An unmanned aerial vehicle-based multispectral image intelligent processing method

Abstract

The present application relates to the technical field of image processing, and more particularly to a multispectral image intelligent processing method based on a UAV, comprising: acquiring real-time data; determining a low-illumination grid; determining a temporary grid; determining an extraction grid; determining an end-member set; adjusting a threshold value and outputting. The present application can effectively deal with complex changes in farmland environments, such as uneven illumination and crop shading, by comprehensively analyzing multi-dimensional data such as real-time illumination intensity, crop coverage, ground height and multispectral images in farmland areas, and accurately screens out grids suitable for extracting end members. By dynamically adjusting the light intensity threshold and the synchronization threshold, the end member extraction result can be optimized according to the actual situation, significantly improving the accuracy and robustness of end member extraction, reducing noise interference, and effectively solving the problems of spectral information loss and unstable processing caused by gray-scale processing and threshold setting relying on manual adjustment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and in particular to a multi-spectral image intelligent processing method based on unmanned aerial vehicles. BACKGROUND

[0002] With the continuous progress of unmanned aerial vehicle technology and remote sensing technology, multi-spectral images based on unmanned aerial vehicles are increasingly widely used in the field of agriculture, especially in precision agriculture monitoring. However, due to the complexity and variability of farmland environments, traditional image processing methods face challenges in endmember extraction, making it difficult to fully utilize the spatial and spectral information in multi-spectral images.

[0003] The patent document with publication number CN105631847A discloses a multi-spectral image processing method and device. The method includes: calculating a gray image of the multi-spectral image and counting a histogram of the gray image; performing mean filter processing on the histogram; constructing a difference sign array and a support array filter, wherein the number of elements of the difference sign array is equal to the dimension of the histogram, the initial value of the elements of the difference sign array is 0, if the difference value obtained by performing difference calculation on the histogram after mean filter processing is positive, the corresponding element in the difference sign array is 1, and if the difference value is negative, the corresponding element in the difference sign array is -1; the number of elements of the support array filter is 2m, and the first m elements are 1 and the last m elements are -1, where m is a positive integer greater than or equal to 5; filtering the difference sign array with the support array filter as a mask, determining whether the obtained filter value is greater than a set threshold, and if so, determining that the filter value is a peak value of the histogram; and taking the number of calculated peak values as the number of clusters, and clustering each pixel point of the multi-spectral image based on a k-means clustering algorithm.

[0004] As can be seen, the multi-spectral image processing method has the following problems: the method first converts the multi-spectral image into a gray image and processes it through histogram statistics, which can cause loss of some spectral information, especially when the spectral channel information in the multi-spectral image has high independence and diversity, the features of each band cannot be fully retained after conversion to a gray image; the method relies on a set threshold to determine whether the filter value is a peak value of the histogram, and the setting of this threshold needs to be adjusted according to the specific image, lacking an automatic optimization mechanism, resulting in unstable performance under different environments or image types. SUMMARY

[0005] Therefore, the present application provides a multi-spectral image intelligent processing method based on unmanned aerial vehicles to overcome the problems of spectral information loss and unstable processing caused by gray-scale processing and threshold setting relying on manual adjustment in the prior art through multi-scale feature fusion and dynamic threshold adjustment.

[0006] To achieve the above object, the application provides a multispectral image intelligent processing method based on a UAV, comprising:

[0007] Obtaining real-time illumination intensity, real-time crop coverage, real-time average ground height and real-time multispectral image of each to-be-processed grid in the UAV collection area of the farmland based on grid division;

[0008] Determining a plurality of low-illumination grids according to the real-time illumination intensity and a preset illumination threshold value;

[0009] Determining a plurality of temporary grids according to the real-time crop coverage and the real-time multispectral image of each low-illumination grid;

[0010] Determining a plurality of extraction grids according to the real-time multispectral image, the real-time average ground height and a preset synchronization threshold value in each temporary grid;

[0011] Determining an endmember set according to the real-time multispectral image of all extraction grids;

[0012] Adjusting the preset illumination threshold value according to the endmember set and a preset standard set to form an adjusted illumination threshold value;

[0013] Adjusting the preset synchronization threshold value according to the endmember set formed based on the adjusted illumination threshold value within a preset adjustment time length to form an adjusted synchronization threshold value;

[0014] Outputting the endmember set formed based on the adjusted synchronization threshold value.

[0015] Further, determining a plurality of temporary grids according to the real-time crop coverage and the real-time multispectral image of each low-illumination grid comprises:

[0016] Extracting real-time spectral reflectance of the real-time multispectral image using a preset endmember extraction model;

[0017] Calculating the average value of the real-time spectral reflectance to form an average spectral reflectance;

[0018] Calculating the standard deviation of the real-time crop coverage within a preset first determination time length to form a coverage fluctuation value;

[0019] Calculating the standard deviation of the average spectral reflectance within the preset first determination time length to form a first reflectance fluctuation value;

[0020] Determining a plurality of temporary grids according to the coverage fluctuation value and the first reflectance fluctuation value.

