Weeding device and method based on visual and dynamic control

Abstract

The invention discloses a weeding device and method based on vision and dynamic control, and the device comprises a support, a soil breaking rake and a soil breaking driving mechanism. The ground breaking driving mechanism comprises a first connecting rod, a second connecting rod, a first electric push rod and a second electric push rod; the upper end of the soil breaking rake is hinged to the support, the lower end of the soil breaking rake is hinged to the first connecting rod, and the other end of the first connecting rod is hinged to the bottom of the second connecting rod. The upper end of the second connecting rod is hinged to a sliding block, and the sliding block is installed on a sliding rail of the support. The first electric push rod is hinged to the support, and a telescopic rod of the first electric push rod is hinged to the hinged position of the first connecting rod and the second connecting rod. The second electric push rod is hinged to the support, and a telescopic rod of the second electric push rod is hinged to the hinged position of the second connecting rod and the sliding block. According to the weeding device, the soil area protruding out of the water surface in the paddy field can be accurately recognized, efficient weeding is conducted, and therefore accurate recognition and efficient operation of the soil area protruding out of the water surface in the paddy field are achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent agriculture, and in particular relates to a weeding device and method based on vision and dynamic control. Background Art

[0002] In rice cultivation, weeding is one of the basic means to ensure the healthy growth of rice. Rice competes with weeds for resources. The growth of weeds not only affects the absorption of nutrients, water and light by rice, but also breeds pests and diseases, thereby reducing yield and quality. Therefore, timely and effective weeding is crucial to increasing rice yield and improving the crop growth environment.

[0003] With the mechanization of agricultural production, mechanized weeding technology has become one of the key means to improve efficiency and reduce labor costs in modern agriculture. Among the many mechanical weeding equipment, passive weeders have been widely used in rice field weeding due to their simple structure and easy operation.

[0004] Passive weeders usually weed between rice rows by pulling weeding wheels, which can quickly remove most weeds and save time and energy for manual weeding. Although passive weeders are easy to operate, the weeding wheels often cannot effectively enter the soil surface on soil protruding from the water surface, especially when the soil is relatively hard, the cutting effect of the weeding wheels is greatly reduced, and the cutting ability of the weeding wheels is limited, resulting in some weeds not being effectively removed; this will cause weeds in some areas to still survive, destroying the uniformity of the crop growth environment, thereby affecting the growth stability of rice and may even have an adverse effect on yield.

[0005] Therefore, in order to further improve the weeding efficiency and the balance of the rice growth environment, agricultural mechanization technology still needs to continue to innovate, especially in the design, function improvement and operation optimization of weeding machinery, to find more efficient and accurate solutions to ensure the stability of the rice growth environment and the sustainable development of agricultural production. Summary of the invention

[0006] In order to overcome the shortcomings of the prior art, the present invention provides a weeding device based on vision and dynamic control, which can accurately identify the soil area protruding from the water surface in the paddy field and perform efficient weeding, thereby achieving accurate identification and efficient operation of the soil area protruding from the water surface in the paddy field.

[0007] The second object of the present invention is to provide a method for weeding soil protruding from the water surface of a paddy field based on vision and dynamic control.

[0008] The technical solution of the present invention to solve the above technical problems is:

[0009] A weeding device based on vision and dynamic control comprises a bracket, a visual recognition system arranged on the bracket, a soil height detection system, a weeding system and a control system, wherein:

[0010] The weeding system comprises a soil-breaking rake arranged on a bracket and a soil-breaking driving mechanism for driving the soil-breaking rake to swing, wherein the soil-breaking driving mechanism comprises a first connecting rod, a second connecting rod, a first electric push rod and a second electric push rod, wherein the upper end of the soil-breaking rake is hinged on the bracket, the lower end is hinged to one end of the first connecting rod, the other end of the first connecting rod is hinged to the lower end of the second connecting rod, the upper end of the second connecting rod is hinged to a slider, the slider is mounted on a slide rail of the bracket, and the slide rail extends straightly; the base of the first electric push rod is hinged to the bracket, and the telescopic rod of the first electric push rod is hinged at the hinge position of the first connecting rod and the second connecting rod; the base of the second electric push rod is hinged to the bracket, and the telescopic rod of the second electric push rod is hinged at the hinge position of the second connecting rod and the slider;

