A rice precision positioning method based on machine vision
Through machine vision technology, the base position of the rice plant stem is determined by using the 'side position, low and downward shooting' acquisition method, Lab color space and K-means algorithm, which solves the problem of low positioning accuracy of rice plant and achieves a higher Positioning accuracy and stability reduce the rate of seedling injury in mechanical weeding.
Patent Information
- Application Number
- CN202111624240.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-28
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2041-12-28
AI Technical Summary
In the prior art, the canopy positioning of rice plants is greatly affected by the regularity of the plant type and lodging state, resulting in a high rate of seedling injury during mechanical weeding and unstable individual positioning accuracy.
Using machine vision-based rice precise positioning method, the image preprocessing is performed through the 'side position, low and top shot' image acquisition method, combined with Lab color space and K-means algorithm, and the base of the rice plant stem is determined as the precise position acquisition point. , and improve positioning reliability through verification modules.
It effectively improves the positioning accuracy of rice plants, avoids the problems of degradation of recognition ability caused by light changes and algae flaking, enhances the stability and reliability of positioning, and reduces the rate of seedling injury.
Smart Images

Figure CN114299144B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of rice plant identification and positioning, and specifically relates to a rice precise positioning method based on machine vision. Background Art
[0002] Environmentally friendly mechanical weeding methods can solve negative problems such as environmental pollution caused by chemical weeding. However, how to effectively avoid rice plants and reduce the seedling injury rate during mechanical weeding has become a major difficulty in the field of mechanical weeding. Since existing weeding machinery lacks an accurate rice plant positioning system, the operating path of the weeding components cannot be adjusted in time according to the changes in the curvature of the seedling belt, resulting in seedling injury problems in the inter-row weeding components, which seriously restricts the development of mechanical weeding technology and equipment. Therefore, conducting research on precise rice plant positioning technology is an effective way to improve the quality of mechanical weeding.
[0003] Machine vision is the mainstream method for crop positioning, with the advantages of high precision and strong real-time performance. Domestic and foreign scholars have conducted relevant research. Crop positioning includes group positioning and individual positioning. Regarding group positioning, ROMEO et al. performed crop-background image segmentation based on image histogram and located crop rows by discriminating image contrast and saturation. Meng Qingkuan et al. selected 2Cg-Cr-Cb feature factors to grayscale the image, established a constraint model for the crop row straight line equation, and used the particle swarm algorithm to optimize the crop row to obtain the baseline. Hu Lian et al. used Otsu image segmentation and morphological operation methods in the super green color space to identify cotton seedlings and lettuce seedlings, and obtained the crop row position information based on the accumulation of row and column pixels combined with the centroid of the connected domain of green plants. The above crop row positioning has high extraction accuracy, but poor positioning accuracy, especially for the complex environment of rice fields, the benchmark accuracy of fitting under the interference of rice field conditions such as rice lodging and missed planting is even more unsatisfactory, and the amount of information processed by the crop row group feature is large, and the complex fitting algorithm has poor real-time performance.
[0004] Regarding the individual positioning of crops, Choi et al. extracted the skeleton of rice plants and the skeleton of leaves based on morphological characteristics, and located the central area of the plant according to its convergence and extensibility. Chen Xueshen et al. adopted a canopy overhead shooting method to extract the centroid image coordinate position of the canopy boundary of rice seedlings, but the positioning accuracy of the above two methods depends on the regularity of the canopy growth of rice plants. However, the actual conditions such as the degree of lodging of rice plants in paddy fields and the influence of wind speed on the canopy shape during image acquisition make the individual positioning accuracy of rice plants based on the canopy extremely unstable. Therefore, the present invention adopts a "side, low, overhead" shooting method, based on the base of the stem as the precise position acquisition point of the rice plants, which effectively improves the positioning accuracy of the rice plants. Summary of the invention
[0005] In view of the technical problems existing in the prior art, the purpose of the present invention is to provide a precise rice positioning method based on machine vision, so as to solve the problem that the traditional rice canopy positioning is affected by the actual conditions such as the regularity of the plant shape and the lodging state, and to provide a perception basis for the navigation of farmland traveling machinery or the alignment of weeding components.
