Farmland status monitoring method based on farmland boundary recognition

Abstract

The present invention discloses a method for monitoring the state of cultivated land based on cultivated land boundary identification, which relates to the field of image analysis technology. The technical solution comprises the following steps: preprocessing periodically acquired real-time images of cultivated land, screening to obtain a grayscale image corresponding to a set optimal spectral band; calculating the comprehensive probability that each local peak pixel point in the current periodic grayscale image represents the presence of seedlings based on the spatial distribution characteristics and temporal difference characteristics of the grayscale image and the seedlings; judging whether each local peak pixel point represents a seedling based on the comprehensive probability; dividing the cultivated land boundary based on the judgment result, and distinguishing between cultivated areas and uncultivated areas; the present invention utilizes the spatial distribution characteristics and temporal difference characteristics of the seedlings to judge the probability of the seedlings' presence, thereby improving the accuracy of cultivated land boundary identification and performing periodic real-time update monitoring of the cultivated land state.

Description

Technical Field

[0001] The present invention relates to the technical field of image analysis, and in particular to a method for monitoring farmland status based on farmland boundary recognition. Background Art

[0002] Cultivated land boundary identification is a key technical tool for monitoring cultivated land status. It is crucial for precision agriculture, cultivated land resource management, cultivated land protection redline control, intelligent digital agriculture, and addressing climate change and the impacts of human activities. It provides strong support for refined management, dynamic monitoring, ecological protection, and policy optimization of cultivated land resources, and is a key technology for achieving sustainable agricultural development and ensuring national food security. Cultivated land boundary identification can be determined by capturing the location of seedlings during the budding phase. Furthermore, the budding phase is the initial stage of crop growth and development, making it a critical time for remedial measures to be taken in uncultivated areas. It is not too late to take remedial measures in uncultivated areas during the budding phase to ensure food production.

[0003] However, due to the small size of seedling leaves in the early stages of growth, some seedlings are difficult to accurately identify and locate in the images captured by the hyperspectral camera, affecting the accuracy of seedling location information. Furthermore, due to differences in soil fertility, moisture, and other conditions within the same farmland, seedling growth patterns can be inconsistent, causing some seedlings to germinate later than others. This makes it difficult to accurately determine cultivated and non-cultivated areas when determining farmland boundaries based on seedling location information. Summary of the Invention

[0004] In order to improve the accuracy of farmland boundary recognition and effectively monitor the state of farmland, and to solve the problems of difficulty in accurately identifying and locating some seedlings in the early stages of growth, as well as inconsistent growth patterns of seedlings, which makes it difficult to accurately determine cultivated and non-cultivated areas when determining farmland boundaries based on seedling position information, the present invention provides a farmland state monitoring method based on farmland boundary recognition, the technical solution of which is as follows:

[0005] A method for monitoring cultivated land status based on cultivated land boundary recognition includes the following steps: preprocessing periodically acquired real-time images of cultivated land to obtain a grayscale image corresponding to a set optimal spectral band; calculating the comprehensive probability that each local peak pixel in the current periodic grayscale image represents the presence of a seedling based on the grayscale image and the spatial distribution characteristics and temporal difference characteristics of the seedlings; determining whether each local peak pixel represents a seedling based on the comprehensive probability; and demarcating the cultivated land boundary based on the determination result to distinguish between cultivated and uncultivated areas.

[0006] Among them, the comprehensive probability that the i-th local peak pixel in the grayscale image of the nth week represents the existence of seedlings is , the calculation formula is:

[0007]

[0008] Where, Represents the weight parameter corresponding to the i-th local peak pixel in the grayscale image of the n-th cycle, Indicates that the i-th local peak pixel in the grayscale image of the n-th cycle represents the spatial probability of the existence of seedlings, The i-th local peak pixel in the grayscale image of the n-th cycle represents the temporal probability of the existence of the seedling.

[0009] Preferably, the optimal spectral band is a fixed band or an adaptive band, and the method for determining the adaptive band includes: obtaining hyperspectral images of cultivated land seedlings during the germination and growth period in previous years as historical data, manually marking several areas where seedlings exist based on the historical hyperspectral images, extracting pixel data of seedlings in each marked area based on the imaging principle of hyperspectral images, obtaining the reflectance values ​​corresponding to the seedling pixels, and counting the average reflectance of the seedling pixels under each spectral band of the hyperspectral image imaging, and taking the spectral band corresponding to the maximum reflectance average as the optimal spectral band for seedling detection.