[0011] The visual recognition system is used to identify the soil protruding from the water surface in the paddy field and send the regional data of the identified soil to the control system;

[0012] The soil height detection system is used to detect the soil height of the area where the soil is located, and send the detected soil height data to the control system;

[0013] The control system is used to receive the soil area data sent by the visual recognition system, control the soil height detection system to perform height detection on the area where the soil is located, and control the weeding system to perform weeding operations according to the soil height data detected by the soil height detection system.

[0014] Preferably, a limited depth component is provided at the bottom of the bracket, and the limited depth component is a floating plate installed at the bottom of the bracket.

[0015] Preferably, the visual recognition system is an industrial camera.

[0016] Preferably, the soil height detection system is an ultrasonic sensor.

[0017] A weeding method based on vision and dynamic control comprises the following steps:

[0018] Step S1: collecting paddy field images;

[0019] Step S2: sending the collected paddy field image to the constructed visual recognition model, and identifying whether there is soil protruding from the water surface in the paddy field image by using the constructed visual recognition model; when it is identified that there is soil protruding from the water surface, marking the area of ​​the soil protruding from the water surface; then, starting the ultrasonic sensor, and measuring the height of the area of ​​the soil protruding from the water surface by using the ultrasonic sensor to obtain the height data of the soil protruding from the water surface;

[0020] Step S3: The control system calculates the movement stroke and rotation angle of the first push rod and the second push rod through the acquired height data, so as to perform weeding operation on the area of ​​the soil protruding from the water surface.

[0021] Preferably, in step S2, the steps of constructing the visual recognition model are:

[0022] Step S201: preprocessing the collected paddy field image;

[0023] Step S202: extracting features from the preprocessed paddy field image, wherein the extracted features include color features, texture features and shape features;

[0024] Step S203: normalizing the feature parameters in the color feature, texture feature and shape feature to construct a high-dimensional feature vector;

[0025] Step S204: The high-dimensional feature vector is input into the neural network model as input, and the neural network model is used for training and verification to obtain a visual recognition model that meets the requirements.

[0026] Preferably, in step S202, the characteristic parameters in the color feature include hue, saturation and brightness; the characteristic parameters in the texture feature include contrast, energy and homogeneity; and the characteristic parameters in the shape feature include perimeter, area and shape complexity.

[0027] Preferably, in step S2, the step of obtaining the height data of the soil protruding from the water surface is as follows: an ultrasonic sensor is used to transmit an ultrasonic signal to the area where the soil is located, and the ultrasonic signal is reflected back after encountering the soil surface. After the ultrasonic sensor receives the reflected ultrasonic signal, the ultrasonic sensor uses the signal propagation time to calculate the distance between the ultrasonic sensor and the soil surface; then, the ultrasonic sensor repeatedly measures different positions of the area where the soil is located multiple times to generate a height map of the soil surface to display the height changes of each detection point in the soil.

[0028] Preferably, in step S3, after receiving the height data, the control system first calculates the height difference between the height of each detection position and the water surface height, and determines whether the soil protrudes above the water surface and the degree of the protrusion by comparing the height difference.

[0029] Preferably, in step S3, the steps for weeding the area with soil protruding from the water surface are as follows: when the industrial camera detects that there is soil protruding from the water surface, the control system calculates the required telescopic stroke and rotation angle of the first push rod and the second push rod through the height data measured by the ultrasonic sensor, so as to prompt the ground-breaking rake to reach the target position for weeding; when the industrial camera continues to monitor the working area and no soil protruding from the water surface is found, the visual recognition system outputs the detection result to the control system to confirm that the working area has been processed; after receiving the feedback signal, the control system controls the first electric push rod and the second electric push rod to reset.

[0030] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0031] 1. The vision- and dynamic-control-based weeding method of the present invention combines visual recognition technology and deep learning models, which can accurately distinguish between the soil area covered by the water layer in the paddy field and the hard soil area protruding above the water surface, thereby realizing efficient targeted regional weeding operations.