[0006] In order to achieve the above object, the present invention adopts the following technical solution:
[0007] A rice precision positioning method based on machine vision, the following steps are implemented through an acquisition module, a processing module, a positioning module and a verification module: step 1: image acquisition; step 2: image acquisition frequency setting; step 3: image preprocessing; step 4: image segmentation and denoising; step 5: determination of image reference points of the base of the rice plant stem as a ROI area; step 6: demarcation of the image boundary of the base of the rice plant stem as a ROI area; step 7: acquisition of the position of the rice plant; step 8: verification of rice plant positioning data; wherein step 1 is implemented by the acquisition module, steps 3-6 are implemented by the processing module, step 7 is implemented by the positioning module, and step 8 is implemented by the verification module.
[0008] As a preferred embodiment, step 1 adopts the image acquisition method of "side, low, and overhead shooting", and adjusts the installation height and overhead shooting angle of the camera according to the specific conditions of the rice row spacing, plant spacing and the vertical field of view of the camera; specifically, the camera is located directly above the adjacent row of the image acquisition rice row, with a height of H, an angle α with the vertical direction, a vertical field of view angle of the camera of θ, and a horizontal vertical distance D between the bottom edge of the image collected in the direction of the rice row and the camera min The vertical distance between the top edge of the image collected in the rice row direction and the camera is D max , then H and α are expressed by the formula and The acquisition module obtains the rice plant image from the side, and the camera position setting parameters will be adjusted according to the growth status of the rice plant. Before the operation, the camera position is manually adjusted in advance according to the actual situation of the rice field.
[0009] As a preferred embodiment, in step 2, the image acquisition frequency is specifically set according to the rice field environment and operating conditions; specifically: assuming that the rice plant spacing is d, the collected image contains n rice plants, and the traveling speed of the traveling equipment is v, then the calculation formula for the image acquisition frequency f is f=v / nd.
[0010] As a preferred embodiment, in step 3, the collected images are preprocessed using a K-means clustering algorithm in Lab color space; wherein K is the number of classifications of the clustering algorithm; specifically, when the water layer in the rice field has an algae background, K is set to 3, which represents the water layer, rice plants, and algae, respectively; when the water layer in the rice field has no obvious algae background, K is set to 2, which represents the water layer and rice plants, respectively.
[0011] As a preferred embodiment, in step 5, the method for determining the image reference point of the base of the rice plant stem as the ROI area is: after segmenting and denoising the preprocessed image, a scanning method is used to find the lowest point in the vertical direction of the image, and the area of the connected region where the point is located is compared with the area threshold S. When the area of the connected region where the point is located is greater than S, the point is taken as the lowest point of the base of the rice plant stem and as the reference point for demarcating the ROI area; wherein the area threshold S is specifically set according to the growth and tillering conditions of the rice plant.
[0012] As a preferred embodiment, in step 6, the boundary delineation method of the image of the base of the rice plant stem as the ROI area is: smoothing the horizontal projection of the processed binary image, obtaining the positions of all maximum points on the projection curve, and setting the horizontal line of the lowest maximum point in the binary image on the projection curve as the upper boundary of the ROI area; setting the distance between the upper boundary and the lowest point of the stem base to h, and taking the horizontal line h / 5 below the lowest point of the stem base as the lower boundary; the width of the ROI area is equal to the width of the original image.
[0013] As a preferred embodiment, in step 7, the rice plant positioning method uses the intersection of the vertical line where the centroid of the stem base image is located and the horizontal line of the lowest point of the rice plant stem base as the positioning point for determining the rice plant position information.
[0014] As a preferred embodiment, in step 8, two cameras are used to respectively obtain the position information of adjacent rice plants in the same rice row direction, and in the same coordinate system, the horizontal distance between adjacent rice plants is numerically compared with the row spacing of rice planting. The closer the two values are, the more reliable the verification of rice plant positioning data is; the two cameras have the same image acquisition method, are arranged opposite to each other, and the horizontal distance is equal to the row spacing of rice planting.
[0015] In view of the existing rice recognition methods and based on the current mainstream image intelligent recognition field, the present invention constructs a rice precise positioning method based on machine vision, which has the following advantages:
[0016] (1) The “side, low, and overhead” acquisition method avoids the problem of large positioning errors caused by the complex and irregular rice canopy pattern.
[0017] (2) The Lab color model is sensitive to brightness, has a wide color gamut, and has uniform perception. In conjunction with the K-means algorithm, the problem of decreased recognition ability of the super-green algorithm caused by lighting changes, algae blooms, and other conditions is solved.