[0010] Preferably, a drone equipped with a hyperspectral camera is used to periodically collect hyperspectral image data of cultivated land; based on the optimal spectral band, all pixel data of the image layer corresponding to the optimal spectral band and the reflectance values ​​corresponding to the pixels are extracted; the obtained reflectance value is divided by the maximum value of the reflectance value to achieve data normalization, the normalized data is multiplied by 255 and rounded down to achieve grayscale conversion, and a grayscale image corresponding to the optimal spectral band is obtained.

[0011] Preferably, local peak extraction processing is performed on the grayscale image corresponding to the optimal spectral band to obtain the position information and corresponding grayscale value information of all local peak pixels on the current period grayscale image. Based on the position coordinates of the 4 neighborhood local peak pixels corresponding to each local peak pixel, the Euclidean distance between each local peak pixel and the corresponding 4 neighborhood local peak pixel is calculated, and the mean is calculated to obtain the average peak distance corresponding to each local peak pixel. The average peak distance corresponding to each local peak pixel is accumulated and divided by the total number of local peak pixels to obtain the mean peak distance on the current period grayscale image.

[0012] Preferably, the ratio of the average peak spacing corresponding to each local peak pixel point in the current period to the mean of the peak spacing on the grayscale image is used as the arrangement normalization value of the position coordinates of each local peak pixel point; the normalization value of the average arrangement of all local peak pixel points is set to 1, and the difference between the arrangement normalization value of the position coordinates of each local peak pixel point and 1 is calculated, and the absolute value is taken as the degree of deviation between the position arrangement of each local peak pixel point and the average arrangement of the local peak pixel points; the natural exponential function is used to reversely map the degree of position deviation of each local peak pixel point, and the mapping value is taken as the spatial probability that each local peak pixel point on the grayscale image corresponding to the current period represents the existence of seedlings.

[0013] Preferably, the position coordinates of each local peak pixel point on the grayscale image of the current cycle and the cycle before the current cycle are aligned, and the single-type region growing method is used to obtain the seedling area corresponding to each local peak pixel point, and the number of pixels in the seedling area is counted; starting from the second cycle, the shape matching method is used to calculate the shape similarity between all seedling areas of the current cycle and the corresponding seedling areas of the cycle before the current cycle, and the difference in the number of pixels between all corresponding seedling areas is calculated; based on one of all seedling areas in the current cycle, the shape similarity from the first cycle to the current cycle is accumulated and the difference in the number of pixels is calculated. The product of the difference in the number of pixels is used as the similarity growth coefficient of the current cycle. The difference in the number of pixels from the first cycle to the current cycle is added and a non-zero constant r is added to serve as the overall growth coefficient of the current cycle. The ratio of the similarity growth coefficient to the overall growth coefficient of the seedling area corresponding to the current cycle is used as the temporal probability that the local peak pixel corresponding to the seedling area represents the existence of seedlings in the current cycle. If a local peak pixel in the current cycle does not exist in the grayscale image of the previous cycle, the temporal probability value corresponding to this local peak pixel in the current cycle is set to 0.

[0014] Preferably, the maximum and minimum values ​​of the number of pixels in all seedling areas on the grayscale image of the current period are obtained, the total difference between the maximum and minimum values ​​is calculated, the difference between the number of pixels in the seedling area corresponding to each local peak pixel and the minimum value is calculated, and the ratio of the difference to the total difference is used as the weight parameter corresponding to the local peak pixel; the weight parameter is used as the coefficient of the temporal probability of the corresponding local peak pixel, 1 minus the value of the weight parameter is used as the coefficient of the spatial probability of the corresponding local peak pixel, and the sum of the products of the spatial probability and temporal probability of the local peak pixel and the corresponding coefficients is used as the comprehensive probability that the local peak pixel represents the existence of the seedling.

[0015] Preferably, a judgment threshold F is set, and the comprehensive probability corresponding to each local peak pixel point on the current period grayscale image is compared with the judgment threshold. When the comprehensive probability is greater than or equal to the judgment threshold F, it is determined that there is a seedling at the position coordinate point corresponding to the local peak pixel point. When the comprehensive probability is less than the judgment threshold F, it is determined that there is no seedling at the position coordinate point corresponding to the local peak pixel point.

[0016] Preferably, the position coordinates corresponding to all local peak pixel points determined to have seedlings on the grayscale image of the current period are extracted, and the mean shift clustering method is used to divide several areas based on the density of local peak pixel points representing seedlings. Then, the axis-aligned bounding box method is used to divide the cultivated land boundary of the current period, and the area within the axis-aligned bounding box is divided into the cultivated area, and the area outside the axis-aligned bounding box is divided into the uncultivated area.