[0021] Further, determining a plurality of temporary grids according to the coverage fluctuation value and the first reflectance fluctuation value comprises:

[0022] draw a coverage change curve according to the coverage fluctuation value;

[0023] draw a first reflectivity change curve according to the first reflectivity fluctuation value;

[0024] calculate a cosine similarity of the coverage change curve and the first reflectivity change curve, to form a change consistency degree;

[0025] when the change consistency degree is greater than a preset consistency threshold, determine that the low-illumination grid is a temporary grid.

[0026] Further, determining a plurality of extraction grids according to real-time multispectral images, the real-time average ground height in each temporary grid, and a preset synchronization threshold comprises:

[0027] calculate a standard deviation of the average spectral reflectivity in a preset second determination time period, to form a second reflectivity fluctuation value;

[0028] calculate a standard deviation of the real-time average ground height in the preset second determination time period, to form a height fluctuation value;

[0029] determine a plurality of abnormal grids according to the second reflectivity fluctuation value and the height fluctuation value;

[0030] exclude the abnormal grids from all the to-be-processed grids, to form a plurality of extraction grids.

[0031] Further, determining a plurality of abnormal grids according to the second reflectivity fluctuation value and the height fluctuation value comprises:

[0032] draw a second reflectivity change curve according to the second reflectivity fluctuation value;

[0033] draw a height change curve according to the height fluctuation value;

[0034] calculate a cosine similarity of the second reflectivity change curve and the height change curve, to form a change synchronization degree;

[0035] when the change synchronization degree is less than the preset synchronization threshold, determine that the temporary grid is an abnormal grid, to form a plurality of abnormal grids.

[0036] Further, determining an endmember set according to the real-time multispectral images of all the extraction grids comprises:

[0037] preprocess the real-time multispectral images, to form processed multispectral images;

[0038] extract real-time spectral reflectivity of all the processed multispectral images using the preset endmember extraction model, to form an endmember set.

[0039] Furthermore, adjusting the preset light intensity threshold according to the endmember set and the preset standard set to form the adjusted light intensity threshold includes:

[0040] Calculating the cosine similarity between the endmember set and the preset standard set to form a set deviation;

[0041] When the collective deviation is greater than the preset deviation threshold, the preset light intensity threshold is adjusted according to the relative deviation between the collective deviation and the preset deviation threshold and a preset first adjustment coefficient to form an adjusted light intensity threshold.

[0042] Further, adjusting the preset synchronization threshold according to the endmember set formed based on the adjusted light intensity threshold within the preset adjustment time period, to form the adjusted synchronization threshold includes:

[0043] Calculating the standard deviation of the number of sets of real-time endmember sets within the preset adjustment time length to form a set number fluctuation value;

[0044] When the set quantity fluctuation value is greater than the preset quantity fluctuation threshold, the preset synchronization threshold is adjusted according to the relative deviation between the set quantity fluctuation value and the preset quantity fluctuation threshold and a preset second adjustment coefficient to form an adjusted synchronization threshold.

[0045] Furthermore, determining a number of low illumination grids according to the real-time illumination intensity and a preset light intensity threshold comprises:

[0046] When the real-time light intensity is less than the preset light intensity threshold, the grid to be processed is determined to be a low-illumination grid.

[0047] Furthermore, preprocessing the real-time multispectral image to form a processed multispectral image includes:

[0048] A preset convex body optimization model is used to remove noise from the processed multispectral image to form a processed multispectral image.

[0049] Compared with the prior art, the beneficial effects of the present application are that, by comprehensively analyzing the real-time light intensity, crop coverage, ground height and multi-spectral image and other multi-dimensional data of the farmland area, the complex farmland environment changes such as uneven light and crop shielding and other factors can be effectively responded to, and the grid suitable for end member extraction can be accurately screened out. By dynamically adjusting the light intensity threshold and the synchronization threshold, the end member extraction result can be optimized according to the actual situation, the precision and robustness of end member extraction are significantly improved, and noise interference is reduced. The method not only improves the accuracy and reliability of multi-spectral image processing, but also provides precise environmental monitoring and decision support for agricultural production, and has important application value. Through real-time feedback adjustment and optimization, it can better adapt to different farmland scenes, improve the intelligent management level of crop growth condition evaluation, pest and disease monitoring, fertilization and irrigation decision-making, and effectively solve the problems of spectral information loss and unstable processing caused by gray processing and threshold setting relying on manual adjustment.

[0050] Further, by analyzing the fluctuation of crop coverage and average spectral reflectance, regions related to changes in the environment or crop growth status can be effectively screened out, helping to accurately identify temporary grids in farmland, improving the accuracy of data processing, making subsequent abnormal grid identification and end member set extraction more accurate, reducing misjudgment caused by environmental changes or image noise, and thus optimizing farmland monitoring and management effects.