[0032] 2. The weeding method based on vision and dynamic control of the present invention measures the soil height in real time and feeds it back to the control system; the control system can dynamically adjust the stroke and force of the first push rod and the second push rod according to the received data information to ensure the accuracy of the ground-breaking operation and the ability to adapt to different soil conditions, thereby greatly improving the weeding efficiency and operation quality.

[0033] 3. The weeding method based on vision and dynamic control of the present invention is designed with a closed-loop feedback mechanism and an automatic reset function. After the weeding system completes the weeding operation, when the visual recognition system detects that the working area has been processed, the control system controls the weeding system to automatically reset to the standby state.

[0034] 4. The weeding device based on vision and dynamic control of the present invention is highly automated, reduces manual intervention, and improves the stability, reliability and continuous operation capability of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a structural schematic diagram of the weeding device based on vision and dynamic control of the present invention (not in working state).

[0036] Figure 2 It is a structural schematic diagram of the weeding device based on vision and dynamic control of the present invention (in working state).

[0037] Figure 3 A schematic diagram of a slider.

[0038] Figure 4 A simplified diagram of the movement of the weed control system when it is working.

[0039] Figure 5 This is a simplified diagram of the movement when the weeding system is reset.

[0040] Figure 6 It is a schematic diagram of the flow of the weeding method based on vision and dynamic control of the present invention.

[0041] In the figure: 1-industrial camera; 2-first electric push rod; 3-second electric push rod; 4-first connecting rod; 5-second connecting rod; 6-earth-breaking rake; 7-bracket; 8-floating plate; 9-ultrasonic sensor; 10-slider: DETAILED DESCRIPTION

[0042] The present invention is further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0043] See also Figure 1-Figure 6 The weeding device based on vision and dynamic control of the present invention includes a bracket, a visual recognition system arranged on the bracket, a soil height detection system, a weeding system and a control system.

[0044] See also Figure 1-Figure 6 The weeding system includes a ground-breaking rake arranged on a bracket and a ground-breaking driving mechanism for driving the ground-breaking rake to swing, wherein the ground-breaking driving mechanism includes a first connecting rod, a second connecting rod, a first electric push rod and a second electric push rod, wherein the upper end of the ground-breaking rake is hinged on the bracket, and the lower end is hinged to one end of the first connecting rod, and the other end of the first connecting rod is hinged to the lower end of the second connecting rod, and the upper end of the second connecting rod is hinged to a slider, and the slider is installed on a slide rail of the bracket, and the slide rail extends straightly; the base of the first electric push rod is hinged on the bracket, and the telescopic rod of the first electric push rod is hinged at the hinge position of the first connecting rod and the second connecting rod; the base of the second electric push rod is hinged on the bracket, and the telescopic rod of the second electric push rod is hinged at the hinge position of the second connecting rod and the slider.

[0045] See also Figure 1-Figure 6 The visual recognition system is used to identify the soil protruding from the water surface in the paddy field and send the regional data of the identified soil to the control system; in this embodiment, the visual recognition system is an industrial camera.

[0046] See also Figure 1-Figure 6 The soil height detection system is used to detect the soil height of the area where the soil is located, and send the detected soil height data to the control system; in this embodiment, an ultrasonic sensor is provided on the bracket.

[0047] See also Figure 1-Figure 6The control system is used to receive the soil area data sent by the visual recognition system, control the soil height detection system to perform height detection on the area where the soil is located, and control the weeding system to perform weeding operations based on the soil height data detected by the soil height detection system.

[0048] See also Figure 1-Figure 6 A limited depth component is provided at the bottom of the bracket, and the limited depth component is a floating plate installed at the bottom of the bracket.

