A soybean harvesting boundary detection method and system in a maize-soybean intercropping mode

Abstract

The application provides a soybean harvesting boundary detection method and system in a corn-soybean intercropping mode, comprising the following steps: acquiring an image in front of a harvester collected by a camera, and performing semantic segmentation on the image, with different color mask images representing different categories of regions; performing color region extraction on the mask image to obtain an RGB image; image preprocessing; converting each pixel point on the image plane after preprocessing into a corresponding pixel point on a world plane, and then into a corresponding pixel point after eliminating perspective distortion; extracting an upper boundary line of a header region, and setting a dynamic ROI according to the upper boundary line of the header region; extracting a soybean region RGB image; extracting left and right boundary lines of the soybean region, and fitting them into straight lines to obtain left and right soybean harvesting boundaries. The application can accurately detect the left and right boundaries of soybeans at the same time, avoid the influence of corn plants on both sides of the soybeans on the observation of the harvesting boundary by the machine hand, causing missed cutting, machine collision damage to the corn plants, and high operation intensity, and achieve better soybean harvesting.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of visual intelligent soybean harvesting boundary detection, and particularly relates to a soybean harvesting boundary detection method and system in a maize-soybean intercropping mode. BACKGROUND

[0002] Maize and soybean are important grain and oil crops related to national food and food safety. Maize-soybean strip intercropping technology can make full use of resources and improve land productivity, and has been selected as a national technology for many consecutive years. When soybeans are harvested first in a maize-soybean intercropping mode, the maize plants on both sides of the soybeans can easily block the view of the harvester operator. In order to ensure that the harvester does not mistakenly harvest maize when harvesting soybeans, the operator needs to have good driving skills, and the long working hours are strenuous and tiring, which cannot guarantee the harvesting efficiency. At present, intelligent agricultural machinery is gradually appearing on the historical stage, which can replace manual harvesting and improve harvesting efficiency while achieving precise operation. Therefore, harvesting boundary detection is particularly important for further meeting the intelligent harvesting needs of the harmonious coexistence planting mode of crops in a maize-soybean intercropping mode.

[0003] At present, there are few reports on soybean harvesting boundary detection in a maize-soybean intercropping mode. Patent CN106508256B "Rice-wheat combined harvester and cutting width detection device and detection method thereof" and patent CN103914071B "Wide-narrow row regenerative rice harvesting regulation system and method based on binocular vision" use laser sensors and binocular cameras to obtain the harvesting boundary by comparing the height of the harvested and non-harvested areas of the crops. Patent CN109215071B "Intelligent rice-wheat harvester cutting width measurement method based on vision" uses a traditional image processing method based on morphology to extract the boundary lines of the harvested and non-harvested crops. The above methods mainly detect the harvesting boundary line by extracting the boundary between the harvested and non-harvested areas of the crops, and cannot distinguish between maize and soybean areas, making it difficult to accurately locate the soybean area boundary. SUMMARY

[0004] In view of the above technical problems, one of the purposes of one embodiment of the present application is to provide a soybean harvesting boundary detection method in a maize-soybean intercropping mode, which comprehensively applies semantic segmentation, image extraction, and dynamic ROI technology, can accurately extract the soybean harvesting boundary in real time, has high detection accuracy and fast detection speed, can solve the problem that the current crop boundary detection method is not suitable for the harvesting scene in a maize-soybean intercropping mode, cannot effectively distinguish between maize and soybean areas, and cannot accurately locate the soybean boundary, and can provide effective and reliable basic data for intelligent control of soybean harvesters in a maize-soybean intercropping mode.

[0005] To solve the above technical problems, one of the purposes of one embodiment of the present application is to provide a soybean harvesting boundary detection system in a corn-soybean intercropping mode, which comprises an image classification module, an image extraction module, an image preprocessing module, an image transformation module, a dynamic ROI module, and a boundary acquisition module. The system comprehensively applies semantic segmentation, image extraction, and dynamic ROI technology, can accurately extract the soybean harvesting boundary in real time, provides basic data for intelligent control of the soybean harvester in the corn-soybean intercropping mode, and realizes better soybean harvesting.