[0051] Further, by analyzing the consistency of the fluctuation of coverage and reflectance, grid regions reflecting consistent crop growth trends can be effectively identified, avoiding misjudgment caused by environmental noise and local interference, thereby improving the accuracy of grid screening.

[0052] Further, by combining the fluctuation values of average spectral reflectance and ground height, abnormal grids can be effectively identified, thereby filtering out unreliable grid data, ensuring that reliable and stable data is selected in the subsequent processing process, improving the accuracy of data analysis and overall processing effect.

[0053] Further, by analyzing the change trend and synchronicity of reflectance and height, abnormal grids can be effectively identified, avoiding false data interference caused by environmental changes or equipment errors, thereby improving the accuracy and reliability of data processing and ensuring the effectiveness of subsequent analysis and decision-making.

[0054] Further, by preprocessing the multi-spectral image and accurately extracting the spectral reflectance, noise and irrelevant factors can be effectively eliminated, improving the quality and reliability of image data. By forming an end member set, the characteristics of different regions can be more accurately represented, providing strong support for subsequent crop monitoring, land management and precision agriculture.

[0055] Further, by calculating the cosine similarity of the end member set and the standard set and dynamically adjusting the light intensity threshold based on set deviation, the light intensity threshold can more accurately reflect the actual image characteristics, thereby improving the system's processing accuracy for low illumination areas. The light intensity threshold can be optimized according to different image characteristics, making it more adaptable in different environments and improving overall acquisition and processing quality.

[0056] Further, by adjusting the synchronization threshold, the response capability to light intensity threshold changes can be optimized, improving recognition accuracy. Using the set number fluctuation value to determine whether to adjust can effectively avoid unnecessary adjustments while ensuring efficient performance in cases of large fluctuations, enhancing the ability to adapt to complex environmental changes.

[0057] Further, by introducing a preset light intensity threshold, areas with insufficient light can be effectively screened out, helping to accurately identify low-light areas that require special attention, optimizing subsequent processing and analysis, and improving the ability to respond to environmental condition changes in crop monitoring, agricultural management, or related fields.

[0058] Further, by applying a preset convex optimization model, image quality can be effectively improved, reducing the impact of noise on subsequent analysis, thereby improving the accuracy and reliability of image processing and providing more accurate data support for subsequent feature extraction, analysis, and decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 Flowchart of the unmanned aerial vehicle-based multispectral image intelligent processing method of the present embodiment;

[0060] Figure 2 Determination logic diagram for determining temporary grids in the present embodiment;

[0061] Figure 3 Determination logic diagram for determining abnormal grids in the present embodiment;

[0062] Figure 4 Determination logic diagram for determining low-illumination grids in the present embodiment. DETAILED DESCRIPTION

[0063] To make the objectives and advantages of the present application clearer, the following further describes the present application with reference to embodiments. It should be understood that the specific embodiments described herein are merely intended to explain the present application and not to limit the present application.

[0064] The preferred embodiments of the present application are described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely intended to explain the technical principles of the present application and not to limit the protection scope of the present application.

[0065] Please refer to Figure 1As shown, it is a flowchart of the unmanned aerial vehicle based multi-spectral image intelligent processing method of the present embodiment;

[0066] The present embodiment provides an unmanned aerial vehicle based multi-spectral image intelligent processing method, comprising:

[0067] Obtaining the real-time light intensity, real-time crop coverage, real-time average ground height and real-time multi-spectral image of each to-be-processed grid in the unmanned aerial vehicle collection area of the farmland based on grid division;

[0068] Determining a plurality of low-illumination grids according to the real-time light intensity and a preset light intensity threshold;

[0069] Determining a plurality of temporary grids according to the real-time crop coverage and the real-time multi-spectral image of each low-illumination grid;

[0070] Determining a plurality of extraction grids according to the real-time multi-spectral image, the real-time average ground height and a preset synchronization threshold in each temporary grid;

[0071] Determining an endmember set according to the real-time multi-spectral image of all the extraction grids;

[0072] Adjusting the preset light intensity threshold according to the endmember set and a preset standard set to form an adjusted light intensity threshold;

[0073] Adjusting the preset synchronization threshold according to the endmember set formed based on the adjusted light intensity threshold within a preset adjustment time to form an adjusted synchronization threshold;

[0074] Outputting the endmember set formed based on the adjusted synchronization threshold.

[0075] Obtaining the real-time light intensity, real-time crop coverage, real-time average ground height and real-time multi-spectral image of each to-be-processed grid in the unmanned aerial vehicle collection area of the farmland based on grid division mainly relies on a variety of sensors and imaging equipment carried by the unmanned aerial vehicle. Specifically, the unmanned aerial vehicle collects the light intensity data of the farmland area in real time through the integrated light sensor, obtains the image data of different wavebands through the high-resolution multi-spectral camera, and uses the image data to calculate the crop coverage and spectral reflectance. At the same time, the ground height information of the farmland is obtained by using laser radar or other ground height measurement technology. According to the spatial distribution of the farmland area, the collected data is divided into grids, so that each grid unit is analyzed one by one, and the real-time light, crop coverage, ground height and multi-spectral image data of each grid are obtained, providing a basis for subsequent processing and analysis.