[0049] See also Figure 1-Figure 6 The weeding method based on vision and dynamic control of the present invention comprises the following steps:

[0050] Step S1: collecting paddy field images;

[0051] Step S2: sending the collected paddy field image to the constructed visual recognition model, and identifying whether there is soil protruding from the water surface in the paddy field image by using the constructed visual recognition model; when it is identified that there is soil protruding from the water surface, marking the area of ​​the soil protruding from the water surface; then, starting the ultrasonic sensor, and measuring the height of the area of ​​the soil protruding from the water surface by using the ultrasonic sensor to obtain the height data of the soil protruding from the water surface;

[0052] Step S3: The control system calculates the movement stroke and rotation angle of the first push rod and the second push rod through the acquired height data, so as to perform weeding operation on the area of ​​the soil protruding from the water surface.

[0053] See also Figure 1-Figure 6 In step S2, the steps of constructing the visual recognition model are:

[0054] Firstly, the collected paddy field images are preprocessed, and then the features are extracted from the preprocessed paddy field images;

[0055] In the visual recognition system of paddy field soil, the selection of feature parameters is crucial for the identification of reflection problems and soil protruding from the water surface. In order to ensure that the soil area can be effectively identified in a complex paddy field environment, multiple feature parameters that are helpful in distinguishing soil from water surface are selected in this embodiment, and the calculation results of these feature parameters are used as the input of the neural network model to improve the recognition accuracy. Among them, the extraction parameters are as follows: color feature, texture feature and shape feature;

[0056] 1) The extraction of color features is to convert the paddy field image from RGB color space to HSV color space, extract characteristic parameters such as hue, saturation and brightness to help distinguish soil from water surface, specifically:

[0057] Convert an RGB image to an HSV image using the following formula:

[0058] C max =max(R, G, B)C min =min(R, G, B)Δ=C max -C min

[0059] For each pixel in the paddy field image, calculate its hue (H), saturation (S) and brightness (V), and then count the mean and standard deviation of these feature parameters in the image area:

[0060]

[0061] Where: N is the number of pixels in the area, Hi, Si, Vi are the hue, saturation and brightness values ​​of the i-th pixel in the paddy field image respectively.

[0062] 2) Texture feature extraction is done through local binary pattern (LBP) or gray level co-occurrence matrix (GLCM). Specifically, for each pixel, the LBP value is calculated by comparing the pixel values ​​of its surrounding neighborhood with the central pixel value:

[0063]

[0064] Where: s k is the binary value (0 or 1) generated after comparison with the central pixel value; P is the number of pixels in the neighborhood (usually 8 neighborhood pixels);

[0065] The gray-level co-occurrence matrix (GLCM) is used to describe the spatial relationship of the gray levels of pixels in the paddy field image. The element M(i, j) in the gray-level co-occurrence matrix (GLCM) represents the number of co-occurrences of gray level i and gray level j at a given distance d and angle θ; the texture features, namely contrast, energy and homogeneity, are calculated by the gray-level co-occurrence matrix (GLCM), which are:

[0066]

[0067] 3) Shape feature extraction is to extract the shape features of the soil area through edge detection algorithm and contour analysis, and further identify the raised parts of the soil, among which,

[0068] Use the Canny green edge detection algorithm to obtain the green edge area in the paddy field image:

[0069] Edges=Canny(I)

[0070] Where: I is the input image; Canny(I) is the edge image generated by the Canny algorithm.

[0071] Use contour analysis to calculate soil area features such as perimeter, area, shape complexity, etc.:

[0072]

[0073] The calculated feature parameters are normalized to form a high-dimensional feature vector as the input of the neural network model, where:

[0074] Combine all calculated features into a vector:

[0075] X=[H mean , H max , H min , Contrast, Energy, Homogeneity, Shape Features];

[0076] In the formula: the parameters in Features mainly include the total length of the target area boundary, the area of ​​the target area, and the shape complexity (measured by the relationship between the area and the boundary length);

[0077] In order to prevent different feature parameters from affecting the performance of the neural network model during training, high-dimensional feature vectors need to be normalized:

[0078]

[0079] Where: μ and σ are the mean and standard deviation of the high-dimensional feature vector, X is the high-dimensional feature vector, X norm It is a normalized high-dimensional feature vector.

