Weeding robot

Abstract

The invention provides a weeding robot, and relates to the field of intelligent weeding, the weeding robot comprises a main case, damping devices, lifting devices, walking devices and a weeding assembly, the outer walls of four corners of the main case are respectively provided with the vertical lifting devices through the damping devices, and the walking devices are arranged at the bottom ends of the lifting devices; the weeding assembly comprises a rotating seat, a three-axis mechanical arm, a four-channel spray head, camera holders, an acquisition module and a radar, the rotating seat is fixedly arranged in the middle of the bottom surface of the main machine box, one end of the three-axis mechanical arm is connected with the rotating seat, the other end of the three-axis mechanical arm is provided with the four-channel spray head, and the two camera holders are symmetrically arranged on the bottom surface of the main machine box on two sides of the rotating seat; the two sets of radars are arranged in the middle of the front side face and the middle of the rear side face of the mainframe box respectively. The robot performs targeted precise pesticide application, is high in environmental adaptability, remarkably reduces the pesticide application amount, and avoids the problems of herbicide waste, environmental pollution and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent weeding, and particularly relates to a weeding robot. Background Art

[0002] Cotton, as an important cash crop in China, occupies an important position in the agricultural economy. It has a long planting history and is widely used in fields such as textiles and medicine. During the cotton planting process, the proliferation of weeds in cotton fields seriously affects the growth and yield of cotton. Weeds not only compete with cotton for nutrients, water, and sunlight but also become a transmission medium for pests and diseases, resulting in a significant reduction in cotton yield and quality. According to statistics, the annual occurrence area of weeds in cotton fields is nearly 20 million mu. If not effectively controlled, it will cause a yield loss of up to 14% - 16%.

[0003] Currently, the control of weeds in cotton fields mainly relies on manual weeding and chemical weeding. Manual weeding has advantages such as being green, safe, and highly accurate, but it has low efficiency and high labor intensity, making it difficult to meet the needs of large-scale planting. Chemical weeding saves time and effort and has high efficiency, and can effectively avoid excessive use of labor costs. However, the large-scale spraying of chemical agents has a large dosage and low precision, which not only causes waste of agents but also the excess agents are easily adhered to cotton and dissipated in the air, resulting in problems such as environmental pollution or crop phytotoxicity. Summary of the Invention Aiming at the above problems existing in the prior art, the purpose of the present invention is to provide a weeding robot that can effectively identify weeds in cotton fields with high precision, so as to perform targeted and precise pesticide application. The amount of pesticide applied is precisely controllable, and it has strong environmental adaptability. It can significantly reduce the use of herbicides while reducing the number of weeds in cotton fields, avoid environmental pollution, and improve the yield and quality of cotton.

[0004] The purpose of the present invention is achieved through the following technical solutions: A weeding robot includes a main chassis, a shock absorption device, a lifting device, a traveling device, and a weeding component. The outer walls at the four corners of the main chassis are respectively provided with vertical lifting devices through the shock absorption device, and the traveling device is arranged at the bottom end of the lifting device. The weeding component includes a rotating base, a three-axis robotic arm, a four-channel nozzle, a camera pan-tilt, a collection module, and a radar. The rotating base is fixedly arranged in the middle of the bottom surface of the main chassis. One end of the three-axis robotic arm is connected to the rotating base, and the other end is provided with a four-channel nozzle. There are two camera pan-tilts, and they are symmetrically arranged on the bottom surface of the main chassis on both sides of the rotating base. The collection modules are respectively arranged on the camera pan-tilts. There are two groups of radars, and they are respectively arranged in the middle of the front and rear sides of the main chassis.

[0005] For further optimization based on the above solution, the shock absorber device includes a shock absorber spring, a damper and a support rod. The shock absorber spring is fixedly installed outside the damper through a buckle, and the upper end of the damper is connected to the side wall of the main chassis, and the lower end is connected to the top of the corresponding lifting device (the damper is in clearance fit with the main chassis and the lifting device and can rotate relative to each other); there are two support rods which are arranged parallel to each other. The two support rods are arranged between the main chassis and the lifting device and are located below the corresponding damper. The two ends of the support rod are respectively connected to the side wall of the main chassis and the corresponding lifting device (the support rod is in clearance fit with the main chassis and the lifting device and can rotate relative to each other).