[0018] (3) Based on the traditional binary image processing method, the stem base image is located and identified by dynamically delineating the ROI area, which takes advantage of the relatively stable and single characteristics of the plant stem base image and effectively improves the positioning accuracy of the rice plant.
[0019] (4) In order to avoid positioning errors caused by incomplete denoising and vibration of the machine during movement, a verification module is added to exclude position points with large rice plant positioning errors, further improving the reliability of rice plant positioning. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 It is the work flow chart of the present invention.
[0021] Figure 2 The original collected image.
[0022] Figure 3a-3c This is the K-means algorithm clustering result under the Lab model when K is 3. Figure 3a For the paddy field background image, Figure 3b is the rice plant image, Figure 3c Noise images such as algae and weeds.
[0023] Figure 4 These are images segmented by the Super Green algorithm and the OTSU algorithm.
[0024] Figure 5 The image after denoising.
[0025] Figure 6 is a binary image (see Figure 5 )Horizontal projection curve after smoothing operation.
[0026] Figure 7 The effect diagram of delineating the ROI area.
[0027] Figure 8 This is the edge extraction effect diagram.
[0028] Fig. 9 This is the effect diagram of positioning in the ROI area.
[0029] Fig.10 Schematic diagram of the relative position between the camera and the rice plant DETAILED DESCRIPTION
[0030] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is now further described in detail in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0031] A rice precision positioning method based on machine vision implements the following steps through an acquisition module, a processing module, a positioning module and a verification module: step 1: image acquisition; step 2: image acquisition frequency setting; step 3: image preprocessing; step 4: image segmentation and denoising; step 5: determination of reference points of the rice plant stem base image (ROI area); step 6: delineation of the boundary of the rice plant stem base image (ROI area); step 7: rice plant position acquisition; step 8: rice plant positioning data verification.
[0032] 1. Introduce the acquisition module, processing module, positioning module and verification module.
[0033] (1) Acquisition module
[0034] The acquisition module must meet the following characteristics: a complete image of the base of the rice stem is required, and a certain distance should be maintained between the image of the base of the rice stem and the edge of the image to prevent the loss of information on the image acquisition object caused by posture changes such as shaking, vibration, and swaying of paddy field machinery.
[0035] (2) Processing module
[0036] The processing module is jointly implemented by software with image processing functions such as Matlab and Labview.
[0037] The function of the processing module is to preprocess, segment and denoise the image. The specific processing is divided into the following steps:
[0038] First, due to different lighting conditions and the number of weeds, the grayscale conversion of plant images by the Super Green Algorithm is unstable. Before grayscale processing, the images are clustered and preprocessed in the Lab color space using the K-means algorithm. The Lab model is sensitive to brightness, has a wide color gamut, and has uniform perception, so that the collected rice plant images are distinguished from the background. The K-means algorithm can divide the image into K categories. When a large number of algae appear in the image, K is set to 3, and the image is divided into three categories: background images without green, such as soil and water layers, green rice images, light green algae, and some weed images. When there is no obvious algae in the image, K is set to 2, and the image is divided into two categories: green rice plant images and background images without green, such as soil, water layers, and reflections.
[0039] Secondly, the green component of the image is extracted by the super green algorithm, and the image is grayed out. The segmentation threshold is obtained by the OTSU algorithm to binarize the image.
[0040] Finally, small area noise such as weeds, leaf adhesion between rice plants, and leaves with large tillering angles are removed through area filtering and morphological methods (opening, closing operations and corrosion, etc.) to obtain a binary image before positioning.
[0041] (3) Positioning module
[0042] After completing the preprocessing, the stem base positioning module segments the ROI area mainly based on the stem base of the rice plant by determining the reference point. The segmentation method is as follows: After completing the image binarization, there are only black and white pixels in the image, and the rice plant image is white. Scan the image to find the lowest point in the vertical direction of the binary image, and compare the area of the connected area where the point is located with the threshold S. When the area is less than S, all points in the connected area are excluded and re-determined to exclude small noise interference such as weeds that are not filtered out in extreme cases; when the area is greater than S, it means that the point belongs to the stem base image and is at the bottom of the image, then the point is used as the reference point for the stem base image (ROI area). Among them, the threshold S can be set according to the specific rice growth conditions and the image projection size.