[0017] Preferably, the sowing period is used as the starting timing moment, and after the set period time t, the hyperspectral image data of the first period is collected, the corresponding grayscale image is obtained, and the spatial probability is calculated. The temporal probability cannot be calculated for the first period, and the weight parameters corresponding to all local peak pixels of the first period are set to 0. The value of the comprehensive probability of the first period is equal to the value of the corresponding spatial probability, thereby obtaining the cultivated land boundary and cultivated land status of the first period; after another set period time t, the hyperspectral image data of the second period is collected, and the corresponding grayscale image is obtained. Starting from the second period, the spatial probability and temporal probability that each local peak pixel on the grayscale image represents the existence of seedlings are calculated, and then the corresponding comprehensive probability is calculated to obtain the cultivated land boundary and cultivated land status of the second period; similarly, the cultivated land boundary and cultivated land status of the next period are obtained in sequence, and the cultivated land boundary and cultivated land status information is periodically updated.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] 1. This method, based on farmland boundary identification, uses the spatial distribution and temporal variation characteristics of seedlings in the germination period to determine the probability of seedling presence. It also performs a secondary assessment of late-germinating seedlings. The method then combines the positional information of seedlings in the germination period to determine the farmland boundary. This improves the accuracy of farmland boundary identification and enables periodic, real-time monitoring of farmland status.

[0020] 2. This method, based on farmland boundary identification, determines farmland boundaries using the positional information of seedlings during the germination period. It then periodically updates and monitors farmland status in real time. This method effectively assesses farmland status in the early stages of crop growth, facilitating timely remedial measures for uncultivated areas and ensuring food production. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1The flowchart for the implementation of the cultivated land status monitoring method is as follows;

[0022] Figure 2 Schematic diagram for finding and determining the local peak pixel points in the 4-neighborhood; DETAILED DESCRIPTION

[0023] The technical features of the present invention are further described in detail below with reference to the accompanying drawings so that those skilled in the art can understand them.

[0024] A method for monitoring farmland status based on farmland boundary identification, the implementation process is as follows Figure 1 The specific implementation steps are as follows:

[0025] Step S1: based on historical data of the germination growth period of cultivated seedlings, obtaining the optimal spectral band for seedling detection;

[0026] Specifically, we obtain hyperspectral images of the germination and growth period of cultivated land seedlings in previous years as historical data. Based on the historical hyperspectral images, we manually mark several areas where seedlings exist. Since hyperspectral images simultaneously image the target area with dozens to hundreds of continuous and subdivided spectral bands, and based on the imaging principle of hyperspectral images, the reflectivity of seedlings at different imaging wavelengths is different, the reflectivity of each pixel point at different imaging wavelengths can be obtained, and the pixel point data of seedlings in each marked area can be extracted to obtain the reflectivity value corresponding to the seedling pixel point, which is statistically recorded in the hyperspectral image. For example, under each spectral band of imaging, the average reflectance of the seedling pixels is taken, and the spectral band corresponding to the maximum reflectance average is taken as the optimal spectral band for seedling detection; in agriculture, the band range for detecting crop physiological indices based on spectral technology is generally between 400nm and 2500nm. Among them, in the near-infrared band of 740nm to 1300nm, due to the reflection of healthy mesophyll cells, the reflectance of the detection image increases sharply. Therefore, for seedlings in the germination growth period, the value range of the optimal spectral band is generally 740nm to 1300nm.

[0027] In addition, since the mean reflectivity obtained when detecting seedlings in the optimal spectral band is the largest and the imaging effect is the best, and the optimal imaging bands corresponding to different varieties of seedlings in the hyperspectral image are consistent, that is, there is no need to consider the plant varieties planted on the cultivated land, and the optimal spectral bands obtained are the same; therefore, it is only necessary to analyze the image data corresponding to the optimal spectral band, which can reduce the amount of data in the image processing process and improve the program running speed without affecting the image analysis effect.

[0028] Step S2: Periodically obtain hyperspectral images of cultivated land and perform grayscale processing to obtain a grayscale image corresponding to the optimal spectral band;

[0029] Specifically, a drone equipped with a hyperspectral camera is used to periodically collect hyperspectral image data of cultivated land. Based on the optimal spectral band, all pixel data of the image layer corresponding to the optimal spectral band and the reflectance values ​​corresponding to the pixels are extracted. The obtained reflectance value is divided by the maximum value of the reflectance value to achieve data normalization, and the normalized data is multiplied by 255 and rounded down to achieve grayscale conversion, thus obtaining a grayscale image corresponding to the optimal spectral band.

[0030] In addition, the total time period for collecting hyperspectral image data of cultivated land is set to T, the interval time of image collection cycle is set to t, the total number of image collections is set to N, and the grayscale image corresponding to the nth image collection cycle is set to , In order to avoid missing the image collection of seedlings that are sown late and germinated late, the total period T needs to exceed the germination growth period of the seedlings. Usually, the sowing period is used as the starting time, the total period T is set to 14 days, and the interval time t of the image collection cycle is set to 6 hours.