[0006] For further optimization based on the above solution, the lifting device adopts a hydraulic telescopic rod, and the hydraulic telescopic rod is controlled by a hydraulic pump arranged in the inner cavity of the corresponding side main chassis.

[0007] For further optimization based on the above solution, the traveling device includes a steering reduction gearbox housing, a steering reduction motor, a reduction gear set, a traveling wheel protection housing, a traveling wheel and a hub motor. The steering reduction gearbox housing is arranged at the bottom end of the lifting device (i.e., the end far from the corresponding shock absorber device), and a steering reduction motor and a steering reduction gear set are installed therein. A Hall angle sensor is arranged in the steering reduction gearbox housing. The upper end of the traveling wheel protection housing passes through the steering reduction gearbox housing and is connected to the internal output stage of the reduction gear set. The lower end of the traveling wheel protection housing is provided with a traveling rotating shaft, and a traveling wheel is sleeved on the outer wall of the traveling rotating shaft. The hub motor is fixedly installed on the traveling wheel protection housing, and its output end is connected to the traveling rotating shaft.

[0008] For further optimization based on the above solution, the four-channel nozzle includes a nozzle front end, a nozzle rear end and a liquid distribution module. One side of the nozzle rear end is connected to the end of the three-axis robotic arm far from the main chassis, and a water inlet channel is opened inside it. The water inlet channel is communicated with a water tank arranged in the inner cavity of the main chassis, and a metering pump is arranged on their connecting pipe; the nozzle front end is connected to the nozzle rear end, and a mixing chamber is opened inside the nozzle front end. The end of the nozzle rear end far from the three-axis robotic arm is located inside the mixing chamber; four liquid distribution modules are respectively arranged on the outer wall of the nozzle front end and at the corresponding end of the nozzle rear end. A liquid medicine channel communicated with the mixing chamber is opened in the middle of the liquid distribution module, and a solenoid valve is arranged on one side of the liquid distribution module. The sides of the liquid distribution module far from the nozzle front end are respectively communicated with the corresponding medicine boxes arranged in the main chassis (i.e., four medicine boxes are arranged in the inner cavity of the main chassis), and metering pumps are respectively arranged between them.

[0009] A weeding method in cotton fields, using the above weeding robot, includes: Step S1: Adjust the overall height of the main chassis according to the plant height of the cotton and control the weeding robot to travel in the cotton field; Step S2: During the traveling process, detect the weeds in the cotton field in real time and generate an image with spectral information annotation; Step S3: Use the image with spectral information annotation to accurately position the robotic arm; Step S4: When the nozzle at the front end of the robotic arm reaches the specified position, control the metering pump of the corresponding classification module to start, and at the same time start the metering pump of the water inlet channel, and turn on the electromagnetic valve of the corresponding classification module to achieve accurate and quantitative spraying of the corresponding liquid medicine.

[0010] Based on the further optimization of the above solution, the specific content of step S2 is as follows: Use the improved YOLOv10 to detect weeds in the cotton field. The improved YOLOv10 uses a multi-spectral feature fusion detection module and adopts a dual-channel input architecture of the visible light channel and the near-infrared channel. Among them, the visible light channel is used to receive the RGB three-band image, and the near-infrared channel is used to receive the NIR single-band image. The channel attention mechanism is used to achieve feature fusion:

[0011] In the formula: F fused Represents the fused feature; Represents the channel attention function; F RGB , F NIR Respectively represent the visible light channel feature and the near-infrared channel feature; Represents the feature addition operation; Represents the backbone network feature extraction function; Then, the classification of cotton plants and weeds is realized through spectral difference discrimination, and the discrimination is carried out through the double thresholds of the vegetation indices of the normalized difference vegetation index (NDVI) and the enhanced difference vegetation index (EDVI):