[0043] After the stem base positioning module determines the lowest point of the stem base image, it smoothes the horizontal projection of the processed binary image and determines the positions of all maximum points on the projection curve. According to the plant type characteristics of the rice plant, the shape gradually widens from the root to the stem base. The lowest extreme point in the corresponding binary image on the projection curve is selected, and its row is used as the upper boundary of the ROI area to select the stem base image with a relatively stable shape for positioning. Let the distance between the upper boundary and the lowest point of the stem base be h, and take the horizontal line h / 5 below the lowest point of the stem base as the lower boundary. The ROI area width is the same as the original image, and the ROI area of the stem base image is delineated.
[0044] The stem base positioning module is used to accurately position the processed rice plant images.
[0045] In the stem base positioning module, after defining the ROI area, the morphological corrosion operation is used to extract the edge of the stem base image.
[0046] In the stem base positioning module, after obtaining the stem base edge image, the connected region algorithm is used to identify the stem base of the rice plant in the ROI area image to determine the number of connected regions in the area. Based on the connected regions, the centroid and the lowest point in the vertical direction of each rice plant stem base image are calculated. The rice plant positioning method uses the intersection of the vertical line where the centroid of the stem base image is located and the horizontal line of the lowest point of the rice plant stem base as the positioning point to determine the rice plant position information.
[0047] (4) Verification Module
[0048] The verification module is a detection module that uses a secondary camera arranged opposite to the main camera to collect images of adjacent rice plants for positioning and verify whether the above positioning data is reliable. The horizontal distance between the two cameras is equal to the row spacing of rice planting. The two cameras are used to obtain the position information of adjacent rice plants in the rice row direction. In the same coordinate system, the horizontal distance between adjacent rice plants is compared with the row spacing of rice planting. The closer the two values are, the more reliable the rice plant positioning data is.
[0049] 2. Give a detailed introduction to a method of precise rice positioning based on machine vision.
[0050] (1) Step 1:
[0051] Observe the growth status of rice plants and the working environment, and adjust the camera frame to a suitable shooting position. Substitute the camera parameters and the horizontal and vertical distances between the camera and the top edge and the camera and the bottom edge in the on-site proofreading image into the formula to calculate the camera height and tilt angle to determine the camera installation position.
[0052] Specifically, the camera is located directly above the adjacent row of the rice row where the image is collected, with a height of H and an angle α with the vertical direction. The vertical field angle of the camera is θ, and the horizontal vertical distance between the bottom edge of the image collected in the direction of the rice row and the camera is D min The vertical distance between the top edge of the image collected in the rice row direction and the camera is D max , then H and α can be expressed by the formula and Calculated to obtain.
[0053] (2) Step 2:
[0054] RGB images of rice plants are collected from paddy fields by “side, low, and overhead” methods; at the same time, the collected images are compressed to avoid the problems of large amount of calculation and slow processing speed caused by the camera’s high resolution.
[0055] The image acquisition frequency is set according to the paddy field environment and operating conditions; specifically, if the rice plant spacing is d, the collected image contains n rice plants, and the travel speed of the traveling equipment is v, then the calculation formula for the image acquisition frequency f is f=v / nd.
[0056] (3) Step 3:
[0057] In Lab color space, the image is clustered by K-means algorithm to filter out the background and retain the green rice image; when a large amount of algae appears in the image, K is set to 3, and the image is divided into three categories: background images without green, such as soil and water layers, green rice images, light green algae, and some weed images. When there is no obvious algae in the image, K is set to 2, and the image is divided into two categories: green rice plant images and background images without green, such as soil, water layers, and reflections.
[0058] Specifically, to convert the color space of an image from RGB to Lab, you need to first convert the RGB model to the XYZ model, expressed as:
[0059] [X, Y, Z] = M * [R, G, B]
[0060] Among them, X, Y, and Z represent the color components of the three channels in the XYZ model; R, G, and B represent the color components of the three channels in the RGB model; M is the gamma function selected by this method for nonlinear tone editing of the image, and the expression is:
[0061]
[0062] After the image is converted from the RGB model to the XYZ model, and then from the XYZ model to the Lab model, the expression is:
[0063]
[0064] Among them, L, a, and b represent the color components of the three channels in the Lab model respectively; according to the gamma function M, in order to map the XYZ model and the RGB model in the same range, X n , Y n , Z n The values are 95.05, 100, and 108.88 respectively; f(t) is the conversion function, and its expression is:
[0065]
[0066] Specifically, the mathematical expression of the K-means algorithm is as follows:
[0067] Assume that the set is partitioned into (C1, C2, ..., C K ), C i For the divided clusters, our goal is to minimize the square error E:
[0068]
[0069] Among them, K is the target number of clusters; μ i is the mean vector of the cluster, i.e. the centroid, expressed as:
[0070]
[0071] (4) Step 4:
[0072] The rice plant images classified by K-means algorithm were grayed out by Super Green algorithm, and the OTSU algorithm was used to automatically obtain the threshold value to binarize the image.
