Corn emergence rate calculation method and system based on unmanned aerial vehicle remote sensing image and machine learning

Abstract

The invention discloses a corn emergence rate calculation method and system based on unmanned aerial vehicle remote sensing images and machine learning, and relates to the field of corn emergence rate calculation. The problems that errors exist when corn seedlings are extracted and recognized through existing unmanned aerial vehicle remote sensing images, pertinence is not high when a common plant seedling rate detection method is used, and the corn seedlings are difficult to recognize at the initial stage are solved. The method comprises the following steps: constructing an aerial image data set; performing wave band registration, general DSM, single orthoimage, mosaic orthoimage and radiation calibration on the aerial image data set, and then completing splicing; performing color transformation on the spliced image to obtain a binary image; extracting corn features in the binary image according to threshold segmentation and image scale scaling transformation; calculating the number of corn seedlings and the emergence rate extracted by the unmanned aerial vehicle according to the seeding quantity; and constructing a corn emergence rate model according to a random forest algorithm and the actual emergence rate, and obtaining the emergence rate according to the corn emergence rate model. The method is applied to the agricultural field.

Description

Technical Field

[0001] The present invention relates to the field of corn emergence rate calculation, and in particular to a method for calculating corn emergence rate based on unmanned aerial vehicle remote sensing images and machine learning. Background Art

[0002] Corn is one of the three major staple crops in my country and an important food and feed crop. Timely acquisition of seedling emergence conditions is of great guiding significance for subsequent field decision-making and agricultural management. Usually, the number of corn seedlings is obtained by manual counting during the corn seedling stage, and this is used as the basis for evaluating the quality of hired machinery and operators, and determining the operating costs of the machinery and operators. However, for large-scale field planting, the workload of counting the number of corn seedlings is large and inefficient, which is not conducive to timely evaluation and will also cause disputes due to human judgment. In production, there is an urgent need for a real-time, efficient, accurate and objective seedling quality evaluation method to supervise the operation of operators and improve the quality of sowing. With the development of drone technology, drones equipped with multispectral imaging equipment have also been used as a new means of information acquisition in the fields of crop growth monitoring, regulatory measures decision-making and yield prediction.

[0003] Traditional corn seedling counting methods are mostly based on the estimation of row spacing and plant spacing. On the one hand, they are based on manual measurement and estimation, which is time-consuming and labor-intensive. On the other hand, they have low accuracy and cannot provide accurate guidance for the next step. In the existing technology of extracting corn seedling number information based on drone remote sensing images, image noise points are not processed in image processing. Image noise points may be mistaken for corn seedlings, resulting in the erroneous identification of non-existent corn seedlings. These noise points usually have irregular shapes or prominent grayscale differences, which may be similar to the background vegetation or surface features, making it difficult for image processing algorithms to distinguish, thus affecting the statistics of seedling numbers. In northern regions, corn needs to return straw to the field after harvest. Although the existing plant seedling rate detection methods are highly universal and suitable for most plants, they are not targeted and it is difficult to remove the influence of non-corn (straw, weeds, etc.). Corn is a monocotyledonous grass crop. It needs to be distinguished from common seedlings in the field, such as soybeans and broad-leaved weeds, by using drones to identify phenotypic traits. However, it is difficult to identify grass weeds in the early stage. Summary of the invention

[0004] Aiming at the problem that there are errors in the existing unmanned aerial vehicle remote sensing image extraction and identification of corn seedlings, and the common plant seedling rate detection method is not targeted and it is difficult to identify corn seedlings in the early stage, the present invention proposes a method for calculating corn seedling rate based on unmanned aerial vehicle remote sensing images and machine learning, and the method comprises:

[0005] S1: Use drone multispectral to collect corn seedling plot image data, and build an aerial image dataset based on the corn seedling plot image data;

[0006] S2: The aerial image dataset is stitched after band registration, rough digital surface model DSM, single orthophoto, mosaic orthophoto, and radiometric calibration;

[0007] S3: Perform color transformation on the spliced ​​image to obtain a binary image;

[0008] S4: Extract corn features from binary images based on threshold segmentation and image scaling transformation;

[0009] S5: Calculate the number of corn seedlings and the emergence rate extracted by drone according to the sowing amount;

[0010] S6: Construct a corn germination rate model based on the random forest algorithm and the actual germination rate, and obtain the germination rate based on the corn germination rate model.

[0011] Furthermore, a preferred embodiment is provided, wherein step S2 comprises:

[0012] Acquire a set of multi-band aerial images from the drone, wherein the aerial images include overlapping areas;

[0013] Mark the known ground control points on the image;

[0014] It is established by aerial triangulation method based on the overlapping parts of aerial images and control points;

[0015] The spatial coordinates of each image point are calculated through the geometric model of the aerial image;

[0016] Matching and stereo matching are performed based on the overlapping areas contained in the aerial images;

[0017] Using stereo matching algorithms, extract three-dimensional point cloud data of the ground surface from images of different perspectives;

[0018] The three-dimensional point cloud data is smoothed by interpolation method to generate a complete digital surface model;

[0019] Align images of different bands into a unified digital surface model;

[0020] According to the known ground control points and the exterior orientation elements of the aerial images, the geometric model of the aerial images is used to perform radiometric correction, differential correction and mosaicking to obtain a single orthophoto.