[0012] In the formula: NIR Represents the near-infrared band reflectance; R Represents the red band reflectance; B Represents the blue band reflectance; Preset the first threshold of the normalized difference vegetation index Y NDVI-1 And the second threshold Y NDVI-2 , as well as the first threshold of the enhanced difference vegetation index Y EDVI-1 And the second threshold Y EDVI-2 : When NDVI is greater than Y NDVI-1 And EDVI is less than Y EDVI-1 It is determined as a cotton plant; when NDVI is greater than YNDVI-2 and when the EDVI is greater than Y EDVI-2 it is determined as a weed; in other cases, it is determined as a background factor or a non-biological obstacle (through the dual discrimination matrix of NDVI and EDVI, a complementary enhancement relationship is formed; not only ensuring the stable recognition of the main vegetation - cotton, but also improving the monitoring accuracy and sensitivity of weeds in the cotton field, so as to quickly and accurately complete the identification and classification of cotton and weeds under complex lighting conditions and field interference conditions); Generate a four-channel annotation matrix by using the method of spectral annotation images ; for each detection region Ri, if the pixel point then:

[0013] In the formula: c i represents the class identifier; W , H represent the spatial resolution (height, width) of the image; NDVI The value is normalized to [0, 1]; For the weed detection frame in the two-dimensional image ( x c , y c , w , h ), its center point is:

[0014] Then the corresponding three-dimensional space coordinates are:

[0015] In the formula: ( c x , c y ) represents the optical center coordinates; f represents the lens focal length; B represents the binocular baseline distance; d represents the pixel value corresponding to the disparity map; represents the compensation coefficient.

[0016] Based on the further optimization of the above scheme, the specific step S3 is: Establish a homogeneous transformation matrix from the camera coordinate system to the base coordinate system of the three-axis robotic arm:

[0017] In the formula: ( T x , T y ,T z represents the offset of the camera installation position; represents the deflection angle of the camera installation; ( X cam , Y cam , Z cam ) represents the three-dimensional point coordinates obtained from the camera image; Compensate for the terrain and correct the Z-axis coordinate of the base coordinate system of the three-axis robotic arm:

[0018] In the formula: represents the ground height difference, obtained by a laser rangefinder; Judge the reachability of the three-axis robotic arm:

[0019] When D is greater than L 1 + L 2, it means that the three-axis robotic arm cannot reach, trigger an alarm, and start the walking device to move; among them, L 1 represents the length of the first link of the three-axis robotic arm, L 2 represents the length of the second link of the three-axis robotic arm; otherwise, it is determined that the three-axis robotic arm can reach; Obtain the joint angles of the first link and the second link respectively q 1, q 2:

[0020] Control the three-axis robotic arm by controlling the joint angles; At the same time, the vertical displacement of the four-channel nozzle is .

[0021] The following are the technical effects achieved by the technical solution of the present invention: The weeding robot provided by the present invention can adjust the corresponding weeding height according to the height of cotton plants, effectively avoiding interference between cotton plants and the weeding components during the weeding process, which may cause damage to cotton during weeding. At the same time, through the structural design of the rotating base, three-axis robotic arm, four-channel nozzle, camera pan-tilt, acquisition module and radar, the weeding robot of the present invention can, firstly, effectively distinguish weeds from cotton plants, avoiding a large amount of weeding agents adhering to cotton plants during spraying, which may lead to problems such as cotton yield reduction and quality decline. Secondly, it can spray pesticides precisely according to the identified weeds (i.e., targeted spraying), with a high spraying accuracy (high utilization rate of weeding agents) and excellent weeding effect. Thirdly, it can precisely control and adjust the dosage of pesticides (i.e., quantitative spraying), avoiding waste during spraying and environmental pollution.

[0022] Through the targeted, quantitative and precise pesticide application of the weeding robot of the present invention, the usage amount of herbicides can be reduced by about 30% - 50%, while ensuring the weed removal rate, reducing the risks of environmental pollution and cotton phytotoxicity, realizing automated weeding, greatly improving the management efficiency of cotton fields, and providing technical support for the sustainable development of agriculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a schematic structural diagram of the weeding robot in an embodiment of the present invention.

[0024] Figure 2 It is a schematic internal diagram of the main chassis of the weeding robot in an embodiment of the present invention.

[0025] Figure 3 It is a schematic structural diagram of the shock absorption device and lifting device of the weeding robot in an embodiment of the present invention.