[0073] Specifically, the formula of the super green algorithm is:
[0074]
[0075] Among them, EXG represents the image after grayscale conversion by the super green algorithm; R, G, and B represent the color components of the three channels in the RGB model; R(x, y), G(x, y), and B(x, y) represent the values of the color component at the coordinates (x, y) in the grayscale target image.
[0076] The OTSU method is also known as the maximum between-class variance method. This algorithm is derived using the mathematical method of least squares. It selects a threshold based on the grayscale histogram, divides the image into two parts, black and white, and makes the separation between the two parts as large as possible. It obtains the segmentation threshold and binarizes the image.
[0077] (5) Step 5:
[0078] Use morphological methods and area filtering to remove noise and adhesion caused by leaf tillering. Among them, small area noise such as weeds can be directly removed by area filtering. Morphological operations can cut off the adhesion of leaf images caused by rice tillering, and then use area filtering to remove fragmented images and burrs;
[0079] Area filtering:
[0080]
[0081] Among them, f(i, j) is the filtered target image, h(i, j) is the image after area filtering, and N is the pixel area threshold.
[0082] Morphological opening operation: is the opening operation of the structure operator B on the set A, then:
[0083]
[0084] Morphological closing operation: Let A·B be the closing operation of the structure operator B on the set A, then:
[0085]
[0086] (6) Step 6:
[0087] The lowest white pixel in the vertical direction of the binary image is determined by scanning. To avoid noise interference in extreme cases, the area threshold is used to determine whether it belongs to the connected area of the stem base image. The lowest point of the stem base image in the binary image is found as the reference point for delineating the ROI area.
[0088] Smooths the horizontal projection of a binary image, see Figure 6 ,According to the plant type characteristics of rice plants, the shape gradually widens from the root to the stem base. ,The lowest extreme point in the corresponding binary image on the projection curve is selected as the ,upper boundary of the ROI area, and the stem base image with a relatively stable ,shape is selected for positioning.
[0089] Specifically, let the extreme point correspond to row c2, the row where the lowest point of the stem base image is located is c1, the straight-line distance between the row coordinate and the lowest point of the stem base image is h = |c1-c2|, take the horizontal line h / 5 below the lowest point of the stem base as the lower boundary, and the ROI area width is the same as the original image, and define the ROI area of the stem base image. Figure 7 , the upper and lower boundaries of the ROI area are c2 and c1 respectively, and the height is
[0090] (7) Step 7:
[0091] The stem base image in the ROI region is subjected to boundary extraction. The boundary extraction method is to perform a 3*3 block model corrosion operation on the original image, and to obtain the boundary extracted image by subtracting the eroded image from the original image.
[0092] Morphological erosion operation: Assume that the relationship between set A and set B in two-dimensional space is Then the corrosion of B to A can be defined as:
[0093]
[0094] (8) Step 8:
[0095] The connected region algorithm is used to locate each stem base image. The connected region algorithm is used to locate the centroid and the lowest point of each rice plant stem base, and a vertical line is drawn through the centroid, and a horizontal line is drawn through the lowest point of the connected region. The rice plant positioning method uses the intersection of the vertical line where the centroid of the stem base image is located and the horizontal line where the lowest point of the rice plant stem base is located as the positioning point to determine the rice plant position information.
[0096] Specifically, let the coordinates of the connected centroid be (x M ,y M ), the coordinate of the lowest point is (x L ,y L ), then the coordinates of the positioning point are (x M ,yL ).
[0097] (9) Step 9:
[0098] Two cameras facing each other are used to collect images of adjacent rice plants in the rice row direction, obtain the positioning points of adjacent rice plants, and calculate and compare whether the horizontal distance between adjacent rice plants is roughly equal to the rice row spacing. The closer the distance is to equality, the more reliable the positioning data is.