[0021] Stitch multiple orthophotos, perform radiation balance processing on the stitched images, and obtain a complete orthophoto map;

[0022] Determine the radiometric calibration coefficients of the image through ground control points;

[0023] According to the radiation calibration model, the brightness grayscale value in the orthophoto is converted into the actual radiation brightness to complete the radiation calibration.

[0024] Furthermore, a preferred embodiment is provided, wherein step S3 comprises:

[0025] Performing nonlinear transformation of the RGB mode on the image to obtain an HSV color space image, wherein the HSV color space image is a binary image;

[0026] Among them, the red in hue H is 0, the green is 120°, and the blue is 240°; the saturation S is 0 represents gray, 100% reaches full saturation, and the value range is 0 to 1.0; the value range of lightness V is 0 to 255;

[0027] The color transformation formula is:

[0028] EXG=2b 555 -b 660 -b 450

[0029] Among them, EXG is the over-green index, b 555 is the reflectivity at 555nm, b 660 is the reflectivity at 660nm, b 450 is the reflectivity at 450nm.

[0030] Furthermore, a preferred embodiment is provided, wherein step S4 comprises:

[0031] The absolute radiance threshold of corn seedlings and non-corn seedlings is extracted using Cursor Location on the binary image;

[0032] The corn seedlings and non-corn seedlings were distinguished based on the brightness threshold;

[0033] The noise pixels in the corn seedling image after threshold segmentation are removed by using a 3×3 convolution kernel function;

[0034] The kernel function image is converted from raster format to vector format by using image conversion, and only the corn seedling vector image is extracted, and the corn seedling vector image is superimposed on the true color image;

[0035] Using vector recognition technology, the characteristics of corn seedlings are identified in the overlay image.

[0036] Furthermore, a preferred embodiment is provided, wherein the convolution kernel function is:

[0037]

[0038] Among them, g(x,y) is the processed matrix, w is the convolution kernel, f(x,y) is the original matrix, -a≤s≤a and -b≤t≤b, s and t are variables used to determine the summation range in the convolution kernel function.

[0039] Furthermore, a preferred embodiment is provided, wherein S5 comprises:

[0040] C=G / P*100%

[0041] Among them, C represents the corn seedling emergence rate extracted by UAV remote sensing, G represents the number of corn seedlings extracted by UAV remote sensing, and P represents the number of corn sown.

[0042] Furthermore, a preferred embodiment is provided, wherein S6 comprises: using a random forest algorithm to perform regression modeling analysis on the germination rate extracted by drone remote sensing and the actual germination rate to construct a corn germination rate model.

[0043] Based on the same inventive concept, the present invention also proposes a corn emergence rate calculation system based on UAV remote sensing images and machine learning, the system comprising:

[0044] An image acquisition unit is used to collect image data of corn seedling plots using multispectral drones, and to construct an aerial image data set based on the image data of corn seedling plots;

[0045] Image stitching unit, used to stitch the aerial image data set after band registration, rough digital surface model DSM, single orthophoto, mosaic orthophoto, and radiometric calibration;

[0046] A transformation unit, used for performing color change on the spliced ​​image to obtain a binary image;

[0047] A feature extraction unit is used to extract corn features in the binary image according to threshold segmentation and image scale scaling transformation;

[0048] a calculation unit for calculating the number of corn seedlings and the emergence rate extracted by the drone according to the sowing amount;

[0049] The seedling emergence rate acquisition unit is used to build a corn seedling emergence rate model according to the random forest algorithm and the actual seedling emergence rate, and obtain the seedling emergence rate according to the corn seedling emergence rate model.

[0050] Based on the same inventive concept, the present invention also proposes a computer device, including a memory and a processor, wherein the memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes a method for calculating corn emergence rate based on drone remote sensing images and machine learning as described in any one of the above items.

[0051] Based on the same inventive concept, the present invention also proposes a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of a method for calculating corn emergence rate based on drone remote sensing images and machine learning as described above are executed.

[0052] The present invention is beneficial in that:

[0053] Traditional plant seedling rate detection methods usually rely on manual visual evaluation, which is greatly affected by subjective factors, and errors are prone to occur in the early stage of corn seedlings due to the small number of seedlings and difficulty in identification. Through unmanned aerial vehicle remote sensing images and machine learning technology, a method for calculating corn seedling emergence rate based on unmanned aerial vehicle remote sensing images and machine learning proposed in the present invention can more accurately extract corn seedling characteristics and calculate a more accurate emergence rate, especially in the early stage of seedlings. The present invention uses unmanned aerial vehicle remote sensing technology to collect high-resolution multispectral images, and performs automated analysis and calculation through image processing and machine learning algorithms, which reduces manual intervention and greatly improves efficiency and accuracy. In particular, the corn seedling emergence rate prediction model established by the random forest algorithm can obtain efficient prediction capabilities based on a large amount of data training, reducing the impact of human factors on the emergence rate.