[0026] Figure 4 It is a schematic structural diagram of the traveling device of the weeding robot in an embodiment of the present invention.

[0027] Figure 5 It is a schematic structural diagram of the weeding component of the weeding robot in an embodiment of the present invention.

[0028] Figure 6 It is a schematic structural diagram of the four-channel nozzle of the weeding robot in an embodiment of the present invention; among them, Figure 6 (a) is a schematic overall structure diagram; Figure 6 (b) is a cross-sectional view.

[0029] Wherein: 10, main chassis; 21, shock-absorbing spring; 22, support rod; 30, lifting device; 41, steering reducer housing; 42, steering reduction motor; 43, traveling wheel protection housing; 44, traveling wheel; 51, rotating base; 52, three-axis robotic arm; 53, four-channel nozzle; 531, rear end of the nozzle; 532, front end of the nozzle; 533, liquid distribution module; 5330, solenoid valve; 54, camera pan-tilt head; 55, acquisition module; 56, radar. Detailed implementation mode

[0030] The technical solutions in the embodiments of the present invention will be clearly and completely described below. In the following description, specific details such as specific system structures and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of the present invention.

[0031] Embodiment 1: A weeding robot includes a main chassis 10, a shock-absorbing device, a lifting device 30, a traveling device, and a weeding component (the overall structure is as Figure 1 shown). The outer walls of the four corners of the main chassis 10 are respectively provided with a vertical lifting device 30 through the shock-absorbing device (as Figure 1 shown), and the traveling device is arranged at the bottom end of the lifting device 30; the shock-absorbing device includes a shock-absorbing spring 21, a damper, and a support rod 22. The shock-absorbing spring 21 is fixedly installed outside the damper through a buckle (that is, the shock-absorbing spring 21 is sleeved on the outer wall of the damper, as Figure 3 shown), and the upper end of the damper is connected to the side wall of the main chassis 10, and the lower end is connected to the top of the corresponding lifting device 30 (the damper and the main chassis 10, the lifting device 30 are all in clearance fit and can rotate relative to each other, as Figure 3 shown); the support rods 22 are two, and the two support rods 22 are arranged in parallel (as Figure 3 shown). The two support rods 22 are arranged between the main chassis 10 and the lifting device 30 and are located below the corresponding damper (as Figure 3 shown), and the two ends of the support rod 22 are respectively connected to the side wall of the main chassis 10 and the corresponding lifting device 30 (the support rod 22 and the main chassis 10, the lifting device 30 are all in clearance fit and can rotate relative to each other, as Figure 3 shown). The lifting device 30 adopts a hydraulic telescopic rod, and the hydraulic telescopic rod is controlled by a hydraulic pump arranged in the inner cavity of the corresponding side of the main chassis 10 (the hydraulic pump and the hydraulic telescopic rod both adopt conventional structures and models in the art, and those skilled in the art can understand).

[0032] The traveling device includes a steering reducer housing 41, a steering reduction motor 42, a reducer gear set, a traveling wheel protection housing 43, a traveling wheel 44, and a hub motor (as Figure 4As shown in the figure, the steering reducer housing 41 is arranged at the bottom end of the lifting device 30 (i.e., the end of the hydraulic telescopic rod away from the corresponding shock absorber), and a steering reduction motor 42 and a steering reducer gear set are installed therein (the steering reducer gear set can adopt a conventional structure in the art and is arranged between the output end of the reduction motor 42 and the walking wheel protection housing 43). A Hall angle sensor is arranged inside the steering reducer housing 41. The upper end of the walking wheel protection housing 43 passes through the steering reducer housing 41 and is connected to the internal output stage of the reducer gear set. The lower end of the walking wheel protection housing 43 is provided with a walking rotating shaft, and the outer wall of the walking rotating shaft is sleeved with a walking wheel 44 (the walking wheel 44 is located inside the walking wheel protection housing 43, as Figure 4 shown). The hub motor is fixedly installed on the walking wheel protection housing 43, and its output end is connected to the walking rotating shaft.