[0076] The preset light intensity threshold is a standard value for distinguishing between low-light areas and high-light areas, and is usually set between 100 lux and 500 lux, and is set to 300 lux in this embodiment, which can ensure that low-light areas that may not be suitable for accurate extraction due to light conditions can be effectively identified, and preparation for further analysis.

[0077] The preset synchronization threshold is a parameter for determining whether the spectral reflectance fluctuation and the ground height fluctuation change synchronously, and is usually set between 0.2 and 0.5, and is set to 0.3 in this embodiment, which can effectively identify and exclude grids whose spectral reflectance and ground height change asynchronously, thereby eliminating errors caused by environmental abnormalities or data noise and improving the accuracy and reliability of data analysis.

[0078] By dividing the farmland area based on the grid, real-time light intensity, crop coverage, ground height, and multispectral image data are obtained, and the grids of the farmland are intelligently processed. First, low-light grids are selected according to the real-time light intensity and the preset light intensity threshold. Then, temporary grids are further selected based on the crop coverage and the multispectral image. Next, the multispectral image, the ground height, and the synchronization threshold are used to determine the extraction grids. The spectral reflectance is extracted through a preset endmember extraction model, and an endmember set is formed. According to the deviation of the endmember set from the preset standard set, the light intensity threshold is dynamically adjusted, and the synchronization threshold is optimized within a preset adjustment duration, and finally the optimized endmember set is output.

[0079] By comprehensively analyzing the real-time light intensity, crop coverage, ground height, and multispectral image data of the farmland area, complex farmland environmental changes such as uneven light and crop shading can be effectively dealt with, and grids suitable for extracting endmembers can be accurately selected. By dynamically adjusting the light intensity threshold and the synchronization threshold, the endmember extraction result can be optimized according to the actual situation, significantly improving the accuracy and robustness of endmember extraction and reducing noise interference. This method not only improves the accuracy and reliability of multispectral image processing, but also provides precise environmental monitoring and decision support for agricultural production, and has important application value. Through real-time feedback adjustment and optimization, it can better adapt to different farmland scenarios, improve the intelligent management level of crop growth condition evaluation, disease and pest monitoring, fertilization and irrigation decision-making, and effectively solve the problems of spectral information loss and unstable processing caused by gray-scale processing and threshold setting relying on manual adjustment.

[0080] Specifically, determining a plurality of temporary grids according to the real-time crop coverage and the real-time multispectral image of each low-light grid includes:

[0081] extracting real-time spectral reflectance of the real-time multispectral image using a preset endmember extraction model;

[0082] calculating a mean value of the real-time spectral reflectance to form a mean spectral reflectance;

[0083] calculating a standard deviation of the real-time crop coverage in a preset first determination duration to form a coverage fluctuation value;

[0084] calculating a standard deviation of the mean spectral reflectance in the preset first determination duration to form a first reflectance fluctuation value;

[0085] determining a plurality of temporary grids according to the coverage fluctuation value and the first reflectance fluctuation value;

[0086] The preset endmember extraction model is an algorithm model designed for multispectral image data, usually based on machine learning or deep learning technology, used to extract representative spectral features from images. In the extraction process, the model first preprocesses the input real-time multispectral image to remove noise and irrelevant background information, and then extracts specific spectral reflectance values from the spectral data of each pixel. These reflectance values reflect information such as crop growth status, ground coverage, etc. The endmember extraction model usually learns a large amount of training data to optimize its parameters to identify and extract features that are important for crop identification or land cover classification. These spectral reflectance values are used for subsequent analysis to help determine key indicators such as crop coverage and growth status.

[0087] The preset first determination duration is the length of the time window used to calculate crop coverage and spectral reflectance fluctuation values, which depends on the growth cycle of crops, environmental factors, and the sampling frequency of sensors, and is usually set between 1 hour and 1 day, in this embodiment, it is set to 6 hours, which can balance short-term and long-term changes, avoid misjudgment due to short-term fluctuations, and capture relatively stable trend changes, improving the reliability and effectiveness of data processing.

[0088] According to the real-time crop coverage and real-time multispectral image of the low-illumination grid, first, use the preset endmember extraction model to extract the real-time spectral reflectance of each grid. Then, calculate the standard deviation of the real-time crop coverage in the preset first determination duration to obtain the coverage fluctuation value; at the same time, calculate the standard deviation of the real-time spectral reflectance in the same duration to form the first reflectance fluctuation value. According to the changes of the two fluctuation values, a plurality of temporary grids are determined, and by comparing the fluctuation of crop coverage and spectral reflectance, the area with potential characteristic changes is identified.

[0089] The fluctuations of crop coverage and average spectral reflectance directly reflect the changes in the growth status of crops and environmental conditions within the farmland area. The fluctuations of crop coverage can reveal the changes in the density and spatial distribution of crop growth, while the fluctuations of spectral reflectance can reflect the differences in plant health and soil types. Through the analysis of the fluctuations of these two parameters, abnormal changes or uneven areas in the farmland can be effectively identified, which helps to more accurately assess potential problems such as crop growth, pest occurrence, etc.