[0006] Note that the recitation of these objects does not preclude the presence of other objects. One embodiment of the present application does not need to achieve all the above-mentioned purposes. The purposes other than the above-mentioned purposes can be extracted from the recitations of the specification, drawings, and claims. The present application achieves the above technical purposes through the following technical means.

[0007] A soybean harvesting boundary detection method in a corn-soybean intercropping mode, comprising the following steps:

[0008] Step S1: image classification: acquiring an image in front of the harvester collected by a camera, and performing semantic segmentation on the image, and representing different categories of regions with different color mask images after pixel-level classification of the harvesting scene;

[0009] Step S2: image region extraction: color region extraction is performed on the mask image obtained in step S1 to obtain an RGB image containing only soybean regions and header regions;

[0010] Step S3: image preprocessing: performing image preprocessing operation on the RGB image obtained in step S2;

[0011] Step S4: image transformation: converting the image preprocessed in step S3 into corresponding pixel points on a two-dimensional plane image eliminating perspective distortion through inverse perspective mapping;

[0012] Step S5: setting dynamic ROI: extracting the upper boundary line of the header region in the two-dimensional plane image eliminating perspective distortion in step S4, and setting a dynamic ROI according to the upper boundary line of the header region;

[0013] Step S6: extracting soybean region RGB image: extracting the color corresponding to the soybean region mask in the ROI region in step S5 to obtain an RGB image containing only soybean regions;

[0014] Step S7: acquiring soybean region boundary: extracting the left and right boundary lines of the soybean region in the RGB image containing only soybean regions in step S6, and fitting the boundary lines into straight lines to obtain the left and right boundaries of the soybean harvesting in the corn-soybean intercropping mode.

[0015] In the scheme, the step S1 classifies the harvesting scene into soybean area, corn area, header area and background at pixel level, and attaches different color masks to each area respectively, and outputs the mask image.

[0016] In the scheme, the step S2 image area extraction only extracts the specified color area of the mask color corresponding to the soybean area and the header area, to obtain the RGB image containing the soybean area and the header area.

[0017] In the scheme, the specific operation of the step S3 image preprocessing is to first erode the image, and then dilate the eroded result.

[0018] In the scheme, the step S4 inverse perspective mapping is specifically: converting each pixel point on the image plane after the step S3 image preprocessing into a corresponding pixel point on the world plane, and then converting each pixel point on the world plane into a corresponding pixel point on the two-dimensional plane image eliminating perspective distortion; wherein the conversion of each pixel point on the image plane after the image preprocessing into a corresponding pixel point on the world plane adopts the following formula:

[0019]

[0020]

[0021] Where (x, y) is the coordinate of the pixel point on the world plane, (u, v) is the coordinate of each pixel point on the image plane after preprocessing, H is the height of the camera from the ground, alpha is half of the camera field of view angle, h represents the height of the image, w represents the width of the image, gamma0 represents the heading angle of the camera, and theta0 represents the pitch angle of the camera.

[0022] In the scheme, the step S4 converts each pixel point on the world plane into a corresponding pixel point on the two-dimensional plane image eliminating perspective distortion, which adopts the following formula:

[0023]

[0024]

[0025] Where (m, n) is the coordinate of the corresponding pixel point on the two-dimensional plane image eliminating perspective distortion.

[0026] In the scheme, the step S5 of setting a dynamic ROI includes the following steps:

[0027] Step S5.1: performing a top-to-bottom scanning process on the two-dimensional planar image described in step S4, when a pixel point with the same RGB value as the mask color corresponding to the header region described in step S1 is scanned, connecting the points with the same vertical coordinate j on the RGB graph to obtain the upper boundary line of the header region;

[0028] Step S5.2: taking the region with the vertical coordinate within (j-s, j) above the upper boundary line of the header region as the dynamic ROI, wherein s is the width of the dynamic ROI.