[0033] The weeding component includes a rotating base 51, a three-axis robotic arm 52, a four-channel nozzle 53, a camera pan-tilt 54, a collection module 55, and a radar 56 (combined with Figure 1 and Figure 5 shown). The rotating base 51 is fixedly arranged in the middle of the bottom surface of the main chassis 10. One end of the three-axis robotic arm 52 is connected to the rotating base 51, and the other end is provided with a four-channel nozzle 53 (as Figure 5 shown). The four-channel nozzle 53 includes a nozzle front end 532, a nozzle rear end 531, and a liquid distribution module 533 (as Figure 6 shown). One side of the nozzle rear end 531 is connected to the end of the three-axis robotic arm 52 away from the main chassis 10, and a water inlet channel is opened inside it (as Figure 6 shown in (b)). The water inlet channel is communicated with the water tank arranged in the inner cavity of the main chassis 10, and a metering pump is arranged on their connecting pipe (as Figure 2 shown). The nozzle front end 532 is connected to the nozzle rear end 531 (by screws), and a mixing chamber is opened inside the nozzle front end 532 (as Figure 6 shown; a cross-shaped partition is arranged inside the mixing chamber, and an intercommunication port is arranged at its bottom). The end of the nozzle rear end 531 away from the three-axis robotic arm 52 is located inside the mixing chamber. Four liquid distribution modules 533 are respectively arranged on the outer wall of the nozzle front end 532 and corresponding to the end of the nozzle rear end 531 (as Figure 6 shown in (a)). A liquid medicine channel communicated with the mixing chamber is opened in the middle of the liquid distribution module 533 (as Figure 6 shown in (b)), and a solenoid valve 5330 is arranged on one side of the liquid distribution module 533. The sides of the liquid distribution module 533 away from the nozzle front end 532 are respectively communicated with the corresponding medicine boxes arranged in the main chassis 10 (that is, four medicine boxes are arranged in the inner cavity of the main chassis, as Figure 2As shown in the figure), and a metering pump is provided between them respectively. There are two camera pan-tilts 54, and they are symmetrically arranged on the bottom surface of the main chassis 10 on both sides of the rotating base 51. Acquisition modules 55 are respectively arranged on the camera pan-tilts 54 (the acquisition module 55 can adopt a binocular camera). There are two groups of radars 56, and they are respectively arranged in the middle of the front and rear sides of the main chassis 10 (as Figure 1 shown).

[0034] Embodiment 2: A weeding method in cotton fields, using the weeding robot as described in Embodiment 1, including: Step S1: Adjust the overall height of the main chassis according to the plant height of the cotton, that is, start the hydraulic telescopic rod to lift according to the plant height of the cotton, and ensure that the bottom surface of the main chassis 10 is located above the cotton plants; Control the weeding robot to travel in the cotton field. The specific control method is: First, use a multi-modal fusion system combining the Global Navigation Satellite System (GNSS) and the Inertial Measurement Unit (IMU) to obtain the pose of the robot: Define the state variables of the robot: ; In the formula: P = x , y , z T , representing the position in the navigation coordinate system; v = v x , v y , x z T , representing the speed in the navigation coordinate system; Q = Q w , Q x , Q y , Q z T , representing the quaternion attitude from the body coordinate system to the navigation coordinate system ( ); a = a x , a y , a z T , representing the zero bias of the accelerometer; b = b x , b y, ​​​​b z T , representing the gyroscope zero bias; The outputs of the accelerometer and gyroscope of the inertial measurement unit (IMU) are:

[0035] Where: a 0, w 0 respectively represent the true values; N a , N g respectively represent Gaussian white noise; Derive the state change through the inertial measurement unit:

[0036] Where: represents the transformation matrix from the body coordinate system (b-frame) to the navigation coordinate system (n-frame) (calculated from the quaternion Q ); g = 0 , 0 , -g T represents the gravity vector; N ba , N bg represent the random walk noise of the zero bias; Discretize using the first-order Euler method to obtain the state transition equation:

[0037] Then the Jacobian matrix F is (the specific block matrix includes the partial derivatives with respect to position, velocity, attitude, and zero bias):

[0038] Global Navigation Satellite System observation model:

[0039] Where: P GNSS , v GNSS respectively represent the position and velocity in the Global Navigation Satellite System coordinate system; n Z represents the observation noise; Perform covariance prediction calculation:

[0040] Where: represents the predicted error covariance matrix;​​ represents the error covariance matrix at the previous moment; Q k represents the process noise covariance (determined by IMU noise and bias random walk): Perform state update and covariance update:

[0041] In the formula: H represents the observation matrix; I represents the identity matrix; K represents the Kalman gain:

[0042] In the formula: R represents the GNSS observation noise covariance; The attitude quaternion is renormalized after update:

[0043] Pre-integrate the inertial measurement unit (IMU) in the time interval t k-1 , t k to reduce the computational load:

[0044] Then, perform path planning for the robot in the cotton field: Extract cotton plant rows through visual semantic segmentation and construct a directional graph structure G ( V , E ) where the node V is the key point of the row and the edge E weight includes row curvature and obstacle density; Given the starting point p s , the ending point p e and the intermediate control points c 1, c 2, obtain the curved path equation:

[0045] In the formula: ; Optimize the intermediate control points to avoid high-density obstacle areas; Finally, through the adaptive dynamic obstacle avoidance method, perform obstacle avoidance during the movement: The multi-sensor fusion system (including lidar, vision camera and ultrasonic sensor) carried on board perceives the road conditions ahead in real time. If an obstacle is detected, a method of fusing LiDAR point cloud clustering (such as the DBSCAN algorithm) and vision segmentation (such as YOLOv10) is used to construct an obstacle set; and dynamic obstacle avoidance is achieved based on the improved artificial potential field method (APF). The improved artificial potential field method is specifically as follows: Introduce a dynamic obstacle speed prediction and relative distance decay factor to construct a dynamic repulsive potential field function (effectively avoiding obstacle avoidance failure caused by the movement of obstacles in the cotton field):

[0046] In the formula: k rep represents the repulsive force coefficient, which is used to adjust the basic intensity of the repulsive potential field; represents the current position of the robot q and the position of the obstacle q obs the distance between; d 0 represents the influence distance threshold, which defines the range of the repulsive force of the obstacle on the robot; q goal represents the target position of the robot, that is, the end coordinate of the planned path; represents the attenuation coefficient; represents the running speed of the obstacle; represents the relative direction unit vector, that is, the relative direction between the robot and the obstacle; Dynamically adjust the attraction gain according to the environment, and then construct an adaptive gravitational potential field:

[0047] In the formula: represents the basic attraction gain coefficient; represents the adjustment factor; dis max represents the maximum environmental span, the largest distance scale in the entire environment; Adopt the velocity obstacle method to generate a velocity-related repulsive force F vel , and then complete the velocity potential field coupling:

[0048]

[0049] In the formula: k v represents the repulsive force adjustment coefficient, which is used to control the overall intensity of the velocity-related repulsive force; ek represents a small constant, which is used to prevent the denominator from being 0; vrobot , v obs respectively represent the speeds of the robot and the obstacle; When reasonably continuously approaching zero, apply a random disturbance force F rand to prevent local minima and thus jump out of the stagnation point:

[0050] In the formula: represents the disturbance intensity; e rand represents the random direction unit vector; Step S2: During the driving process, continuously detect the weeds in the cotton field in real time and generate an image with spectral information annotation, specifically: Use the improved YOLOv10 to detect the weeds in the cotton field. The improved YOLOv10 uses the Dysample upsampling technique to improve the Neck part in the yolov10 network, and at the same time introduces the RMTBlock to improve the Bottleneck structure in the C2F module; in addition, the improved YOLOv10 uses a multi-spectral feature fusion detection module, adopting a dual-channel input architecture of the visible light channel and the near-infrared channel, where the visible light channel is used to receive the RGB three-band image, and the near-infrared channel is used to receive the NIR single-band image, and uses the channel attention mechanism to achieve feature fusion:

[0051] In the formula: F fused represents the fused feature; represents the channel attention function; F RGB , F NIR respectively represent the visible light channel feature and the near-infrared channel feature; represents the feature addition operation; represents the backbone network feature extraction function; Then, classify the cotton plants and weeds through spectral difference discrimination, and discriminate through the dual thresholds of the vegetation indices of the normalized difference vegetation index (NDVI) and the enhanced difference vegetation index (EDVI):