[0090] By analyzing the fluctuations of crop coverage and average spectral reflectance, areas related to changes in environmental or crop growth status can be effectively screened out, helping to accurately identify temporary grids in the farmland, improving the accuracy of data processing, making subsequent abnormal grid identification and endmember set extraction more accurate, reducing false positives caused by environmental changes or image noise, and thus optimizing the effect of farmland monitoring and management.

[0091] Please continue to refer to Figure 2 as shown in the determination logic diagram for determining temporary grids in this embodiment;

[0092] Specifically, determining a number of temporary grids according to the coverage fluctuation value and the first reflectance fluctuation value includes:

[0093] Drawing a coverage change curve according to the coverage fluctuation value;

[0094] Drawing a first reflectance change curve according to the first reflectance fluctuation value;

[0095] Calculating the cosine similarity of the coverage change curve and the first reflectance change curve to form a change consistency degree;

[0096] When the change consistency degree is greater than a preset consistency threshold, determining the low-illumination grid as a temporary grid.

[0097] The preset consistency threshold is used to measure the similarity between the coverage change curve and the reflectance change curve, and is usually set according to the characteristics of agricultural data and application requirements. Generally, it is set between 0 and 1, and in this embodiment, it is set to 0.8. This can effectively filter out some interference and noise while ensuring the accuracy of identification, ensure that the extracted temporary grids reflect the actual changes in crop growth, reduce errors, and improve the accuracy of subsequent processing. The purpose of setting this value is to ensure the accuracy of grid screening and exclude false positives caused by small-scale fluctuations, thereby improving the stability and reliability of the analysis.

[0098] According to the coverage fluctuation value and the first reflectivity fluctuation value, first, the coverage change curve and the reflectivity change curve are drawn, and then the cosine similarity of the two curves is calculated to form the change consistency. When the change consistency is greater than a preset consistency threshold, the low irradiation grid is determined as a temporary grid, and through the comprehensive characteristics of the crop coverage and the spectral reflectivity, the grid with stable and similar growth state is accurately identified.

[0099] By analyzing the change consistency of the coverage and reflectivity fluctuation, the grid area reflecting the consistent growth trend of crops can be effectively identified, avoiding misjudgment caused by environmental noise and local interference, thereby improving the accuracy of grid screening.

[0100] Specifically, according to the real-time multispectral image in each of the temporary grids, the real-time average ground height, and a preset synchronization threshold, a plurality of extraction grids are determined, including:

[0101] The standard deviation of the average spectral reflectivity in a preset second determination duration is calculated to form a second reflectivity fluctuation value;

[0102] The standard deviation of the real-time average ground height in the preset second determination duration is calculated to form a height fluctuation value;

[0103] According to the second reflectivity fluctuation value and the height fluctuation value, a plurality of abnormal grids are determined;

[0104] The abnormal grids in all the to-be-processed grids are excluded to form a plurality of extraction grids.

[0105] The preset second determination duration refers to the time interval for calculating the standard deviation of real-time spectral reflectivity and ground height, which is determined according to the growth cycle of crops, the frequency of environmental fluctuations, and the accuracy requirement of analysis. It is usually between a few minutes and a few tens of minutes, and is set to 10 minutes in this embodiment. It can effectively capture the fluctuation change in a short period of time while balancing the stability and response speed of data, thereby improving the identification accuracy of abnormal grids and ensuring the accuracy of the processing result.

[0106] First, the standard deviation of the average spectral reflectivity in a preset second determination duration is calculated to obtain a second reflectivity fluctuation value, and the standard deviation of the real-time average ground height is calculated to obtain a height fluctuation value. Subsequently, according to the two fluctuation values, the abnormal conditions of each temporary grid are analyzed, the grids with abnormal fluctuations are identified, these abnormal grids are excluded, and finally a group of extraction grids are determined.

[0107] The second reflectivity fluctuation value and the height fluctuation value can determine the abnormal grid, because the crop coverage and the ground height should maintain a certain stability under normal conditions, and the fluctuation values reflect abnormal changes caused by environmental factors or collection errors. Larger reflectivity fluctuation values and height fluctuation values generally indicate that there are obvious disturbances or abnormal phenomena in the region, such as equipment failure, weather changes, etc., which will cause abnormalities in crops or ground features, thereby helping to identify these grids as abnormal grids. Therefore, by calculating and analyzing the two fluctuation values, abnormal grids can be effectively screened out, noise interference can be removed, and the accuracy of subsequent processing can be ensured.

[0108] By combining the average spectral reflectivity and the fluctuation value of the ground height, abnormal grids can be effectively identified, thereby filtering out unreliable grid data, ensuring that reliable and stable data is selected in subsequent processing, and improving the accuracy of data analysis and overall processing effect.