[0029] In the above scheme, the specific operation of step S6 to extract the RGB graph of the soybean region is to extract the pixel points with the corresponding RGB value in the ROI according to the RGB value of the mask color corresponding to the soybean region, to obtain the RGB graph in the ROI containing only the mask of the soybean region.

[0030] In the above scheme, the specific steps of step S7 to obtain the boundary of the soybean region include:

[0031] Step S7.1: performing a top-to-bottom scanning process on the RGB graph in the ROI containing only the soybean region, when a pixel point with the same RGB value as the mask color corresponding to the soybean region described in step S1 is scanned, saving the coordinate point corresponding to the pixel;

[0032] Step S7.2: arranging the coordinate points according to the points with the same vertical coordinate, selecting the points with the minimum and maximum horizontal coordinates as the points on the boundary line of the soybean region, connecting the pixel points with the minimum horizontal coordinates in all rows as the left boundary line of the soybean region, and connecting the pixel points with the maximum horizontal coordinates in all rows as the right boundary line of the soybean region;

[0033] Step S7.3: fitting the left and right boundary lines of the soybean region into straight lines by the least square method to obtain the left and right boundaries of the soybean harvesting in the corn-soybean intercropping mode.

[0034] A soybean harvesting boundary detection system in a corn-soybean intercropping mode, comprising an image classification module, an image extraction module, an image preprocessing module, an image transformation module, a dynamic ROI module, and a boundary acquisition module;

[0035] The image classification module is used to acquire the image in front of the harvester collected by the camera, and perform semantic segmentation on the image. After classifying the harvesting scene at the pixel level, different color mask graphs are used to represent different categories of regions.

[0036] The image extraction module is used to extract the color region of the acquired mask graph to obtain the corresponding RGB image.

[0037] The image preprocessing module is used to perform image preprocessing operations on the acquired RGB image.

[0038] The image transformation module is used for converting each pixel point on the image preprocessed image plane into a corresponding pixel point on a world plane, and then converting each pixel point on the world plane into a corresponding pixel point on a two-dimensional plane image with perspective distortion eliminated; and extracting the color corresponding to the soybean region mask in the ROI region to obtain an RGB image containing only the soybean region;

[0039] The dynamic ROI module is used for extracting the upper boundary line of the cutting table region in the two-dimensional plane image with perspective distortion eliminated, and setting a dynamic ROI according to the upper boundary line of the cutting table region.

[0040] The boundary acquisition module is used for extracting the left and right boundary lines of the soybean region, and fitting the boundary lines into straight lines to obtain the left and right boundaries of the soybean harvesting in the maize-soybean intercropping mode.

[0041] Compared with the prior art, the present application has the following advantages:

[0042] According to one mode of the present application, the present application extracts the soybean region and the cutting table region through semantic segmentation for pixel-level segmentation of the harvesting scene, sets a dynamic ROI according to the position of the cutting table after traditional image processing, extracts the left and right boundary lines of the soybean region in the ROI, and fits the boundary lines into straight lines. The method provided by the present application has high detection precision and fast detection speed, can solve the problems that the current crop boundary detection method is not suitable for the harvesting scene in the maize-soybean intercropping mode, it is difficult to effectively distinguish the maize and soybean regions, and it cannot accurately locate the soybean boundary, and can provide effective and reliable basic data for intelligent control of the soybean harvester in the maize-soybean intercropping mode.

[0043] According to one mode of the present application, a soybean harvesting boundary detection system in a maize-soybean intercropping mode is provided, which comprises an image classification module, an image extraction module, an image preprocessing module, an image transformation module, a dynamic ROI module and a boundary acquisition module. The semantic segmentation, image extraction and dynamic ROI technology are comprehensively applied, the soybean harvesting boundary can be accurately extracted in real time, basic data for intelligent control of the soybean harvester in the maize-soybean intercropping mode is provided, and better soybean harvesting is achieved.

[0044] Note that the description of these effects does not preclude the presence of other effects. One mode of the present application does not necessarily have all the above-mentioned effects. Effects other than the above-mentioned effects can be clearly seen and extracted from the description, drawings, claims, etc. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 is a flowchart of one embodiment of the present application.

