:

[0074] Construct a fitting function. The input of the fitting function is the at least one vegetation index and the crop growth stage, and the output of the fitting function is the fertilization amount of the nutrient components. The fitting function is used as the fertilization amount model.

[0075] By mathematical or statistical methods, construct a fitting function that represents the relationship between the input (vegetation index and crop growth stage) and the output (fertilization amount of nutrient components). For example, input at least one vegetation index with the highest correlation with crop nutrient requirements, such as NDVI or NDRE, and the crop growth stages, such as the germination stage, vegetative growth stage, filling stage, etc., into the fitting function, and the output is the fertilization amount of the nutrient components, such as the specific fertilization amounts of nitrogen, phosphorus, and potassium.

[0076] In another optional embodiment, the at least one vegetation index with the highest correlation is selected to construct the fertilization amount model, specifically:

[0077] Construct training samples. The features of the training samples at least include the at least one index with the highest correlation and the crop growth stage. The training samples also include the fertilization amount corresponding to the nutrient components. Use the training samples to train the ensemble learning model, and use the trained ensemble learning model as the fertilization amount model.

[0078] In a second aspect, the present invention provides a crop fertilization decision-making system, which includes the following modules:

[0079] A spectral acquisition module, which is used to obtain the spectral map of the photographed crops, determine the weights according to the central wavelengths of the bands. If the pixel value in the spectral map of the band is equal to the preset value, adjust the pixel value to 1, otherwise adjust it to 0, multiply the weights and the adjusted spectral map bit by bit to obtain the adjusted spectral map of the band, and accumulate the spectral maps of all bands to obtain the feature map corresponding to the preset value;

[0080] A spectral correction module, which is used to correct the spectral map of each band based on the feature map corresponding to the preset value and the central wavelength of the band;

[0081] A fertilization control module, which is used to obtain the vegetation index by using the corrected spectral map, obtain the fertilization amount map of the crops according to the vegetation index; and adjust the fertilization amount in real time according to the fertilization amount map during the fertilization process.

[0082] Combined with the second aspect, in some implementation manners, the correction of the spectral map of each band based on the feature map corresponding to the preset value and the central wavelength of the band is specifically:

[0083] Determine the division interval of the band based on the central wavelength of the band, and determine the corresponding selection number of the division interval. Sort the preset values in ascending order, determine the preset number of preset values starting from the first one after sorting, and add the feature maps corresponding to the preset values to obtain the feature map of the band;

[0084] Obtain the adjustment factor of the original spectral map of the band using the feature map of the band, subtract the adjustment factor from each pixel value in the original spectral map of the band to obtain the corrected spectral map; if the pixel value minus the adjustment factor is negative, set the pixel value to 0.

[0085] Combined with the second aspect, in some implementation manners, the obtaining the adjustment factor of the original spectral map of the band using the feature map of the band is specifically:

[0086] Determine the position of the maximum value in the feature map of the band, and use the pixel value at the position in the original spectral map of the band as the adjustment factor; or, determine the positions in the feature map of the band that are not 0, and use the average value of the pixel values at the positions in the original spectral map of the band as the adjustment factor; or, perform element summation after multiplying the feature map of the band and the original spectral map of the band bit by bit, perform element summation on the feature map of the band, and use the ratio of the previous summation result and the latter summation result as the adjustment factor.

[0087] Combined with the second aspect, in some implementation manners, the obtaining the optimal fertilization amount map of the crop according to the vegetation index is specifically:

[0088] Obtain the correlations between different vegetation indices and different growth stages and required nutrient elements of the crop, select at least one vegetation index with the largest correlation to construct a fertilization amount model, and input the vegetation index and the crop growth stage into the fertilization amount model to obtain the fertilization amount map.

[0089] Combined with the second aspect, in some implementation manners, the selecting at least one vegetation index with the largest correlation to construct a fertilization amount model is specifically:

[0090] Construct a fitting function, the input of the fitting function is the at least one vegetation index and the crop growth stage, and the output of the fitting function is the fertilization amount of the nutrient component. Use the fitting function as the fertilization amount model.

[0091] Combined with the second aspect, in some implementation manners, the selecting at least one vegetation index with the largest correlation to construct a fertilization amount model is specifically:

[0092] Construct training samples, the features of the training samples at least include at least one index with the greatest correlation and the crop growth stage, the training samples also include the fertilization amount corresponding to the nutrient components, and use the training samples to train the ensemble learning model, and use the trained ensemble learning model as the fertilization amount model.

[0093] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, the processes or functions described in the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wire (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that contains one or more sets of available media. The available media can be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as high-density digital video discs (DVDs)), or semiconductor media. The semiconductor media can be SSDs.