[0109] Please continue to refer to Figure 3 as shown in the determination logic diagram for determining abnormal grids of the embodiment;

[0110] Specifically, determining a plurality of abnormal grids according to the second reflectivity fluctuation value and the height fluctuation value includes:

[0111] Drawing a second reflectivity change curve according to the second reflectivity fluctuation value;

[0112] Drawing a height change curve according to the height fluctuation value;

[0113] Calculating the cosine similarity of the second reflectivity change curve and the height change curve to form a change synchronization degree;

[0114] When the change synchronization degree is less than the preset synchronization threshold, determining that the temporary grid is an abnormal grid to form a plurality of abnormal grids.

[0115] First, a reflectivity change curve and a height change curve are drawn according to the second reflectivity fluctuation value and the height fluctuation value. These curves reflect the change trend of the crop coverage and the ground height in a specific period. Then, by calculating the cosine similarity of the two curves, a change synchronization degree is obtained. When the change synchronization degree is less than the preset synchronization threshold, it indicates that the changes in reflectivity and height are not synchronized, which may be due to abnormal factors, thereby determining that the grid is an abnormal grid.

[0116] By analyzing the change trend of reflectivity and height and their synchronization, abnormal grids can be effectively identified, avoiding errors caused by environmental changes or device errors, thereby improving the accuracy and reliability of data processing and ensuring the effectiveness of subsequent analysis and decision-making.

[0117] Specifically, determining an endmember set according to the real-time multispectral images of all the extraction grids includes:

[0118] Preprocessing the real-time multispectral images to form processed multispectral images;

[0119] Extracting real-time spectral reflectance of all the processed multispectral images using the preset endmember extraction model to form an endmember set.

[0120] First, the real-time multispectral images of all the extraction grids are preprocessed to remove noise and irrelevant information and ensure image quality. Then, real-time spectral reflectance is extracted from the processed multispectral images using a preset endmember extraction model, which can represent the characteristic information of the ground and crops. Finally, the extracted spectral reflectance is collected to form an endmember set, which includes the spectral characteristics of different regions in the image for subsequent analysis and processing.

[0121] Through preprocessing of the multispectral images and accurate extraction of spectral reflectance, noise and irrelevant factors can be effectively eliminated, and the quality and reliability of the image data can be improved. By forming an endmember set, the characteristics of different regions can be more accurately represented, providing strong support for subsequent crop monitoring, land management, and precision agriculture.

[0122] Specifically, adjusting the preset light intensity threshold according to the endmember set and a preset standard set to form an adjusted light intensity threshold includes:

[0123] Calculating the cosine similarity of the endmember set and the preset standard set to form a set deviation;

[0124] When the set deviation is greater than a preset deviation threshold, adjusting the preset light intensity threshold according to the relative deviation of the set deviation and the preset deviation threshold and a preset first adjustment coefficient to form an adjusted light intensity threshold, wherein the adjustment method is to increase, and the relative deviation of the set deviation and the preset deviation threshold and the adjusted light intensity threshold are positively correlated.

[0125] The preset deviation threshold is a standard for judging whether the deviation between the endmember set and the preset standard set is significant, which depends on the characteristics of the data and the requirements of the actual application, and is usually set between 0.1 and 0.3. In this embodiment, it is set to 0.2, which helps to adjust within a relatively loose deviation range and avoid excessive sensitivity or excessive tolerance.

[0126] The preset first adjustment coefficient is a coefficient for adjusting the light intensity threshold, which depends on the sensitivity requirement of the adjustment, and is usually set between 0.5 and 2. In this embodiment, it is set to 1.5, which can balance the light intensity threshold adjustment after calculating the deviation, so that the adjustment process is not too sensitive or too conservative.

[0127] First, the cosine similarity is calculated according to the end member set and the preset standard set, and the set deviation is obtained. Then, when the set deviation is greater than the preset deviation threshold, the relative deviation of the set deviation and the preset deviation threshold is further calculated, and the original light intensity threshold is adjusted in combination with the preset first adjustment coefficient. Finally, the adjusted light intensity threshold is obtained.

[0128] By calculating the cosine similarity of the end member set and the standard set and dynamically adjusting the light intensity threshold based on the set deviation, the light intensity threshold can more accurately reflect the actual image features, thereby improving the processing accuracy of the system for low illumination areas. The system can be optimized according to different image features, making the light intensity threshold more adaptive in different environments, and improving the overall acquisition effect and processing quality.

[0129] Specifically, the preset synchronization threshold is adjusted based on the end member set formed by adjusting the light intensity threshold within the preset adjustment duration, and the adjusted synchronization threshold is formed, including:

[0130] Calculating the standard deviation of the number of sets of the real-time end member set within the preset adjustment duration to form a set number fluctuation value;

[0131] When the set number fluctuation value is greater than the preset number fluctuation threshold, the preset synchronization threshold is adjusted according to the relative deviation of the set number fluctuation value and the preset number fluctuation threshold and the preset second adjustment coefficient to form the adjusted synchronization threshold, wherein the adjustment mode is to increase, and the relative deviation of the set number fluctuation value and the preset number fluctuation threshold is positively correlated with the adjusted synchronization threshold.