[0046] Figure 2is a harvested scene image captured by a camera in an embodiment of the present application, and (b) is a corresponding mask image.

[0047] Figure 3 is a harvested scene image captured by a camera in an ROI region in an embodiment of the present application.

[0048] Figure 4 is a detection effect image of a soybean harvesting boundary in an ROI region in an embodiment of the present application. DETAILED DESCRIPTION

[0049] Embodiments of the present application are described in detail below with reference to the accompanying drawings, in which like reference numerals indicate like elements or elements having the same or similar function throughout the several views. The embodiments described below are examples of the present application, which are intended to explain the present application, and should not be understood as limiting the present application.

[0050] In the description of the present application, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "front", "back", "left", "right", "up", "down", "axial", "radial", "vertical", "horizontal", "inner", "outer", and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only used to facilitate the description of the present application and simplify the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise explicitly specified and limited.

[0051] In the present application, unless otherwise explicitly specified and limited, the terms "mounting", "connection", "connection", "fixing" and the like should be understood broadly, for example, can be fixedly connected, or can be detachably connected, or integrally connected; can be mechanically connected, or can be electrically connected; can be directly connected, or can be indirectly connected through an intermediate medium, or can be connected inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0052] Figure 1 A preferred embodiment of the soybean harvesting boundary detection method in the corn-soybean intercropping mode is shown.

[0053] The soybean harvest boundary detection method under the corn-soybean intercropping pattern includes the following steps:

[0054] Step S1: Image Classification: The camera is mounted at the top center of the harvester's cab. During harvesting, the camera captures images of the area in front of the harvester, such as... Figure 3 As shown, the images captured by the camera are input into a pre-trained semantic segmentation model, and the images are semantically segmented. After pixel-level classification of the captured scene, different colored masks are used to represent regions of different categories.

[0055] Step S2: Image region extraction: Extract color regions from the mask image obtained in step S1 to obtain an RGB image containing only the soybean region and the header region;

[0056] Step S3: Image preprocessing: Perform image preprocessing operations on the RGB image obtained in step S2;

[0057] Step S4: Image transformation: Convert each pixel on the image plane after image preprocessing in step S3 into a corresponding pixel on the world plane through inverse perspective mapping, and then convert each pixel on the world plane into a corresponding pixel on the two-dimensional plane image after perspective distortion is eliminated;

[0058] Step S5: Set dynamic ROI: Extract the upper boundary line of the truncated beam region from the two-dimensional planar image after eliminating perspective distortion in step S4, and set the dynamic ROI according to the upper boundary line of the truncated beam region;

[0059] Step S6: Extract the RGB image of the soybean region: Extract the color of the corresponding soybean region mask within the ROI region in step S5 to obtain an RGB image containing only the soybean region;

[0060] Step S7: Obtain soybean region boundaries: Extract the left and right boundary lines of the soybean region from the RGB image containing only the soybean region in Step S6, and fit the boundary lines into straight lines to obtain the left and right boundaries of soybean harvesting under the corn-soybean intercropping pattern.

[0061] like Figure 2 As shown, Figure 2 (a) shows the harvest scene captured by the camera. Figure 2 In the image (b), the corresponding mask image is shown. In step S1, the harvesting scene is classified into soybean area, corn area, header area and background at the pixel level, and a mask of different color is attached to each area to output the mask image. According to this embodiment, preferably, green, yellow, red and black masks are attached to soybean area, corn area, header area and background respectively.

[0062] According to the embodiment, preferably, the step S2 image region extraction only extracts images of the soybean region and the header region, i.e. green and red regions, to obtain an RGB image containing the soybean region and the header region.

[0063] The specific operation of the step S3 image preprocessing is to first erode the image and then dilate the eroded result. According to the embodiment, preferably, a 5*5 kernel is used for image preprocessing.