[0080] The normalized high-dimensional feature vector is used as the input of the neural network model:

[0081] Output = X norm

[0082] The neural network learns the relationship between these features and finally outputs the recognition result to determine the state of the soil area, that is, "soil protruding from the water surface". At the same time, when the visual recognition system detects soil protruding from the water surface, it sends a signal to the control system, and the control system activates the ultrasonic sensor to measure the soil height of the target area, specifically:

[0083] The ultrasonic sensor transmits an ultrasonic signal to the surface of the soil. The ultrasonic signal is reflected back after encountering the soil surface. After receiving the reflected signal, the ultrasonic sensor uses the signal propagation time (i.e. the time difference from transmission to reception) to calculate the distance. The formula is as follows:

[0084]

[0085] In the formula: v is the propagation speed of ultrasound in the air, Δt is the time difference from the emission to the reception of ultrasound, in seconds; the distance from the ultrasonic sensor to the soil surface, that is, the height information, is obtained by calculation.

[0086] The ultrasonic sensor takes multiple measurements at different locations in the soil area and generates a height map of the soil surface that shows the height variation at each detection point within the soil area.

[0087] The ultrasonic sensor’s measurement data needs to be combined with the visual recognition model’s recognition results to accurately identify the convexity of the soil; after the visual recognition system detects the soil, the ultrasonic sensor’s task is to provide the height information of the area where the soil is located. This process includes the following steps:

[0088] The visual recognition model is used to identify the area where the soil is located in the paddy field image and mark the boundary of the area. The visual recognition model sends a signal to the control system, indicating that further height measurement of the marked area is required. The control system instructs the ultrasonic sensor to measure the height of the specific area based on the visually recognized area. The ultrasonic sensor obtains a height map of the area through distance measurement and transmits the height data to the control system.

[0089] After receiving the height data, the control system needs to further analyze the height data to determine whether the soil protrudes above the water surface and the degree of protrusion; this process includes the following steps:

[0090] The control system calculates the height difference between the height of each measuring point and the water surface height:

[0091] Height Difference=H soil -H water

[0092] Where: H soil is the height of the soil area, H water is the height of the water surface;

[0093] By comparing these height differences, the control system can determine whether the soil protrudes above the water surface and the degree of the protrusion; that is, based on the height difference, the control system determines whether the soil area is above the water surface, and determines the degree of soil protrusion based on the size of the height difference; if the height difference is greater than the predetermined threshold, it means that the soil is relatively protruding and special operations may be required (such as using a ground-breaking rake for weeding); if the height difference is small, it means that the soil is still covered by water, and the control system can adjust subsequent operations based on this information (such as using a weeding wheel at the bottom of the bracket for weeding); the control system determines the working stroke of the first electric push rod and the second electric push rod based on the analysis results of the degree of soil protrusion. Assuming that the degree of soil protrusion is large, the control system may instruct the first electric push rod and the second electric push rod to make a larger adjustment; if the degree of protrusion is small, the actions of the first electric push rod and the second electric push rod can be appropriately reduced to avoid unnecessary soil interference.

[0094] When the industrial camera detects soil protruding above the water surface, the ultrasonic sensor starts working, and after measuring the soil height data, the control system analyzes and processes it, and calculates the required telescopic stroke of the first electric push rod and the second electric push rod, wherein the extension action of the first electric push rod and the second electric push rod is accurately executed through electric control, so that it can drive the first connecting rod, the second connecting rod and the ground-breaking rake to gradually descend until the telescopic rods of the first electric push rod and the second electric push rod reach the target position, wherein, since the second push rod will first reach the specified position and then lock, the second push rod will then accurately adjust the position of the ground-breaking rake to ensure that the four-bar linkage (i.e., composed of the first connecting rod, the second connecting rod, the ground-breaking rake and the bracket) is in a dead point state, thereby achieving dynamic control.