[0054] Traditional seedling rate detection methods generally have the problem of insufficient sensitivity, especially when the number of seedlings is scarce in the early stage, it is difficult to detect accurately. The present invention proposes a method for calculating corn emergence rate based on drone remote sensing images and machine learning. Through analysis based on multispectral remote sensing images, it can be highly adaptable to different plots and different environmental conditions and has stronger pertinence. The present invention provides a complete process from data acquisition, preprocessing, image analysis to machine learning modeling, which can systematically optimize the data analysis process, thereby more accurately identifying the emergence of corn seedlings, saving a lot of time and cost for manual detection, and the emergence accuracy rate is as high as 98.3%.

[0055] The present invention proposes a method for calculating corn emergence rate based on unmanned aerial vehicle remote sensing images and machine learning, which uses a multispectral sensor carried by an unmanned aerial vehicle to collect image data in a corn seedling plot. Multispectral images can capture spectral information of different bands (such as red light, green light, blue light, near infrared, etc.), which is crucial for crop identification and health status analysis. The reflection characteristics of vegetation in the corn seedling stage are significantly different in different bands, so corn seedlings can be distinguished from other objects or weeds by band combination. The collected aerial images need to undergo a series of processing steps such as band registration, digital surface model (DSM) construction, orthophoto stitching, and radiation calibration. The purpose of these processing steps is to stitch multiple images into a complete high-resolution map, eliminate distortion and errors, and ensure the accuracy and consistency of image data. The processed images are analyzed for color changes, and the images are converted into binary images to highlight the features of the corn seedlings (such as green areas). In the binary image, the difference between the corn seedlings and the background is more obvious, which is helpful for subsequent feature extraction and analysis. Through methods such as threshold segmentation and scale transformation, the area of ​​corn seedlings can be extracted, and the density and number of seedlings can be calculated. After extracting the corn seedling area and calculating the number of seedlings, the present invention uses a random forest algorithm to construct a prediction model for corn emergence rate. Classification or regression analysis is performed through voting of multiple decision trees, and it performs well when processing high-dimensional data and noise data. By training the model, it can accurately predict the emergence rate of corn seedlings based on the input remote sensing image data. According to the output of the seeding amount and the machine learning model, the emergence rate of the current corn plot is calculated. This calculation result is more accurate than the traditional method, can reflect the growth status of corn seedlings, and provide farmers or agricultural managers with more reliable data support.

[0056] The invention is applied in the field of agriculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a flow chart of a method for calculating corn emergence rate based on UAV remote sensing images and machine learning according to implementation mode 1;

[0058] Figure 2 This is a schematic diagram of absolute radiance threshold segmentation according to the fourth implementation mode;

[0059] Figure 3 This is a schematic diagram of image overlay described in Implementation Method 4. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0061] Implementation method 1, see Figure 1 The present embodiment describes a method for calculating corn emergence rate based on drone remote sensing images and machine learning, the method comprising:

[0062] S1: Use drone multispectral to collect corn seedling plot image data, and build an aerial image dataset based on the corn seedling plot image data;

[0063] S2: The aerial image dataset is stitched after band registration, rough digital surface model DSM, single orthophoto, mosaic orthophoto, and radiometric calibration;

[0064] S3: Perform color transformation on the spliced ​​image to obtain a binary image;

[0065] S4: Extract corn features from binary images based on threshold segmentation and image scaling transformation;

[0066] S5: Calculate the number of corn seedlings and the emergence rate extracted by drone according to the sowing amount;

[0067] S6: Construct a corn germination rate model based on the random forest algorithm and the actual germination rate, and obtain the germination rate based on the corn germination rate model.

[0068] When extracting corn seedlings from traditional drone remote sensing images, errors often occur, especially in the early stage of corn growth, when the morphology of the seedlings is relatively similar, resulting in the inability of traditional image recognition technology to accurately extract corn seedlings, and prone to missed detection or false detection. The method proposed in this embodiment combines multispectral remote sensing image data with machine learning, using multispectral image data collected by drones to extract the characteristics of corn seedlings. Multispectral images can capture information in different bands, which helps to identify and distinguish the differences between corn and the surrounding environment in the early corn seedling stage, thereby effectively reducing recognition errors.

[0069] Traditional plant seedling rate detection methods usually use general algorithms or technologies, which cannot perform customized identification for seedling characteristics of specific crops such as corn, resulting in unsatisfactory detection results. The method proposed in this embodiment combines the growth characteristics of corn seedlings to construct a method for calculating corn seedling rate based on drone remote sensing images, especially by performing color change processing, binary segmentation and other processing on the images, which can specifically extract the characteristics of corn seedlings, thereby improving the detection accuracy of corn seedlings.

[0070] At the same time, in the early stage of growth, corn seedlings are difficult to accurately identify due to their small size and similar color to weeds or soil background, resulting in low initial recognition accuracy of traditional detection methods. In the method proposed in this embodiment, the influence of factors such as illumination, viewing angle, and sensor characteristics are eliminated through image preprocessing steps such as band registration, digital surface model (DSM) construction, and orthophoto stitching, thereby enhancing the recognition characteristics of corn seedlings. In addition, feature extraction through color change and binarization technology can more clearly identify the outline of corn seedlings, and even in the early seedling stage, the distribution of corn seedlings can be accurately detected.