[0132] The preset number fluctuation threshold is a standard value for judging whether the set number fluctuation is significant, which depends on the tolerance of the system to the change of the number of sets, and is usually set between 0.05 and 0.2. In this embodiment, it is set to 0.1, aiming to balance the stability and response speed of the system, to ensure timely adjustment of the synchronization threshold when the set number fluctuation is large, thereby improving the sensitivity and adaptability of the system.

[0133] The preset second adjustment coefficient is a proportional coefficient for adjusting the synchronization threshold, which depends on the reaction degree of the system to the fluctuation, and is usually set between 0.1 and 0.5. In this embodiment, it is set to 0.3, to ensure that the adjustment amplitude can effectively respond to the fluctuation and not be adjusted too much, maintaining the stability and accuracy of the system.

[0134] The preset second adjustment coefficient is a proportional coefficient for adjusting the synchronization threshold, which depends on the reaction degree of the system to the fluctuation, and is usually set between 0.1 and 0.5. In this embodiment, it is set to 0.3, to ensure that the adjustment amplitude can effectively respond to the fluctuation and not be adjusted too much, maintaining the stability and accuracy of the system.

[0135] First, the standard deviation of the set number of the real-time endomembers is calculated within the preset adjustment duration, thereby obtaining a set number fluctuation value. When the set number fluctuation value exceeds a preset number fluctuation threshold, the system adjusts the preset synchronization threshold according to the relative deviation of the set number fluctuation value from the preset number fluctuation threshold and a preset second adjustment coefficient, and finally forms an adjusted synchronization threshold.

[0136] By adjusting the synchronization threshold, the response capability to the change of light intensity threshold can be optimized, and the recognition accuracy can be improved. Using the set number fluctuation value to determine whether to adjust can effectively avoid unnecessary adjustment, while ensuring high working performance in the case of large fluctuation and enhancing the ability to adapt to complex environmental changes.

[0137] Please continue to refer to Figure 4 The determination logic diagram of the low-illumination grid determined by the embodiment is shown in the figure;

[0138] Specifically, a plurality of low-illumination grids are determined according to the real-time light intensity and a preset light intensity threshold, including:

[0139] When the real-time light intensity is less than the preset light intensity threshold, the system determines that the to-be-processed grid is a low-illumination grid.

[0140] By monitoring the light intensity in real time and comparing it with the preset light intensity threshold, when the real-time light intensity is lower than the threshold, the system automatically identifies and marks these grids as low-illumination grids.

[0141] By introducing the preset light intensity threshold, the areas with insufficient light can be effectively screened out, helping to accurately identify the low-light areas that need special attention, optimizing subsequent processing and analysis, and improving the response capability to environmental condition changes in crop monitoring, agricultural management or related fields.

[0142] Specifically, the real-time multispectral image is preprocessed to form a processed multispectral image, including:

[0143] A preset convex optimization model is used to remove noise from the processed multispectral image to form a processed multispectral image.

[0144] The preset convex optimization model is an image processing method based on convex optimization theory, aiming to remove noise and maintain effective signals by optimizing spectral reflectance and other features in the image. This model usually uses mathematical optimization techniques to minimize the impact of noise by setting an objective function, and makes the spectral features of the image more consistent with the preset standards. By introducing convex constraints, the model ensures that the structure of the image remains convex during the optimization process, effectively removing external interference while preserving the effective information of the image. Common applications include noise removal, contrast enhancement, etc. In this embodiment, the preset convex optimization model is a Total Variation (TV) denoising model based on convex optimization theory, which removes noise by minimizing the total variation of image gradients and preserves edges and details in the image. This method treats noise in the image as high-frequency components by setting an objective function, and removes these components through a convex optimization process to preserve the effective spectral information of the image.

[0145] First, the real-time multispectral image is preprocessed, and the preset convex optimization model is applied to remove noise in the image to obtain a clear processed multispectral image. This optimization model optimizes the spectral features of the image to eliminate unnecessary interference signals.

[0146] By applying the preset convex optimization model, the image quality can be effectively improved, and the impact of noise on subsequent analysis can be reduced, thereby improving the accuracy and reliability of image processing and providing more accurate data support for subsequent feature extraction, analysis and decision-making.

[0147] So far, the technical solutions of the present application have been described in conjunction with the preferred embodiments shown in the drawings, but those skilled in the art will readily understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to related technical features without departing from the principles of the present application, and the technical solutions after such changes or replacements will fall within the protection scope of the present application.