[0064] The step S4: According to the embodiment, preferably, the camera is installed on the top of the harvester cab, the camera coordinate system is set as (O1, X1, Y1, Z1), wherein the coordinate origin O1 is located at the center point of the camera, the harvester coordinate system is (O2, X2, Y2, Z2), wherein the coordinate origin O2 is located at the ground directly below O1, the height of the camera from the ground is H=2600mm, the installation pitch angle is θ0=45°, the heading angle is γ0=0°, the height and width of the image collected by the camera are h=1080 and w=1920 respectively, and the preprocessed image is transformed into a two-dimensional plane image through inverse perspective mapping technology, specifically as follows:

[0065] S4.1: Each pixel point on the image plane after image preprocessing is converted into a corresponding pixel point on the world plane, and the formula is as follows:

[0066]

[0067]

[0068] wherein (x, y) is the coordinate of the pixel point on the world plane, (u, v) is the coordinate of each pixel point on the image plane after preprocessing, H is the height of the camera from the ground, a is half of the camera field of view angle, h represents the height of the image, w represents the width of the image, γ0 represents the heading angle of the camera, and θ0 represents the pitch angle of the camera.

[0069] S4.2: Each pixel point on the world plane is converted into a corresponding pixel point on the two-dimensional plane image without perspective distortion, and the formula is as follows:

[0070]

[0071]

[0072] wherein (m, n) is the coordinate of the corresponding pixel point on the two-dimensional plane image without perspective distortion.

[0073] The specific steps of the step S5 setting a dynamic ROI include:

[0074] Step S5.1: performing a top-to-bottom scanning process on the two-dimensional planar image described in step S4, when a pixel point with the same RGB value as the mask color corresponding to the header region described in step S1 is scanned, connecting the points with the same vertical coordinate j on the RGB graph to obtain the upper boundary line of the header region;

[0075] Step S5.2: taking the region with a vertical coordinate within (j-s, j) above the upper boundary line of the header region as the dynamic ROI, wherein s is the width of the dynamic ROI, as shown in Figure 2 According to the present embodiment, the ROI width s is preferably set to 200.

[0076] The specific operation of step S6 for extracting the RGB graph of the soybean region is to extract the pixel points corresponding to the RGB value in the ROI according to the RGB value of the mask color corresponding to the soybean region, to obtain an RGB graph within the ROI containing only the green mask corresponding to the soybean region.

[0077] As shown in Figure 4 The specific steps of step S7 for obtaining the boundary of the soybean region include:

[0078] Step S7.1: performing a top-to-bottom scanning process on the RGB graph within the ROI containing only the soybean region, when a pixel point with the same RGB value as the mask color corresponding to the soybean region described in step S1 is scanned, saving the coordinate point corresponding to the pixel;

[0079] Step S7.2: arranging the coordinate points according to the points with the same vertical coordinate, selecting the points with the smallest and largest horizontal coordinates as the points on the boundary line of the soybean region, connecting the pixel points with the smallest horizontal coordinates in all rows as the left boundary line of the soybean region, and connecting the pixel points with the largest horizontal coordinates in all rows as the right boundary line of the soybean region;

[0080] Step S7.3: fitting the left and right boundary lines of the soybean region into straight lines using the least squares method to obtain the left and right boundaries of the soybean harvest in the corn-soybean intercropping mode.

[0081] A soybean harvest boundary detection system in a corn-soybean intercropping mode includes an image classification module, an image extraction module, an image preprocessing module, an image transformation module, a dynamic ROI module, and a boundary acquisition module.

[0082] The image classification module is used to obtain the image in front of the harvester collected by the camera, and perform semantic segmentation on the image. After classifying the harvesting scene at the pixel level, different color mask graphs are used to represent different categories of regions.

[0083] The image extraction module is used to extract the color region of the obtained mask graph to obtain the corresponding RGB image.

[0084] The image preprocessing module is used for image preprocessing operation on the acquired RGB image.

[0085] The image transformation module is used for converting each pixel point on the image plane after image preprocessing into a corresponding pixel point on a world plane, and then converting each pixel point on the world plane into a corresponding pixel point on a two-dimensional plane image after eliminating perspective distortion; and extracting the color corresponding to the soybean region mask in the ROI region to obtain an RGB image containing only the soybean region.