[0071] Many traditional corn emergence rate calculation methods, especially algorithms based on visual recognition, are often affected by noise and uneven illumination, resulting in large deviations in the calculated emergence rate. The method proposed in this embodiment combines the random forest algorithm with actual emergence rate data to construct a calculation model for corn emergence rate. As a powerful machine learning method, the random forest algorithm can effectively handle complex nonlinear relationships, optimize the influence of features, reduce the interference of noise, and enhance the robustness and accuracy of the model. Therefore, the emergence rate of corn can be accurately calculated under different environmental conditions.

[0072] The present invention proposes a method for calculating corn emergence rate based on UAV remote sensing images and machine learning, the principle of which is:

[0073] Image data was collected from corn seedling plots using multispectral sensors carried by drones. Multispectral images can capture spectral information in different bands (such as red light, green light, blue light, near infrared, etc.), which is crucial for crop identification and health status analysis. The reflectance characteristics of vegetation in the corn seedling stage are significantly different in different bands, so corn seedlings can be distinguished from other objects or weeds by band combination. The collected aerial images need to undergo a series of processing steps such as band registration, digital surface model (DSM) construction, orthophoto stitching, and radiometric calibration. The purpose of these processing steps is to stitch multiple images into a complete high-resolution map, eliminate distortion and errors, and ensure the accuracy and consistency of image data. The processed images are analyzed for color changes, and the images are converted into binary images to highlight the features of corn seedlings (such as green areas). In the binary image, the difference between corn seedlings and the background is more obvious, which is helpful for subsequent feature extraction and analysis. Through methods such as threshold segmentation and scale transformation, the area of ​​corn seedlings can be extracted, and the density and number of seedlings can be calculated. After extracting the corn seedling area and calculating the number of seedlings, the present invention uses a random forest algorithm to construct a prediction model for the corn seedling emergence rate. Classification or regression analysis is performed through voting of multiple decision trees, and it performs well when processing high-dimensional data and noise data. By training the model, it can accurately predict the emergence rate of corn seedlings based on the input remote sensing image data. According to the seeding amount and the output of the machine learning model, the emergence rate of the current corn plot is calculated. This calculation result is more accurate than the traditional method, can reflect the growth status of corn seedlings, and provide farmers or agricultural managers with more reliable data support.

[0074] Implementation method 2: This implementation method further limits the method for calculating corn emergence rate based on drone remote sensing images and machine learning described in implementation method 1. Step S2 includes:

[0075] Acquire a set of multi-band aerial images from the drone, wherein the aerial images include overlapping areas;

[0076] Mark the known ground control points on the image;

[0077] It is established by aerial triangulation method based on the overlapping parts of aerial images and control points;

[0078] The spatial coordinates of each image point are calculated through the geometric model of the aerial image;

[0079] Matching and stereo matching are performed based on the overlapping areas contained in the aerial images;

[0080] Using stereo matching algorithms, extract three-dimensional point cloud data of the ground surface from images of different perspectives;

[0081] The three-dimensional point cloud data is smoothed by interpolation method to generate a complete digital surface model;

[0082] Align images of different bands into a unified digital surface model;

[0083] According to the known ground control points and the exterior orientation elements of the aerial images, the geometric model of the aerial images is used to perform radiometric correction, differential correction and mosaicking to obtain a single orthophoto.

[0084] Stitch multiple orthophotos, perform radiation balance processing on the stitched images, and obtain a complete orthophoto map;

[0085] Determine the radiometric calibration coefficients of the image through ground control points;

[0086] According to the radiation calibration model, the brightness grayscale value in the orthophoto is converted into the actual radiation brightness to complete the radiation calibration.

[0087] In this embodiment, by calibrating known ground control points in the image and performing aerial triangulation, the spatial position of the image can be accurately calculated. This method can effectively eliminate geometric distortion caused by factors such as camera tilt and aerial photography angle, and improve the spatial accuracy of the image. The three-dimensional point cloud data is extracted from the multi-angle image using a stereo matching algorithm, and a high-precision digital surface model is generated through interpolation and smoothing. In this way, the slight differences in ground undulation and vegetation growth can be accurately captured, and the accuracy of the surface model can be improved. Images of different bands are registered to a unified digital surface model, so that the geometric matching between images of different bands is more accurate, and the distortion problems caused by factors such as camera posture changes and terrain undulations are reduced. By using aerial triangulation and external orientation elements for image geometric correction, a more accurate orthophoto can be obtained, providing a more reliable data basis for subsequent analysis. By calibrating the image by ground control points, the influence of factors such as atmosphere and lighting conditions on the image brightness can be eliminated, and the grayscale value of the image can be converted into actual radiation brightness, ensuring that the radiation information in the image is consistent with the actual ground conditions, and providing more accurate reflectivity data for the calculation of vegetation index and analysis of corn emergence rate.

[0088] Furthermore, in the process of stitching aerial images, the use of radiation balance processing can effectively eliminate the brightness differences between different images caused by shooting angles, time and other factors, so that the stitched images have better uniformity and consistency. This is especially important for the analysis of corn emergence rate in a large area, which can avoid the error caused by stitching.