[0099] In view of the existing rice recognition methods, the present invention is based on the current mainstream image intelligent recognition field. Firstly, the "side, low, and overhead" acquisition method is used to avoid the problem of complex and uneven rice canopy graphics; the Lab color space and K-means algorithm are used to solve the problem of reduced recognition ability of the super green algorithm caused by lighting changes, algae patches, etc.; on the basis of the traditional binary image processing method, the ROI area is dynamically delineated for recognition, and the characteristic that the base position of the rice stem can accurately reflect the position of the rice plant is used to achieve an ideal positioning effect.
[0100] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.
Claims
1. A rice precision positioning method based on machine vision, characterized in that: The following steps are implemented through the acquisition module, processing module, positioning module and verification module: Step 1: Image acquisition; Step 2: Setting the image acquisition frequency; Step 3: Image preprocessing; Step 4: Image segmentation and denoising; Step 5: Determine the image reference point of the base of the rice stem as the ROI area; Step 6: Delineate the image boundary of the base of the rice stem as the ROI area; Step 7: Obtaining the location of rice plants; Step 8: Verification of rice plant location data; Among them, step 1 is implemented by the acquisition module, steps 3-6 are implemented by the processing module, step 7 is implemented by the positioning module, and step 8 is implemented by the verification module; In step 1, a side, low, and overhead image acquisition method is adopted, and the installation height and the overhead angle of the camera are adjusted according to the specific conditions of the rice row spacing, plant spacing, and the vertical field of view of the camera; Specifically, the camera is located above the adjacent row of the rice row where the image is collected, with a height of H and an angle α with the vertical direction. The vertical field of view of the camera is θ, and the horizontal vertical distance between the bottom edge of the image collected in the direction of the rice row and the camera is D min The vertical distance between the top edge of the image collected in the rice row direction and the camera is D max , then H and α are expressed by the formula and Calculated to obtain; In step 5, the method for determining the reference point of the image of the base of the rice plant stem as the ROI area is as follows: after segmenting and denoising the preprocessed image, a scanning method is used to find the lowest point in the vertical direction of the image, and the area of the connected region where the point is located is compared with the area threshold S. When the area of the connected region where the point is located is greater than S, the point is taken as the lowest point of the base of the rice plant stem and used as the reference point for demarcating the ROI area; wherein the area threshold S is specifically set according to the growth and tillering conditions of the rice plant; In step 6, the boundary delineation method of the rice plant stem base image as the ROI area is as follows: the horizontal projection of the processed binary image is smoothed, the positions of all maximum points on the projection curve are obtained, and the horizontal line corresponding to the lowest maximum point in the binary image on the projection curve is set as the upper boundary of the ROI area; the distance between the upper boundary and the lowest point of the stem base is set to h, and the horizontal line h / 5 below the lowest point of the stem base is taken as the lower boundary; the width of the ROI area is equal to the width of the original image; In step 7, the rice plant positioning method uses the intersection of the vertical line where the centroid of the stem base image is located and the horizontal line of the lowest point of the rice plant stem base as the positioning point for determining the rice plant position information.
2. A rice precision positioning method based on machine vision according to claim 1, characterized in that: In step 2, the image acquisition frequency is specifically set according to the paddy field environment and operation conditions; Specifically, assuming that the rice plant spacing is d, the collected image contains n rice plants, and the travel speed of the traveling machine is v, the calculation formula for the image collection frequency f is f=v / nd.
3. A rice precision positioning method based on machine vision according to claim 1, characterized in that: In step 3, the collected images are preprocessed using a K-means clustering algorithm in the Lab color space; K is the number of classifications of the clustering algorithm; specifically, when the water layer in the paddy field has an algae background, K is set to 3, which represents the water layer, rice plants, and algae; when the water layer in the paddy field has no obvious algae background, K is set to 2, which represents the water layer and rice plants.
4. The method for accurately locating rice based on machine vision according to claim 1, characterized in that: In step 8, two cameras are used to obtain the position information of adjacent rice plants in the same rice row direction. In the same coordinate system, the horizontal distance between adjacent rice plants is compared with the row spacing of rice planting. The closer the two values are, the more reliable the rice plant positioning data is. The two cameras have the same image acquisition method and are arranged opposite to each other, and the horizontal distance between them is equal to the row spacing of rice planting.