[0095] In the above dynamic control process, it is necessary to obtain the movement stroke and rotation angle of the first electric push rod and the second electric push rod to ensure accurate adjustment of the height of the earth-breaking rake, wherein the stroke of the first electric push rod can be calculated by the following formula, and similarly, the stroke of the second electric push rod can also be calculated by the following formula:

[0096]

[0097] In order to ensure that the ground-breaking rake is in the correct position, in addition to controlling the telescopic length of the first electric push rod and the second electric push rod, it is also necessary to control the rotation angle of the first electric push rod and the second electric push rod; the rotation angle of the first electric push rod and the second electric push rod can be calculated by the following formula:

[0098]

[0099] After the control system calculates the stroke and rotation angle of the first electric push rod and the second electric push rod, the position of the earth-breaking rake can be determined by the geometric positions of the first electric push rod and the second electric push rod. Assuming that the earth-breaking rake is driven by the first electric push rod and the second electric push rod, its position can be expressed as:

[0100] x=l2 new ·cos(θ2)+l3 new ·cos(θ3)

[0101] y=l2 new ·sin(θ2)+l3 new sin(θ3)

[0102] Wherein, x and y represent the position of the ground breaking harrow on the horizontal plane.

[0103] When the industrial camera continuously monitors the working area and finds that no soil protruding from the water surface is detected, the visual recognition system outputs the detection result to confirm that the target area has been processed. At this time, the control system receives the feedback signal and instructs the first electric push rod and the second electric push rod to enter the reset state; during the reset process, the first electric push rod and the second electric push rod begin to shrink, driving the first connecting rod and the second connecting rod to gradually return to the initial position; as the first electric push rod and the second electric push rod shrink, the first connecting rod, the second connecting rod and the ground-breaking rake gradually rise, and the ground-breaking rake is retracted from the bottom of the floating plate to the top of the floating plate and remains in a non-working state; and this reset action ensures that the weeding device returns to the standby state and is ready for the next weeding operation.

[0104] Through the above design, the resetting action of the first electric push rod and the second electric push rod not only ensures the continuity and stability of the weeding system, but also effectively avoids the misoperation of the first electric push rod and the second electric push rod due to failure to reset; the entire weeding process is completed by the visual recognition system and the control system, realizing the automated closed-loop control of the weeding operation, thereby improving the operation efficiency and system reliability.

[0105] In addition, the present invention can accurately identify the soil covered by the water layer and the hard soil area protruding from the water surface in the paddy field through the combination of industrial cameras with deep learning models and reflection suppression technology, and intelligently switch the operation mode according to the soil status: use a weeding wheel to cut weeds in the water-covered area, and use the first electric push rod and the second electric push rod to drive the ground-breaking rake to break the soil and weed in the hard soil area protruding from the water surface.

[0106] In addition, the present invention combines an ultrasonic sensor to measure the soil height in real time, and the control system dynamically adjusts the telescopic stroke of the first electric push rod and the second electric push rod to ensure the accuracy and efficiency of the weeding operation; the closed-loop control mechanism and automatic reset function adopted by the present invention ensure the stable operation and continuous operation of the weeding system, thereby significantly improving the quality of weeding operations in the entire area.

[0107] The above is a preferred embodiment of the present invention, but the embodiment of the present invention is not limited to the above content. 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 weeding device based on vision and dynamic control, characterized in that: It includes a support, a visual recognition system arranged on the support, a soil height detection system, a weeding system and a control system, wherein: The weeding system comprises a soil-breaking rake arranged on a bracket and a soil-breaking driving mechanism for driving the soil-breaking rake to swing, wherein the soil-breaking driving mechanism comprises a first connecting rod, a second connecting rod, a first electric push rod and a second electric push rod, wherein the upper end of the soil-breaking rake is hinged on the bracket, the lower end is hinged to one end of the first connecting rod, the other end of the first connecting rod is hinged to the lower end of the second connecting rod, the upper end of the second connecting rod is hinged to a slider, the slider is mounted on a slide rail of the bracket, and the slide rail extends straightly; the base of the first electric push rod is hinged to the bracket, and the telescopic rod of the first electric push rod is hinged at the hinge position of the first connecting rod and the second connecting rod; the base of the second electric push rod is hinged to the bracket, and the telescopic rod of the second electric push rod is hinged at the hinge position of the second connecting rod and the slider; The visual recognition system is used to identify the soil protruding from the water surface in the paddy field and send the regional data of the identified soil to the control system; The soil height detection system is used to detect the soil height of the area where the soil is located, and send the detected soil height data to the control system; The control system is used to receive the soil area data sent by the visual recognition system, control the soil height detection system to perform height detection on the area where the soil is located, and control the weeding system to perform weeding operations according to the soil height data detected by the soil height detection system.