[0089] This embodiment is explained in conjunction with the first embodiment. The method described in this embodiment is combined with machine learning technology. By analyzing the calibrated image data and its spatial information, the model is trained to achieve efficient calculation of corn emergence rate. Through comprehensive processing of multiple band images, features related to corn growth can be automatically extracted, reducing manual intervention and improving processing efficiency and accuracy.

[0090] At the same time, the method also uses advanced aerial triangulation, image matching and radiation calibration technology, so that it can adapt to aerial image processing under various complex terrains and different environmental conditions. Whether it is flat farmland or undulating mountainous areas, accurate prediction of corn emergence rate can be obtained, which has high adaptability. The method proposed in this embodiment is not only suitable for small-scale sample analysis, but also can process large-scale regional data. In agricultural monitoring and crop growth prediction, especially in the management of large-scale farmland, this method has significant advantages.

[0091] Furthermore, the brightness value of the image is converted into the real radiation brightness through radiometric calibration, so that the image data can reflect the surface conditions more realistically. This is crucial for accurately identifying vegetation characteristics and crop growth conditions (such as corn seedling emergence), and can significantly improve the accuracy of seedling emergence rate calculation.

[0092] Implementation method 3: This implementation method further limits the method for calculating corn emergence rate based on drone remote sensing images and machine learning described in implementation method 1. Step S3 includes:

[0093] Performing nonlinear transformation of the RGB mode on the image to obtain an HSV color space image, wherein the HSV color space image is a binary image;

[0094] Among them, the red in hue H is 0, the green is 120°, and the blue is 240°; the saturation S is 0 represents gray, 100% reaches full saturation, and the value range is 0 to 1.0; the value range of lightness V is 0 to 255;

[0095] The color transformation formula is:

[0096] EXG=2b 555 -b 660 -b 450

[0097] Among them, EXG is the over-green index, b 555 is the reflectivity at 555nm, b 660 is the reflectivity at 660nm, b 450 is the reflectivity at 450nm.

[0098] In this embodiment, the RGB image is converted into the HSV color space by nonlinear transformation. The HSV color space is closer to the perception of the human eye than RGB, especially in hue (H), which can better separate color information. The converted HSV image helps to more easily extract the color characteristics of the plant in subsequent processing, especially in agricultural monitoring, where the change of color can often reflect the growth status and health of the plant. Through nonlinear transformation, the reflectivity of different bands can be adjusted to more accurately reflect the characteristics of the corn seedlings. Such a method can effectively reduce the interference of background noise and improve the recognition accuracy of the target object. EXG, as an index to measure the "green" degree of vegetation, plays an important role in distinguishing vegetation from non-vegetation background. By calculating the overgreen index based on the reflectivity of specific bands (555nm, 660nm, 450nm), the greenness of the plant can be effectively identified and highlighted, thereby more clearly locating the area where the corn seedlings emerge. Through the calculation of the EXG formula, it is possible to more accurately determine which areas are vegetation areas, thereby excluding non-vegetation backgrounds. Especially in complex backgrounds, EXG can effectively avoid interference from non-vegetation areas such as soil and buildings.

[0099] In the HSV color space, hue (H) can better reflect the color changes in the image, especially the green plant area, avoiding the influence of factors such as light changes on the image processing results in the RGB mode. Saturation (S) and brightness (V) can provide additional image features when further extracting vegetation information, increasing the robustness of the model to data under different environmental and weather conditions. The EXG index is particularly sensitive to green vegetation areas and can improve the recognition rate of corn seedlings, especially in the early growth stages, when the color difference between the plant and the soil is small and the image quality is poor. This method can still effectively identify the growth area of ​​corn. In this embodiment, based on the reflectivity of different bands (555nm, 660nm, 450nm), different types of remote sensing data can be processed to adapt to data provided by a variety of drone sensors (such as RGB, NDVI or other multispectral images).

[0100] Implementation method 4: See Figure 2 and Figure 3 This embodiment further defines the method for calculating corn emergence rate based on UAV remote sensing images and machine learning described in Embodiment 1, and the step S4 includes:

[0101] The absolute radiance threshold of corn seedlings and non-corn seedlings is extracted using Cursor Location on the binary image;

[0102] The corn seedlings and non-corn seedlings were distinguished based on the brightness threshold;

[0103] The noise pixels in the corn seedling image after threshold segmentation are removed by using a 3×3 convolution kernel function;

[0104] The kernel function image is converted from raster format to vector format by using image conversion, and only the corn seedling vector image is extracted, and the corn seedling vector image is superimposed on the true color image;

[0105] Using vector recognition technology, the characteristics of corn seedlings are identified in the overlay image.