[0094] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for making fertilization decisions for crops, characterized in that, The method includes the following steps: Obtain the spectral map of the photographed crops, determine the weights according to the central wavelengths of the bands. If the pixel value in the spectral map of a band is equal to the preset value, adjust the pixel value to 1; otherwise, adjust it to 0. Multiply the weights and the adjusted spectral map bit by bit to obtain the adjusted spectral map of the band, and accumulate the spectral maps of all bands to obtain the feature map corresponding to the preset value; Correct the spectral map of each band based on the feature map corresponding to the preset value and the central wavelength of the band; Obtain the vegetation index using the corrected spectral map, and obtain the fertilization amount map of the crops according to the vegetation index; during the fertilization process, adjust the fertilization amount in real time according to the fertilization amount map; The correcting the spectral map of each band based on the feature map corresponding to the preset value and the central wavelength of the band is specifically: Determine the division interval of the band based on the central wavelength of the band, and determine the corresponding selection number of the division interval. Sort the preset values in ascending order, and determine the preset number of preset values starting from the first one after sorting. Add the feature maps corresponding to the preset values to obtain the feature map of the band; Obtain the adjustment factor of the original spectral map of the band using the feature map of the band, subtract the adjustment factor from each pixel value in the original spectral map of the band to obtain the corrected spectral map; if the pixel value minus the adjustment factor is negative, set the pixel value to 0.

2. The method according to claim 1, wherein The obtaining the adjustment factor of the original spectral map of the band using the feature map of the band is specifically: Determine the position of the maximum value in the feature map of the band, and use the pixel value at this position in the original spectral map of the band as the adjustment factor; or, determine the positions in the feature map of the band that are not 0, and use the average value of the pixel values at these positions in the original spectral map of the band as the adjustment factor; Or, perform element summation after multiplying the feature map of the band and the original spectral map of the band bit by bit, perform element summation on the feature map of the band, and use the ratio of the previous summation result to the latter summation result as the adjustment factor.

3. The method according to claim 1, characterized in that, The obtaining the fertilization amount map of the crops according to the vegetation index is specifically: Obtain the correlations between different vegetation indices and different growth stages and required nutrient elements of the crops, select at least one vegetation index with the largest correlation to construct a fertilization amount model, and input the vegetation index and the crop growth stage into the fertilization amount model to obtain the fertilization amount map.

4. The method according to claim 3, wherein The selecting at least one vegetation index with the largest correlation to construct a fertilization amount model is specifically: Construct a fitting function, the input of the fitting function is the at least one vegetation index and the crop growth stage, and the output of the fitting function is the fertilization amount of the nutrient component. Use the fitting function as the fertilization amount model.

5. The method according to claim 3, wherein The selecting at least one vegetation index with the largest correlation to construct a fertilization amount model is specifically: Construct a training sample, the features of the training sample at least include the at least one index with the largest correlation and the crop growth stage, and the training sample also includes the fertilization amount corresponding to the nutrient component. Use the training sample to train an ensemble learning model, and use the trained ensemble learning model as the fertilization amount model.

6. A crop fertilization decision-making system, characterized in that, The system includes the following modules: A spectral acquisition module, which is used to obtain the spectral map of crops taken, determine weights according to the central wavelength of the band. If the pixel value in the spectral map of the band is equal to the preset value, the pixel value is adjusted to 1, otherwise it is adjusted to 0. Multiply the weight and the adjusted spectral map bit by bit to obtain the adjusted spectral map of the band, and accumulate the spectral maps of all bands to obtain the feature map corresponding to the preset value; A spectral correction module, which is used to correct the spectral map of each band based on the feature map corresponding to the preset value and the central wavelength of the band; A fertilization control module, which is used to obtain the vegetation index by using the corrected spectral map, and obtain the fertilization amount map of the crops according to the vegetation index; during the fertilization process, adjust the fertilization amount in real time according to the fertilization amount map; The correction of the spectral map of each band based on the feature map corresponding to the preset value and the central wavelength of the band is specifically as follows: Determine the division interval of the band based on the central wavelength of the band, and determine the selection number corresponding to the division interval. Sort the preset values in ascending order, and determine the preset number of preset values starting from the first one after sorting. Add the feature maps corresponding to the preset values to obtain the feature map of the band; Obtain the adjustment factor of the original spectral map of the band by using the feature map of the band, and subtract the adjustment factor from each pixel value in the original spectral map of the band to obtain the corrected spectral map; if the pixel value minus the adjustment factor is negative, set the pixel value to 0.

7. The system according to claim 6, wherein The obtaining of the adjustment factor of the original spectral map of the band by using the feature map of the band is specifically as follows: Determine the position of the maximum value in the feature map of the band, and use the pixel value at this position in the original spectral map of the band as the adjustment factor; or, determine the position where the value in the feature map of the band is not 0, and use the average value of the pixel values at this position in the original spectral map of the band as the adjustment factor; Or, sum the elements after multiplying the feature map of the band and the original spectral map of the band bit by bit, sum the elements of the feature map of the band, and use the ratio of the previous summation result and the latter summation result as the adjustment factor.

8. A computer storage device, on which a computer program is stored, characterized in that, The computer program, when executed by a processor, implements the method according to any one of claims 1-5.

Images