[0148] The above description is only the preferred embodiments of the present application and is not intended to limit the present application; those skilled in the art can make various changes and modifications to the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for unmanned aerial vehicle (UAV) based multispectral image intelligent processing, comprising: The method comprises: acquiring real-time light intensity, real-time crop coverage, real-time average ground height and real-time multispectral image of each grid in the farmland unmanned aerial vehicle collection area based on grid division; determining a plurality of low-illumination grids according to the real-time light intensity and a preset light intensity threshold value; determining a plurality of temporary grids according to the real-time crop coverage and the real-time multispectral image of each low-illumination grid; determining a plurality of extraction grids according to the real-time multispectral image, the real-time average ground height in each temporary grid and a preset synchronization threshold value; determining an endmember set according to the real-time multispectral image of all the extraction grids; adjusting the preset light intensity threshold value according to the endmember set and a preset standard set to form an adjusted light intensity threshold value; adjusting the preset synchronization threshold value according to the endmember set formed based on the adjusted light intensity threshold value within a preset adjustment time length to form an adjusted synchronization threshold value; outputting the endmember set formed based on the adjusted synchronization threshold value; adjusting the preset light intensity threshold value according to the endmember set and a preset standard set to form an adjusted light intensity threshold value comprises: calculating the cosine similarity of the endmember set and the preset standard set to form a set deviation; when the set deviation is greater than a preset deviation threshold value, adjusting the preset light intensity threshold value according to the relative deviation of the set deviation and the preset deviation threshold value and a preset first adjustment coefficient to form an adjusted light intensity threshold value; adjusting the preset synchronization threshold value according to the endmember set formed based on the adjusted light intensity threshold value within a preset adjustment time length to form an adjusted synchronization threshold value comprises: calculating the standard deviation of the number of real-time endmember sets within the preset adjustment time length to form a set number fluctuation value; when the set number fluctuation value is greater than a preset number fluctuation threshold value, adjusting the preset synchronization threshold value according to the relative deviation of the set number fluctuation value and the preset number fluctuation threshold value and a preset second adjustment coefficient to form an adjusted synchronization threshold value. 2.The unmanned aerial vehicle based multi-spectral image intelligent processing method according to claim 1, characterized in that, determining a plurality of temporary grids according to the real-time crop coverage and the real-time multispectral image of each low-illumination grid comprises: extracting real-time spectral reflectance of the real-time multispectral image using a preset endmember extraction model; calculating the average value of the real-time spectral reflectance to form an average spectral reflectance; calculating the standard deviation of the real-time crop coverage within a preset first determination time length to form a coverage fluctuation value; calculating the standard deviation of the average spectral reflectance within the preset first determination time length to form a first reflectance fluctuation value; determining a plurality of temporary grids according to the coverage fluctuation value and the first reflectance fluctuation value. 3.The UAV-based multispectral image intelligent processing method of claim 2, wherein, determining a plurality of temporary grids according to the coverage fluctuation value and the first reflectance fluctuation value comprises: drawing a coverage change curve according to the coverage fluctuation value; drawing a first reflectance change curve according to the first reflectance fluctuation value; calculating the cosine similarity of the coverage change curve and the first reflectance change curve to form a change consistency degree; when the change consistency degree is greater than a preset consistency degree threshold value, determining that the low-illumination grid is a temporary grid. 4.The UAV-based multispectral image intelligent processing method of claim 3, wherein, determining a plurality of extraction grids according to the real-time multispectral image, the real-time average ground height in each temporary grid and a preset synchronization threshold value comprises: calculating a standard deviation of the average spectral reflectance in a preset second determination duration, to form a second reflectivity fluctuation value; calculating a standard deviation of the real-time average ground height in the preset second determination duration, to form a height fluctuation value; determining a plurality of abnormal grids according to the second reflectivity fluctuation value and the height fluctuation value; excluding the abnormal grids from all the to-be-processed grids, to form a plurality of extracted grids. 5.The UAV-based multispectral image intelligent processing method of claim 4, wherein, determining a plurality of abnormal grids according to the second reflectivity fluctuation value and the height fluctuation value includes: plotting a second reflectivity change curve according to the second reflectivity fluctuation value; plotting a height change curve according to the height fluctuation value; calculating a cosine similarity of the second reflectivity change curve and the height change curve, to form a change synchronization degree; when the change synchronization degree is less than a preset synchronization threshold, determining that the temporary grid is an abnormal grid, to form a plurality of abnormal grids. 6.The UAV-based multispectral image intelligent processing method of claim 5, wherein, determining an endmember set according to the real-time multispectral images of all the extracted grids includes: preprocessing the real-time multispectral images, to form processed multispectral images; extracting real-time spectral reflectance of all the processed multispectral images using the preset endmember extraction model, to form an endmember set. 7.The UAV-based multispectral image intelligent processing method of claim 6, wherein, determining a plurality of low-illumination grids according to the real-time illumination intensity and a preset light intensity threshold includes: when the real-time illumination intensity is less than the preset light intensity threshold, determining that the to-be-processed grid is a low-illumination grid. 8.The UAV-based multispectral image intelligent processing method of claim 7, wherein, preprocessing the real-time multispectral images, to form processed multispectral images includes: removing noise of the processed multispectral images using a preset convex optimization model, to form processed multispectral images.

Images