[0106] In this embodiment, the absolute radiance threshold of corn seedlings and non-corn seedlings is extracted by Cursor Location / Value (cursor position / value), and corn seedlings and background noise can be accurately distinguished in remote sensing images. This method relies on the contrast of image brightness, which can improve the accuracy of recognition, especially under the condition of certain environmental noise, and can effectively filter out the interference of other vegetation or ground features. The 3×3 convolution kernel function is used to remove the noise pixels in the image after threshold segmentation, which further improves the image quality. The convolution operation can effectively smooth the image and reduce noise through local pixel processing, thereby ensuring that subsequent feature extraction is more accurate. This method also enhances the adaptability to low-quality images and is suitable for remote sensing data under different lighting, weather and other conditions. The image processed by the kernel function is converted from raster format to vector format, and a vector diagram containing only corn seedlings is extracted, making subsequent analysis more efficient. The vector format usually takes up less storage space than the raster format, and the processing speed is faster in the later image analysis, feature extraction and other processes. In addition, the vector format is conducive to the further application of vector recognition technology and the improvement of the recognition ability of corn seedling features. The vector map is superimposed on the true color image, so that the distribution characteristics of corn seedlings can be more intuitively compared with the actual image, providing a more intuitive and accurate reference. This superposition method combines the real scene of the remote sensing image and the analysis results after machine learning processing, which helps to enhance the credibility and interpretability of the results. Vector recognition technology can further accurately identify the characteristics of corn seedlings in the vector map, especially when performing seedling emergence statistics, it can effectively distinguish corn seedlings in different areas and calculate the precise distribution. Vector recognition technology can usually handle features of complex shapes and structures, which is particularly important for identifying corn seedlings in complex terrain and multiple vegetation backgrounds.

[0107] Implementation mode 5: This implementation mode is a further limitation of the method for calculating corn emergence rate based on UAV remote sensing images and machine learning described in Implementation mode 4, and the convolution kernel function is:

[0108]

[0109] Among them, g(x,y) is the processed matrix, w is the convolution kernel, f(x,y) is the original matrix, -a≤s≤a and -b≤t≤b, s and t are variables used to determine the summation range in the convolution kernel function.

[0110] Implementation method 6: This implementation method further limits the method for calculating corn emergence rate based on drone remote sensing images and machine learning described in implementation method 1, and S5 includes:

[0111] C=G / P*100%

[0112] Among them, C represents the corn seedling emergence rate extracted by UAV remote sensing, G represents the number of corn seedlings extracted by UAV remote sensing, and P represents the number of corn sown.

[0113] Implementation method seven. This implementation method is a further limitation of the method for calculating corn emergence rate based on UAV remote sensing images and machine learning described in Implementation method one. S6 includes: using a random forest algorithm to perform regression modeling analysis on the emergence rate extracted by UAV remote sensing and the actual emergence rate to construct a corn emergence rate model.

[0114] In this embodiment, the use of drone remote sensing technology can quickly and extensively obtain soil and crop information in the field through high-resolution images. Compared with traditional manual inspections or ground surveys, the use of drones for remote sensing can greatly improve the efficiency and coverage of data collection, especially in the monitoring of large areas of fields. Remote sensing images can achieve real-time and comprehensive data collection, reducing the time and cost of manual surveys. Using the random forest algorithm, the emergence rate information in the drone remote sensing image is regressed and modeled with the actual observation value, which can effectively capture and identify the complex relationship between the emergence rate and other environmental factors (such as soil moisture, temperature, light, etc.).

[0115] Through regression modeling analysis and using the random forest algorithm to learn from historical data, the model can better fit the actual emergence rate, thereby improving the accuracy of the prediction. Especially in large-scale planting areas, traditional seedling emergence rate measurements may have large errors, and methods based on remote sensing images and machine learning can significantly improve the prediction accuracy of seedling emergence rate.

[0116] Embodiment 8: A corn emergence rate calculation system based on drone remote sensing images and machine learning described in this embodiment, the system includes:

[0117] An image acquisition unit is used to collect image data of corn seedling plots using multispectral drones, and to construct an aerial image data set based on the image data of corn seedling plots;

[0118] Image stitching unit, used to stitch the aerial image data set after band registration, rough digital surface model DSM, single orthophoto, mosaic orthophoto, and radiometric calibration;

[0119] A transformation unit, used for performing color change on the spliced ​​image to obtain a binary image;

[0120] A feature extraction unit is used to extract corn features in the binary image according to threshold segmentation and image scale scaling transformation;

[0121] a calculation unit for calculating the number of corn seedlings and the emergence rate extracted by the drone according to the sowing amount;

[0122] The seedling emergence rate acquisition unit is used to build a corn seedling emergence rate model according to the random forest algorithm and the actual seedling emergence rate, and obtain the seedling emergence rate according to the corn seedling emergence rate model.

[0123] Implementation method 9. A computer device described in this implementation method includes a memory and a processor, wherein a computer program is stored in the memory. When the processor runs the computer program stored in the memory, the processor executes a method for calculating corn emergence rate based on drone remote sensing images and machine learning as described in any one of implementation methods 1 to 7.

[0124] Implementation method ten, a computer-readable storage medium described in this implementation method, on which a computer program is stored, and when the computer program is executed by a processor, the steps of a method for calculating corn emergence rate based on drone remote sensing images and machine learning as described in any one of implementation methods one to seven are executed.

[0125] Implementation eleven: This implementation provides a specific example of a method for calculating corn emergence rate based on drone remote sensing images and machine learning described in Implementation One, and is also used to explain Implementations Two to Seven. Specifically:

[0126] Step 1: Use the M300RTK drone equipped with the MS 600pro multispectral camera and the global navigation and positioning system (GNSS) device to take aerial photos of the target plots of corn seedlings to obtain a series of high-precision plot images. To ensure the successful generation of orthophotos, the heading overlap rate and the lateral overlap rate should be greater than 80% during drone aerial photography. The aerial photography height is 15m.