[0031] Step S3: Based on the grayscale image and the spatial distribution characteristics and temporal difference characteristics of the seedlings, the comprehensive probability that each local peak pixel in the grayscale image of the current period represents the presence of the seedling is calculated;

[0032] When the seedlings sprout and break through the soil, the reflectivity value of the corresponding position on the image increases, and the grayscale value in the grayscale image becomes higher, while the reflectivity of the land part does not change much. There is a grayscale value difference between the seedlings and the cultivated land, which is manifested as the seedlings being brighter than the cultivated land on the grayscale image. Due to the grayscale value difference between the seedlings and the cultivated land, the local area peak pixel point on the grayscale image can approximately represent the position of the seedling. If the coordinate position of the local area peak pixel point does not exist on the grayscale image, it can be judged that there is no seedling at the corresponding coordinate position in the current cycle. Therefore, it is only necessary to perform image analysis on the coordinate position of the local area peak pixel point. According to the spatial distribution characteristics and temporal difference characteristics of the seedling growth on the cultivated land, excluding environmental interference such as weeds, it is judged whether the plant represented by the local area peak pixel point is a seedling, and then the cultivated land boundary is divided and the cultivated land status is determined according to the coordinate position distribution corresponding to the local area peak pixel point representing the seedling.

[0033] Regarding the spatial distribution characteristics of seedling growth on cultivated land, seeds are sown at regular intervals. Therefore, the seedlings are regularly and evenly spaced in the horizontal and vertical directions. This equidistant arrangement of seedlings on cultivated land can be used to determine the probability that a local peak pixel represents a seedling.

[0034] Regarding the temporal difference characteristics of seedling growth on cultivated land, seedlings in the germination growth stage gradually grow over time, which is manifested on the grayscale image as the number of pixels contained in the seedlings gradually increasing and having similar shapes at the same time interval; thus, the probability that the local peak pixel represents the seedling can be judged by the temporal difference based on the growth characteristics of the seedling area in the germination growth stage gradually increasing and having similar shapes.

[0035] Specifically, by performing local peak extraction processing on the grayscale image corresponding to the optimal spectral band, the position information and corresponding grayscale value information of all local peak pixels on the current period grayscale image are obtained, and the pixel point is a three-dimensional information, including the horizontal and vertical coordinates and grayscale value of the pixel point in the image; based on the position coordinates of the four neighboring local peak pixels corresponding to each local peak pixel point, the Euclidean distance between each local peak pixel point and the corresponding four neighboring local peak pixels is calculated, and the average value is calculated to obtain the average peak distance corresponding to each local peak pixel point, and the average peak distance corresponding to each local peak pixel point is accumulated and divided by the total number of local peak pixels to obtain the average peak distance on the current period grayscale image;

[0036] Among them, get the i-th local peak pixel point in the grayscale image of the n-th cycle The corresponding 4 neighborhood local peak pixel points include: first find the distance according to the known position coordinate information corresponding to each local peak pixel point The nearest local peak pixel, marked as point p1, is connected and point p1 and extend the connecting line, marking the extended connecting line as L1. Draw a perpendicular line L2 to L1 as the center; set the line width LK with L1 and L2 as the center lines respectively, and use the cross-shaped area covered by the line width as the search area. The corresponding range of the 4-neighborhood local peak pixel points is as follows: Figure 2 As shown; the line width LK can be set according to the actual application scenario and requirements. The solution of the present invention sets the line width LK to a spacing width of 10 pixels, that is, the line width on one side of the center line is a spacing width of 5 pixels;

[0037] For the cross-shaped area covered by the line width, find the distance in the other direction relative to point p1 within the area of ​​L1 The nearest local peak pixel point is recorded as point p2; find the correct pixel in the L2 area. Distance on both sides The nearest local peak pixel points are recorded as points p3 and p4; points p1, p2, p3, and p4 are The corresponding 4 neighborhood local peak pixel points; determine After the corresponding 4 neighborhood local peak pixel points are obtained, the position coordinates can be calculated based on them. The corresponding average peak spacing .

[0038] Among them, the total number of local peak pixels in the grayscale image of the nth cycle is , the i-th local peak pixel in the grayscale image of the n-th cycle is , the mean peak spacing on the n-th cycle grayscale image is , the calculation formula is:

[0039]

[0040] Where, Represents the i-th local peak pixel in the grayscale image of the n-th cycle The corresponding average peak spacing, .