[0086] The dynamic ROI module is used for extracting the upper boundary line of the header region in the two-dimensional plane image after eliminating perspective distortion, and setting a dynamic ROI according to the upper boundary line of the header region.

[0087] The boundary acquisition module is used for extracting the left and right boundary lines of the soybean region, and fitting the boundary lines into straight lines to obtain the left and right boundaries of the soybean harvesting in the corn-soybean interplanting mode.

[0088] The present application extracts the soybean region and the header region by pixel-level segmentation of the harvesting scene through semantic segmentation, sets a dynamic ROI according to the header position after traditional image processing, extracts the left and right boundary lines of the soybean region in the ROI, and fits the boundary lines into straight lines. The method provided by the present application has high detection accuracy and fast detection speed, can solve the problems that the current crop boundary detection method is not suitable for the harvesting scene in the corn-soybean interplanting mode, it is difficult to effectively distinguish the corn and soybean regions, and it cannot accurately locate the soybean boundary, and can provide effective and reliable basic data for intelligent control of the soybean harvester in the corn-soybean interplanting mode.

[0089] It should be understood that although the present specification is described in terms of various embodiments, not every embodiment contains only one independent technical solution, and the description manner of the specification is only for the sake of clarity, and those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be combined appropriately to form other implementation manners that can be understood by those skilled in the art.

[0090] The series of detailed descriptions listed above are only specific descriptions of the feasible embodiments of the present application, and are not used to limit the protection scope of the present application, and any equivalent embodiments or changes made without departing from the spirit of the present application should be included in the protection scope of the present application.

Claims

1. A method for soybean harvesting boundary detection in a maize-soybean interplanting mode, characterized in that, The method comprises the following steps: Step S1: image classification: obtaining an image collected by a camera in front of a harvester, and performing semantic segmentation on the image, and performing pixel-level classification on a harvesting scene, and using a mask map with different colors to represent regions of different categories; Step S2: image region extraction: performing color region extraction on the mask map obtained in step S1 to obtain an RGB image containing only soybean regions and header regions; Step S3: image preprocessing: performing image preprocessing on the RGB image obtained in step S2; Step S4: image transformation: converting the image preprocessed in step S3 into corresponding pixel points on a two-dimensional plane image free of perspective distortion through inverse perspective mapping; Step S5: setting a dynamic ROI: extracting an upper boundary line of the header region in the two-dimensional plane image free of perspective distortion in step S4, and setting a dynamic ROI according to the upper boundary line of the header region; Step S6: extracting a soybean region RGB image: extracting the color of the mask corresponding to the soybean region in the ROI region in step S5 to obtain an RGB image containing only the soybean region; Step S7: obtaining a soybean region boundary: extracting left and right boundary lines of the soybean region in the RGB image containing only the soybean region in step S6, and fitting the boundary lines into straight lines to obtain left and right boundaries of soybean harvesting in a maize-soybean intercropping mode.

2. The method of claim 1, wherein the soybean harvesting boundary is detected in a corn-soybean interplanting mode. The step S1 classifies the harvesting scene at the pixel level into soybean regions, maize regions, header regions, and backgrounds, and attaches different color masks to each region, respectively, and outputs a mask map.

3. The method of claim 1, wherein the method is used in a maize-soybean interplanting mode. The step S2 image region extraction only extracts the specified color region of the mask corresponding to the soybean region and the header region to obtain an RGB image containing the soybean region and the header region.

4. The method of claim 1, wherein the soybean harvesting boundary is detected in a corn-soybean interplanting mode. The specific operation of the step S3 image preprocessing is to first erode the image and then dilate the eroded result.

5. The method of claim 1, wherein the soybean harvesting boundary is detected in a corn-soybean interplanting mode. In the step S4, the inverse perspective mapping is specifically: converting each pixel point on the image plane after image preprocessing into a corresponding pixel point on a world plane, and then converting each pixel point on the world plane into a corresponding pixel point on a two-dimensional plane image free of perspective distortion; wherein the conversion of each pixel point on the image plane after image preprocessing into a corresponding pixel point on the world plane adopts the following formula: where (x, y) is the coordinate of the pixel point on the world plane, (u, v) is the coordinate of each pixel point on the image plane after preprocessing, H is the height of the camera from the ground, a is half of the camera field of view angle, h represents the height of the image, w represents the width of the image, γ0 represents the heading angle of the camera, and θ0 represents the pitch angle of the camera.