[0127] Step 2: Using the UAV remote sensing 3D reconstruction software, a group of aerial images of the sample site taken at the same time were stitched together after band registration (aerial triangulation), rough DSM, single orthophoto, mosaic orthophoto, and radiometric calibration.

[0128] Band registration adopts the aerial triangulation method: using continuously captured aerial photographs with a certain overlap, based on a small number of field control points, a flight route model or regional network model (optical or digital) corresponding to the actual field is established by photogrammetry, thereby obtaining the plane coordinates and elevation of the encrypted points.

[0129] DSM: Digital Surface Model. It contains not only the elevation information of the terrain but also the elevation of other surface information (plants, buildings, etc.) other than the ground.

[0130] Orthophoto: A remote sensing image that has been subjected to radiation correction, differential correction and mosaicking on a pixel-by-pixel basis and cropped to a specified map range to generate image data, a plan view with surface and map outline information, etc.

[0131] Radiometric calibration: When calculating the spectral reflectance of an object, or when comparing images acquired at different times and by different sensors, the brightness grayscale value of the image is converted into absolute radiometric brightness. This process is called radiometric calibration.

[0132] Step 3: Transform the image to obtain an HSV color space image, i.e., a binary image (HSV is a nonlinear transformation of the RGB mode, where H (hue) red is 0, green is 120°, and blue is 240°; S (saturation) 0 is gray, 100% is fully saturated, and the value range is 0 to 1.0. The larger the value, the more saturated the color; V (brightness) ranges from 0 to 255, with a value of 0 being black and 255 reaching the maximum brightness being white). The color transformation formula is:

[0133] EXG=2b 555 -b 660 -b 450

[0134] In the formula, EXG is the over-green index, b 555 is the reflectivity at 555nm, b 660 is the reflectivity at 660nm, b 450 is the reflectivity at 450nm band.

[0135] Step 4: Use Cursor Location to extract the absolute radiation brightness threshold of corn and non-corn on the binary image (0.128>corn>0.015), and then use this brightness threshold to quickly distinguish corn seedlings from non-corn seedlings (soil, weeds, shadows, etc.).

[0136] Step 5: After threshold segmentation, the image still has many noise points, which will affect the subsequent recognition. (It can also be understood as a more refined removal of impurities, removing some very small pixel points with thresholds close to the corn seedlings) Since the pixels are composed of matrices, a 3×3 (moderate and will not damage the image quality of the corn seedlings themselves) convolution kernel function is used to remove image noise pixels. The convolution kernel, also called the convolution matrix or mask, is essentially a very small matrix. It mainly uses the convolution operation between the kernel and the image to achieve image processing, and can produce blurring, sharpening, bumping, edge detection and other effects. Convolution is to take each element in the original matrix and take the weighted average of its adjacent elements according to the weight distribution of the convolution kernel. The formula for the convolution operation is as follows:

[0137]

[0138] Among them, g(x,y) is the processed matrix, w is the convolution kernel, f(x,y) is the original matrix, -a≤s≤a and -b≤t≤b.

[0139] Image conversion is used to convert the kernel function image from raster format to vector format, and the vector image of only corn seedlings is extracted and superimposed on the true color image; vector recognition technology is used to identify the characteristics of corn seedlings in the superimposed image, and the number of corn seedlings is extracted.

[0140] Step 6: Calculate the number of corn seedlings and the emergence rate extracted by the drone according to the sowing amount;

[0141] The formula for calculating the emergence rate based on the actual sowing amount is:

[0142] C=G / P*100%

[0143] In the formula, C represents the corn seedling emergence rate extracted by UAV remote sensing, G represents the number of corn seedlings extracted by UAV remote sensing, and P represents the number of corn sown.

[0144] Step 7: Use the random forest algorithm to perform regression modeling analysis on the seedling emergence rate extracted by drone remote sensing and the actual seedling emergence rate to improve the accuracy of drone remote sensing seedling emergence rate.

[0145] Y={C}

[0146] In the formula, Y is the seedling emergence rate, C is the seedling emergence rate extracted by UAV remote sensing, and {} is the random forest algorithm model.

[0147] In this embodiment, the area with more impurities was selected as the experimental sample. The actual sowing amount was 227 plants, the actual number of seedlings was 197, and the seedling rate was 86.7%. The number of seedlings identified by the image was 180, and the seedling rate was 79.2%. 79.2% was imported into the random forest model, and the output seedling rate was 85.3%. The accuracy rate was as high as 98.3%.

[0148] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0149] The present disclosure is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram and the combination of processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0150] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present disclosure rather than to limit its protection scope. Although the present disclosure has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that after reading the present disclosure, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the invention, but these changes, modifications or equivalent substitutions are all within the protection scope of the disclosed claims to be approved.