2. The weeding device based on vision and dynamic control according to claim 1, characterized in that: A limited depth component is arranged at the bottom of the bracket, and the limited depth component is a floating plate installed at the bottom of the bracket.

3. The weeding device based on vision and dynamic control according to claim 2, characterized in that: The visual recognition system is an industrial camera.

4. The weeding device based on vision and dynamic control according to claim 3, characterized in that: The soil height detection system is an ultrasonic sensor.

5. A weeding method for the weeding device based on vision and dynamic control according to any one of claims 1 to 4, characterized in that: The following steps are involved: Step S1: collecting paddy field images; Step S2: sending the collected paddy field image to the constructed visual recognition model, and identifying whether there is soil protruding from the water surface in the paddy field image by using the constructed visual recognition model; when it is identified that there is soil protruding from the water surface, marking the area of ​​the soil protruding from the water surface; then, starting the ultrasonic sensor, and measuring the height of the area of ​​the soil protruding from the water surface by using the ultrasonic sensor to obtain the height data of the soil protruding from the water surface; Step S3: The control system calculates the movement stroke and rotation angle of the first push rod and the second push rod through the acquired height data, so as to perform weeding operation on the area of ​​the soil protruding from the water surface.

6. The weeding device based on vision and dynamic control according to claim 5, characterized in that: In step S2, the steps of constructing the visual recognition model are: Step S201: preprocessing the collected paddy field image; Step S202: extracting features from the preprocessed paddy field image, wherein the extracted features include color features, texture features and shape features; Step S203: normalizing the feature parameters in the color feature, texture feature and shape feature to construct a high-dimensional feature vector; Step S204: The high-dimensional feature vector is input into the neural network model as input, and the neural network model is used for training and verification to obtain a visual recognition model that meets the requirements.

7. The weeding device based on vision and dynamic control according to claim 6, characterized in that: In step S2, in step S202, the characteristic parameters in the color feature include hue, saturation and brightness; the characteristic parameters in the texture feature include contrast, energy and homogeneity; the characteristic parameters in the shape feature include perimeter, area and shape complexity.

8. The weeding device based on vision and dynamic control according to claim 7, characterized in that: In step S2, in step S2, the step of obtaining the height data of the soil protruding from the water surface is as follows: an ultrasonic sensor transmits an ultrasonic signal to the area where the soil is located, and the ultrasonic signal is reflected back after encountering the soil surface. After the ultrasonic sensor receives the reflected ultrasonic signal, the ultrasonic sensor uses the signal propagation time to calculate the distance between the ultrasonic sensor and the soil surface; then, the ultrasonic sensor repeatedly measures different positions of the area where the soil is located to generate a height map of the soil surface to display the height changes of each detection point in the soil.

9. The weeding device based on vision and dynamic control according to claim 8, characterized in that: In step S2 and step S3, after receiving the height data, the control system first calculates the height difference between the height of each detection position and the water surface height, and determines whether the soil protrudes above the water surface and the degree of the protrusion by comparing the height difference.

10. The weeding device based on vision and dynamic control according to claim 9, characterized in that: In step S2, in step S3, the steps of weeding the area of ​​soil protruding from the water surface are as follows: when the industrial camera detects that there is soil protruding from the water surface, the control system calculates the required telescopic stroke and rotation angle of the first push rod and the second push rod through the height data measured by the ultrasonic sensor, so as to prompt the ground-breaking rake to reach the target position for weeding; when the industrial camera continues to monitor the working area and does not find any soil protruding from the water surface, the visual recognition system outputs the detection result to the control system to confirm that the working area has been processed; After receiving the feedback signal, the control system controls the first electric push rod and the second electric push rod to reset.

Images