[0041] In addition, the ratio of the average peak spacing corresponding to each local peak pixel point in the current cycle to the mean peak spacing on the grayscale image is used as the arrangement normalization value of the position coordinates of each local peak pixel point; the normalization value of the average arrangement of all local peak pixels is set to 1, and the difference between the arrangement normalization value of the position coordinates of each local peak pixel point and 1 is calculated, and the absolute value is taken as the degree of deviation between the position arrangement of each local peak pixel point and the average arrangement of local peak pixels; the natural exponential function is used to reversely map the position deviation degree of each local peak pixel point, and the mapping value is taken as the spatial probability that each local peak pixel point on the grayscale image of the current cycle represents the existence of a seedling;

[0042] Among them, the i-th local peak pixel in the grayscale image of the n-th cycle The spatial probability representing the existence of seedlings is , the calculation formula is:

[0043]

[0044] Where, represents the natural exponential function, Indicates the i-th local peak pixel in the grayscale image of the n-th cycle The arrangement normalized value of the position coordinates, ; Indicates the i-th local peak pixel on the grayscale image in the n-th cycle The degree of deviation between the position arrangement and the average arrangement of local peak pixels, , The closer the value of is to 0, the more it indicates that the i-th local peak pixel point on the grayscale image of the n-th cycle The closer the position coordinates are to the standard coordinates of the average arrangement of local peak pixels, the smaller the deviation is. , The closer the value is to 1, the more the pixel point is at the local peak in the nth cycle. The greater the probability that there is a seedling at the corresponding coordinate position, that is, the greater the probability that the pixel information at the corresponding coordinate position in the image corresponds to the seedling image information.

[0045] Specifically, align the position coordinates of each local peak pixel point on the grayscale image of the current cycle and the previous cycle of the current cycle, adopt a single-type region growing method to obtain the seedling area corresponding to each local peak pixel point, and count the number of pixels in the seedling area; because the seedlings will gradually grow larger with the passage of time when they are in the germination growth period, and their shapes remain basically unchanged, therefore, it is necessary to segment and determine the seedling area corresponding to each local peak pixel point, so as to compare and judge the shape similarity, and at the same time count the number of pixels in the seedling area to judge the growth of the seedling; for the segmentation and determination of the seedling area, generally, the threshold segmentation method, edge extraction method, region growing method, watershed algorithm and other methods can be adopted for determination. The present invention adopts a single-type region growing method to determine the seedling area corresponding to each local peak pixel point;

[0046] The specific steps of the single-type region growing method are as follows: taking the local peak pixel point as the seed point position, based on the similarity criterion in the single-type region growing method, setting the gray value difference threshold to f, and based on the termination criterion of the single-type region growing method, setting the growth to stop when there is no pixel point with a gray value difference less than or equal to f in the 8-neighborhood, or when the growing area exceeds the specified range, wherein the local peak pixel point is set as the center and the side length is set to stop growing. The square area is taken as the specified range; based on the growth criterion in the single-type region growing method, the pixels with grayscale value differences less than or equal to f within the 8-neighborhood range corresponding to the seed point position are merged, and the 8-neighborhood range corresponding to the merged pixel continues to grow outward, and the pixels with grayscale value differences less than or equal to f within the 8-neighborhood range corresponding to the merged pixel are merged again until the termination criterion is triggered; then the seedling area corresponding to each local peak pixel is obtained, and the number of pixels in the seedling area corresponding to each local peak pixel is obtained by counting;

[0047] Starting from the second cycle, the shape matching method is used to calculate the shape similarity between all seedling areas of the current cycle and the corresponding seedling areas of the previous cycle of the current cycle, and the difference in the number of pixels between all corresponding seedling areas is calculated; based on one of all seedling areas of the current cycle, the product of the shape similarity and the difference in the number of pixels from the first cycle to the current cycle is accumulated as the similarity growth coefficient of the current cycle, and the difference in the number of pixels from the first cycle to the current cycle is added with a constant r approaching 0 as the overall growth coefficient of the current cycle; the ratio of the similarity growth coefficient of the seedling area corresponding to the current cycle to the overall growth coefficient is used as the temporal probability that the local peak pixel corresponding to the seedling area represents the existence of the seedling in the current cycle; for a local peak pixel in the current cycle that does not exist in the grayscale image of the previous cycle of the current cycle, such as a newly sprouted plant, since the corresponding shape similarity cannot be obtained, the temporal probability value corresponding to this local peak pixel in the current cycle is set to 0, and the temporal probability value corresponding to this local peak pixel is specifically calculated in the next cycle;

[0048] Among them, the i-th local peak pixel in the grayscale image of the n-th cycle The corresponding seedling area is , the number of pixels is , and The shape similarity between , and The difference between , ;

[0049] The i-th local peak pixel in the grayscale image of the n-th cycle The temporal probability representing the existence of seedlings is , the calculation formula is:

[0050]

[0051] Where, Represents the i-th local peak pixel in the grayscale image of the n-th cycle The corresponding similarity growth coefficient is calculated as follows:

[0052]

[0053] Represents the i-th local peak pixel in the grayscale image of the n-th cycle The corresponding overall growth coefficient is calculated as follows:

[0054]

[0055] In order to avoid the overall growth coefficient as the denominator having a value of 0, which affects the calculation process, and to avoid the calculation result deviating too much from the actual value and having a large error when the normal value is calculated, it is necessary to add a constant r that approaches 0. The value of r can be 0.01.