6. The method of claim 5, wherein the soybean harvesting boundary is detected based on the soybean planting boundary and the corn planting boundary. The step S4 converts each pixel point on the world plane into a corresponding pixel point on a two-dimensional plane image free of perspective distortion, which adopts the following formula: where (m, n) is the coordinate of the corresponding pixel point on the two-dimensional plane image free of perspective distortion.

7. The method of claim 1, wherein the soybean harvesting boundary is detected in a corn-soybean interplanting mode. The specific steps of setting the dynamic ROI in step S5 include: Step S5.1: performing a top-to-bottom scanning process on the two-dimensional plane image described in step S4, when a pixel point with the same RGB value as the mask color corresponding to the header region described in step S1 is scanned, connecting the points with the same vertical coordinate j on the RGB image to obtain the upper boundary line of the header region; Step S5.2: taking the region with the vertical coordinate within (j-s, j) above the upper boundary line of the header region as the dynamic ROI, wherein s is the width of the dynamic ROI.

8. The method of claim 1, wherein the soybean harvesting boundary is detected in a corn-soybean interplanting mode. The specific operation of step S6 for extracting the RGB image of the soybean region is to extract the pixel points corresponding to the RGB value in the ROI according to the RGB value of the mask color corresponding to the soybean region, to obtain an RGB image containing only the mask of the soybean region in the ROI.

9. The method of claim 1, wherein the soybean harvesting boundary is detected in a corn-soybean interplanting mode. The specific steps of step S7 for obtaining the boundary of the soybean region include: Step S7.1: performing a top-to-bottom scanning process on the RGB image containing only the soybean region in the ROI, when a pixel point with the same RGB value as the mask color corresponding to the soybean region described in step S1 is scanned, saving the coordinate point corresponding to the pixel; Step S7.2: arranging the coordinate points according to the same vertical coordinate, selecting the points with the minimum and maximum horizontal coordinates as the points on the boundary line of the soybean region, connecting the pixel points with the minimum horizontal coordinate in all rows as the left boundary line of the soybean region, and connecting the pixel points with the maximum horizontal coordinate in all rows as the right boundary line of the soybean region; Step S7.3: fitting the left and right boundary lines of the soybean region into straight lines by the least square method to obtain the left and right boundaries of the soybean harvesting in the maize-soybean intercropping mode.

10. A detection system for applying the soybean harvesting boundary detection method according to any one of claims 1-9 in a maize-soybean intercropping pattern, characterized in that, The system comprises an image classification module, an image extraction module, an image preprocessing module, an image transformation module, a dynamic ROI module, and a boundary acquisition module; The image classification module is used to obtain the image in front of the harvester collected by the camera, and perform semantic segmentation on the image, so as to classify the harvesting scene at the pixel level and represent different regions of different categories by mask images with different colors; The image extraction module is used to extract the color regions of the obtained mask image to obtain the corresponding RGB image; The image preprocessing module is used to perform image preprocessing operations on the obtained RGB image; The image transformation module is used to convert each pixel point on the image plane after image preprocessing into the corresponding pixel point on the world plane, and then convert each pixel point on the world plane into the corresponding pixel point on the two-dimensional plane image after eliminating perspective distortion; and extract the color corresponding to the soybean region mask in the ROI region to obtain an RGB image containing only the soybean region; The dynamic ROI module is used to extract the upper boundary line of the header region in the two-dimensional plane image after eliminating perspective distortion, and set a dynamic ROI according to the upper boundary line of the header region; The boundary acquisition module is used to extract the left and right boundary lines of the soybean region, and fit the boundary lines into straight lines to obtain the left and right boundaries of the soybean harvesting in the maize-soybean intercropping mode.

Images