Claims

1. A method for calculating corn emergence rate based on UAV remote sensing images and machine learning, characterized in that: The method comprises: S1: Use drone multispectral to collect corn seedling plot image data, and build an aerial image dataset based on the corn seedling plot image data; S2: The aerial image dataset is stitched after band registration, rough digital surface model DSM, single orthophoto, mosaic orthophoto, and radiometric calibration; S3: Perform color transformation on the spliced ​​image to obtain a binary image; S4: Extract corn features from binary images based on threshold segmentation and image scaling transformation; S5: Calculate the number of corn seedlings and the emergence rate extracted by drone according to the sowing amount; S6: Construct a corn germination rate model based on the random forest algorithm and the actual germination rate, and obtain the germination rate based on the corn germination rate model.

2. The method for calculating corn emergence rate based on UAV remote sensing images and machine learning according to claim 1, characterized in that: The step S2 comprises: Acquire a set of multi-band aerial images from the drone, wherein the aerial images include overlapping areas; Mark the known ground control points on the image; It is established by aerial triangulation method based on the overlapping parts of aerial images and control points; The spatial coordinates of each image point are calculated through the geometric model of the aerial image; Matching and stereo matching are performed based on the overlapping areas contained in the aerial images; Using stereo matching algorithms, extract three-dimensional point cloud data of the ground surface from images of different perspectives; The three-dimensional point cloud data is smoothed by interpolation method to generate a complete digital surface model; Align images of different bands into a unified digital surface model; According to the known ground control points and the exterior orientation elements of the aerial images, the geometric model of the aerial images is used to perform radiometric correction, differential correction and mosaicking to obtain a single orthophoto. Stitch multiple orthophotos, perform radiation balance processing on the stitched images, and obtain a complete orthophoto map; Determine the radiometric calibration coefficients of the image through ground control points; According to the radiation calibration model, the brightness grayscale value in the orthophoto is converted into the actual radiation brightness to complete the radiation calibration.

3. The method for calculating corn emergence rate based on UAV remote sensing images and machine learning according to claim 1, characterized in that: The step S3 comprises: Performing nonlinear transformation of the RGB mode on the image to obtain an HSV color space image, wherein the HSV color space image is a binary image; Among them, the red in hue H is 0, the green is 120°, and the blue is 240°; the saturation S is 0 represents gray, 100% reaches full saturation, and the value range is 0 to 1.0; the value range of lightness V is 0 to 255; The color transformation formula is: EXG=2b 555 -b 660 -b 450 Among them, EXG is the over-green index, b 555 is the reflectivity at 555nm, b 660 is the reflectivity at 660nm, b 450 is the reflectivity at 450nm band.

4. The method for calculating corn emergence rate based on UAV remote sensing images and machine learning according to claim 1, characterized in that: The step S4 comprises: The absolute radiance threshold of corn seedlings and non-corn seedlings is extracted using Cursor Location on the binary image. The corn seedlings and non-corn seedlings were distinguished based on the brightness threshold; The noise pixels in the corn seedling image after threshold segmentation are removed by using a 3×3 convolution kernel function; The kernel function image is converted from raster format to vector format by using image conversion, and only the corn seedling vector image is extracted, and the corn seedling vector image is superimposed on the true color image; Using vector recognition technology, the characteristics of corn seedlings are identified in the overlay image.

5. The method for calculating corn emergence rate based on UAV remote sensing images and machine learning according to claim 4, characterized in that: The convolution kernel function is: Among them, g(x,y) is the processed matrix, w is the convolution kernel, f(x,y) is the original matrix, -a≤s≤a and -b≤t≤b, s and t are variables used to determine the summation range in the convolution kernel function.

6. The method for calculating corn emergence rate based on UAV remote sensing images and machine learning according to claim 1, characterized in that: The S5 includes: C=G / P*100% Among them, C represents the corn seedling emergence rate extracted by UAV remote sensing, G represents the number of corn seedlings extracted by UAV remote sensing, and P represents the number of corn sown.

7. The method for calculating corn emergence rate based on UAV remote sensing images and machine learning according to claim 1, characterized in that: The S6 includes: using a random forest algorithm to perform regression modeling analysis on the germination rate extracted by drone remote sensing and the actual germination rate to construct a corn germination rate model.

8. A corn emergence rate calculation system based on UAV remote sensing images and machine learning, characterized in that: The system comprises: An image acquisition unit is used to collect image data of corn seedling plots using multispectral drones, and to construct an aerial image data set based on the image data of corn seedling plots; Image stitching unit, used to stitch the aerial image data set after band registration, rough digital surface model DSM, single orthophoto, mosaic orthophoto, and radiometric calibration; A transformation unit, used for performing color change on the spliced ​​image to obtain a binary image; A feature extraction unit is used to extract corn features in the binary image according to threshold segmentation and image scale scaling transformation; a calculation unit for calculating the number of corn seedlings and the emergence rate extracted by the drone according to the sowing amount; The seedling emergence rate acquisition unit is used to build a corn seedling emergence rate model according to the random forest algorithm and the actual seedling emergence rate, and obtain the seedling emergence rate according to the corn seedling emergence rate model.

9. A computer device, characterized in that: It includes a memory and a processor, wherein a computer program is stored in the memory. When the processor runs the computer program stored in the memory, the processor executes a method for calculating corn emergence rate based on unmanned aerial vehicle remote sensing images and machine learning according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of a method for calculating corn emergence rate based on unmanned aerial vehicle remote sensing images and machine learning as described in any one of claims 1 to 7.

Images