[0056] In addition, based on the spatial probability and the temporal probability, the comprehensive probability that each local peak pixel on the current period grayscale image represents the existence of the seedling is calculated, including obtaining the maximum and minimum values ​​of the number of pixels in all seedling areas on the current period grayscale image, calculating the total difference between the maximum and minimum values, calculating the difference between the number of pixels in the seedling area corresponding to each local peak pixel and the minimum value, and taking the ratio of the difference to the total difference as the weight parameter corresponding to the local peak pixel; taking the weight parameter as the coefficient of the temporal probability of the corresponding local peak pixel, taking 1 minus the value of the weight parameter as the coefficient of the spatial probability of the corresponding local peak pixel, and taking the sum of the products of the spatial probability and temporal probability of the local peak pixel and the corresponding coefficients as the comprehensive probability that the local peak pixel represents the existence of the seedling;

[0057] In the early stage of the germination growth period, the seedling area is small and contains fewer pixels. The number of pixels does not change much over time, and the value of the weight parameter is small. At this time, spatial probability plays a dominant role in the judgment of seedling probability. As the seedling grows gradually, its area becomes larger and larger, and the value of the weight parameter is adjusted accordingly. At this time, temporal probability plays a dominant role in the judgment of seedling probability.

[0058] Among them, the comprehensive probability that the i-th local peak pixel in the grayscale image of the nth week represents the existence of seedlings is , the calculation formula is:

[0059]

[0060] Where, The weight parameter corresponding to the i-th local peak pixel in the grayscale image of the n-th cycle is calculated as follows:

[0061]

[0062] Indicates the maximum number of pixels in all seedling areas on the grayscale image of the nth cycle, Indicates the minimum number of pixels in all seedling areas on the grayscale image of the nth cycle.

[0063] Step S4: Based on the comprehensive probability, determine whether each local peak pixel in the grayscale image of the current period represents a seedling;

[0064] Specifically, a judgment threshold F is set, and the comprehensive probability corresponding to each local peak pixel point on the current period grayscale image is compared with the judgment threshold. When the comprehensive probability is greater than or equal to the judgment threshold F, it is determined that there are seedlings at the position coordinate point corresponding to the local peak pixel point. When the comprehensive probability is less than the judgment threshold F, it is determined that there are no seedlings at the position coordinate point corresponding to the local peak pixel point. The present invention sets the value of the judgment threshold F to 0.7, and the specific value of the judgment threshold F can be adjusted according to actual application scenarios and needs.

[0065] Step S5: Based on the determination result, the cultivated land boundary is divided to distinguish the cultivated area from the uncultivated area;

[0066] Specifically, the position coordinates corresponding to all local peak pixel points determined to have seedlings on the grayscale image of the current period are extracted, and the mean shift clustering method is used to divide several areas based on the density of local peak pixel points representing seedlings. The axis-aligned bounding box method is then used to divide the cultivated land boundary of the current period, and the maximum and minimum values ​​of the horizontal coordinates and the maximum and minimum values ​​of the vertical coordinates of each area are obtained to determine the coordinates of the corner points of the bounding box, and then the cultivated land boundary can be determined by the bounding box; the area within the axis-aligned bounding box is divided into the cultivated area, and the area outside the axis-aligned bounding box is divided into the uncultivated area.

[0067] Step S6: periodically updating cultivated land boundary and cultivated land status information based on the periodically acquired hyperspectral images;

[0068] Specifically, taking the sowing period as the starting timing moment, after the set cycle time t, the hyperspectral image data of the first cycle is collected, the grayscale image corresponding to the first cycle is obtained, and the spatial probability that each local peak pixel on the grayscale image of the first cycle represents the existence of seedlings is calculated. The temporal probability cannot be calculated for the first cycle, and the weight parameters corresponding to all local peak pixels of the first cycle are set to 0. The value of the comprehensive probability corresponding to all local peak pixels of the first cycle is equal to the value of the spatial probability, thereby obtaining the cultivated land boundary and cultivated land status of the first cycle; after another set cycle time t, the hyperspectral image data of the second cycle is collected, and the grayscale image corresponding to the second cycle is obtained. Starting from the second cycle, the spatial probability and temporal probability that each local peak pixel on the grayscale image represents the existence of seedlings are calculated, and then the corresponding comprehensive probability is calculated to obtain the cultivated land boundary and cultivated land status of the second cycle; similarly, the cultivated land boundary and cultivated land status of the next cycle are obtained in turn, and the cultivated land boundary and cultivated land status information is periodically updated.

[0069] In addition, based on the periodically updated cultivated land boundary information and cultivated land status information, remedial measures can be taken in a timely manner for cultivated land classified as uncultivated areas, and the cultivated land areas where remedial measures are taken can be marked. In subsequent cycle detections, it can be determined whether seedlings have grown to determine whether the remediation is successful.

[0070] The embodiments described in the present invention are merely descriptions of preferred implementations of the present invention and are not limited to the precise structures described above and shown in the accompanying drawings. Various modifications and changes can be made without departing from the scope of protection thereof. Without departing from the design concept of the present invention, various variations and improvements made to the technical solutions of the present invention by engineers and technicians in this field should fall within the scope of protection of the present invention.

Claims

1. A method for monitoring farmland status based on farmland boundary recognition, characterized by: The periodically acquired real-time images of cultivated land are preprocessed to obtain a grayscale image corresponding to the set optimal spectral band. Based on the grayscale image and the spatial distribution characteristics and temporal difference characteristics of the seedlings, the comprehensive probability that each local peak pixel point in the current period grayscale image represents the presence of a seedling is calculated. Based on the comprehensive probability, it is determined whether each local peak pixel represents a seedling; Based on the determination results, the boundaries of cultivated land are divided to distinguish between cultivated areas and uncultivated areas; Among them, by performing local peak extraction processing on the grayscale image corresponding to the optimal spectral band, the position information and corresponding grayscale value information of all local peak pixels on the current period grayscale image are obtained, based on the position coordinates of the four neighboring local peak pixels corresponding to each local peak pixel, the Euclidean distance between each local peak pixel and the corresponding four neighboring local peak pixels is calculated, and the average value is calculated to obtain the average peak distance corresponding to each local peak pixel, the average peak distance corresponding to each local peak pixel is accumulated and divided by the total number of local peak pixels to obtain the average peak distance on the current period grayscale image; The ratio of the average peak spacing corresponding to each local peak pixel point in the current period to the mean peak spacing on the grayscale image is used as the arrangement normalization value of the position coordinates of each local peak pixel point; the normalization value of the average arrangement of all local peak pixel points is set to 1, and the difference between the arrangement normalization value of the position coordinates of each local peak pixel point and 1 is calculated, and the absolute value is taken as the degree of deviation between the position arrangement of each local peak pixel point and the average arrangement of the local peak pixel points; the natural exponential function is used to reversely map the position deviation degree of each local peak pixel point, and the mapping value is taken as the spatial probability that each local peak pixel point on the grayscale image of the current period represents the existence of a seedling; The comprehensive probability that the i-th local peak pixel in the grayscale image of the n-th week represents the existence of seedlings is , the calculation formula is: Where, Represents the weight parameter corresponding to the i-th local peak pixel in the grayscale image of the n-th cycle, Indicates that the i-th local peak pixel in the grayscale image of the n-th cycle represents the spatial probability of the existence of seedlings, The i-th local peak pixel in the grayscale image of the n-th cycle represents the temporal probability of the existence of the seedling.

2. The method for monitoring cultivated land status according to claim 1, characterized in that: The optimal spectral band is a fixed band or an adaptive band. The method for determining the adaptive band includes: obtaining hyperspectral images of cultivated land seedlings during the germination and growth period in previous years as historical data, manually marking several areas where seedlings exist based on the historical hyperspectral images, extracting pixel point data of seedlings in each marked area based on the imaging principle of hyperspectral images, obtaining the reflectance values ​​corresponding to the seedling pixels, and calculating the average reflectance of the seedling pixels under the spectral band of each imaging segment of the hyperspectral image, and taking the spectral band corresponding to the maximum reflectance average as the optimal spectral band for seedling detection.

3. The method for monitoring cultivated land status according to claim 1, wherein: The method preprocesses the periodically acquired real-time images of cultivated land to obtain a grayscale image corresponding to a set optimal spectral band, including: using an unmanned aerial vehicle equipped with a hyperspectral camera to periodically collect hyperspectral image data of the cultivated land; based on the optimal spectral band, extracting all pixel point data of the image layer corresponding to the optimal spectral band and the reflectance values ​​corresponding to the pixels; dividing the acquired reflectance value by the maximum value of the reflectance value to achieve data normalization, multiplying the normalized data by 255 and rounding down to achieve grayscale conversion, and obtaining a grayscale image corresponding to the optimal spectral band.

4. The method for monitoring cultivated land status according to claim 1, wherein: Based on the position information of the local peak pixel points, the temporal probability that each local peak pixel point on the grayscale image of the current cycle represents the existence of the seedling is calculated, including: aligning the position coordinates of each local peak pixel point on the grayscale image of the current cycle and the previous cycle of the current cycle, using a single type region growing method to obtain the seedling area corresponding to each local peak pixel point, and counting the number of pixels in the seedling area; starting from the second cycle, using a shape matching method to calculate the shape similarity between all seedling areas of the current cycle and the corresponding seedling areas of the previous cycle of the current cycle, and calculating the difference in the number of pixels between all corresponding seedling areas; based on the number of all seedling areas in the current cycle One of them, accumulate the product of the shape similarity and the difference in the number of pixels from the first cycle to the current cycle as the similarity growth coefficient of the current cycle, accumulate the difference in the number of pixels from the first cycle to the current cycle and add a non-zero constant r as the overall growth coefficient of the current cycle; take the ratio of the similarity growth coefficient to the overall growth coefficient of the seedling area corresponding to the current cycle as the temporal probability that the local peak pixel corresponding to the seedling area represents the existence of the seedling in the current cycle; for a local peak pixel in the current cycle that does not exist in the grayscale image of the previous cycle of the current cycle, the temporal probability value corresponding to this local peak pixel in the current cycle is set to 0.

5. The method for monitoring cultivated land status according to claim 4, characterized in that: Based on the spatial probability and the temporal probability, the comprehensive probability that each local peak pixel point on the current period grayscale image represents the existence of a seedling is calculated, including: obtaining the maximum and minimum values ​​of the number of pixels in all seedling areas on the current period grayscale image, calculating the total difference between the maximum and minimum values, calculating the difference between the number of pixels in the seedling area corresponding to each local peak pixel point and the minimum value, and taking the ratio between the difference and the total difference as the weight parameter corresponding to the local peak pixel point; taking the weight parameter as the coefficient of the temporal probability of the corresponding local peak pixel point, taking 1 minus the value of the weight parameter as the coefficient of the spatial probability of the corresponding local peak pixel point, and taking the sum of the products of the spatial probability and the temporal probability of the local peak pixel point and the corresponding coefficients as the comprehensive probability that the local peak pixel point represents the existence of the seedling.

6. The method for monitoring cultivated land status according to claim 1 or 5, characterized in that: Based on the comprehensive probability, it is determined whether each local peak pixel point represents a seedling, including: setting a judgment threshold F, comparing the comprehensive probability corresponding to each local peak pixel point on the current period grayscale image with the judgment threshold, when the comprehensive probability is greater than or equal to the judgment threshold F, it is determined that there is a seedling at the position coordinate point corresponding to the local peak pixel point; when the comprehensive probability is less than the judgment threshold F, it is determined that there is no seedling at the position coordinate point corresponding to the local peak pixel point.

7. The method for monitoring cultivated land status according to claim 6, characterized in that: Based on the determination result of whether there are seedlings at each coordinate point, the cultivated land boundary is divided and the cultivated area and the uncultivated area are distinguished, including: extracting the position coordinates corresponding to all local peak pixel points determined to have seedlings on the grayscale image of the current period, and using the mean shift clustering method to divide several areas based on the density of local peak pixel points representing seedlings, and then using the axis-aligned bounding box method to divide the cultivated land boundary of the current period, dividing the area within the axis-aligned bounding box range into the cultivated area, and dividing the area outside the axis-aligned bounding box range into the uncultivated area.

8. The method for monitoring cultivated land status according to claim 1, wherein: Based on the periodically acquired real-time images of cultivated land, the cultivated land boundary and cultivated land status information are periodically updated, including: taking the sowing period as the starting timing moment, after the interval set period time t, collecting the hyperspectral image data of the first period, obtaining the corresponding grayscale image, calculating the spatial probability, the temporal probability cannot be calculated in the first period, setting the weight parameters corresponding to all local peak pixel points of the first period to 0, and the value of the comprehensive probability of the first period is equal to the value of the corresponding spatial probability, thereby obtaining the cultivated land boundary and cultivated land status of the first period; after the interval set period time t again, collecting the hyperspectral image data of the second period, obtaining the corresponding grayscale image, starting from the second period, calculating the spatial probability and temporal probability that each local peak pixel point on the grayscale image represents the existence of seedlings, and then calculating the corresponding comprehensive probability, and obtaining the cultivated land boundary and cultivated land status of the second period; similarly, obtaining the cultivated land boundary and cultivated land status of the next period in turn, and periodically updating the cultivated land boundary and cultivated land status information.

Images