[0052] In the formula: NIR represents the near-infrared band reflectance; R represents the red light band reflectance; B represents the blue light band reflectance; preset first threshold of the normalized difference vegetation index YNDVI-1 With the second threshold Y NDVI-2 , and the first threshold of the enhanced difference vegetation index Y EDVI-1 With the second threshold Y EDVI-2 (In this embodiment, Y NDVI-1 = 0.6, Y EDVI-1 = 0.3, Y NDVI-2 = 0.55, Y EDVI-2 = 0.35): When NDVI is greater than Y NDVI-1 And EDVI is less than Y EDVI-1 , it is determined as a cotton plant; when NDVI is greater than Y NDVI-2 And EDVI is greater than Y EDVI-2 , it is determined as a weed; in other cases, it is determined as a background factor or a non - biological obstacle (through the dual discrimination matrix of NDVI and EDVI, a complementary enhancement relationship is formed; not only ensuring the stable identification of the main vegetation - cotton, but also improving the monitoring accuracy and sensitivity of weeds in the cotton field, so as to quickly and accurately complete the identification and classification of cotton and weeds under complex lighting conditions and field interference conditions); Generate a four - channel annotation matrix by using the method of spectral annotation images ; for each detection area Ri, if the pixel point , then:

[0053] In the formula: c i Represents the class identifier; W , H Represents the spatial resolution (height, width) of the image; NDVI The value is normalized to [0, 1]; For the weed detection box in the two - dimensional image ( x c , y c , w , h ), its center point is:

[0054] Then the corresponding three - dimensional space coordinates are:

[0055] Wherein: ( c x , c y ) represents the optical center coordinates; f represents the lens focal length; B represents the binocular baseline distance; d represents the pixel value corresponding to the disparity map; represents the compensation coefficient.

[0056] Step S3: Use the image annotated with spectral information to perform precise positioning of the robotic arm, specifically: Establish the homogeneous transformation matrix from the camera coordinate system to the base coordinate system of the three-axis robotic arm:

[0057] Wherein: ( T x , T y , T z ) represents the camera mounting position offset; represents the camera mounting deflection angle; ( X cam , Y cam , Z cam ) represents the three-dimensional point coordinates obtained from the camera image; Correct the Z-axis coordinate of the base coordinate system of the three-axis robotic arm through terrain compensation:

[0058] Wherein: represents the ground height difference, obtained by a laser rangefinder; Judge the reachability of the three-axis robotic arm:

[0059] When D is greater than L 1 + L 2, it means that the three-axis robotic arm cannot reach, trigger an alarm, and start the walking device to move; among them, L 1 represents the length of the first link of the three-axis robotic arm, L 2 represents the length of the second link of the three-axis robotic arm; otherwise, it is determined that the three-axis robotic arm can reach; Obtain the joint angles of the first link and the second link q 1, q 2:

[0060] Control the three-axis robotic arm by controlling the joint angles; Meanwhile, the vertical displacement of the four-channel nozzle is .

[0061] Step S4: After the nozzle at the front end of the robotic arm reaches the specified position, start the metering pump of the corresponding classification module and simultaneously start the metering pump of the water inlet channel, and open the electromagnetic valve of the corresponding classification module to achieve accurate and quantitative spraying of the corresponding liquid medicine; After the four-channel nozzle at the front end of the robotic arm is accurately positioned at the preset coordinate point, the integrated motion control system sends control instructions to the execution unit through the CAN bus; at this time, the high-precision metering pump is started under the drive of the servo motor, and its flow control accuracy can reach ±0.5%; meanwhile, the piezoelectric solenoid valve with four-channel independent control is opened under the excitation of a 24V DC pulse signal, and the response time is less than 10ms. Under this control, the liquid medicine is accurately sprayed in the form of atomization from the fan-shaped nozzle through the 316L stainless steel flow channel under the system pressure of 0.2 - 0.5MPa, and the atomization particle size distribution is controlled within the range of 50 - 150μm to ensure that the liquid medicine evenly covers the target area.

[0062] The whole process is monitored in real time by the PLC, and the operation parameters are displayed through the HMI to achieve accurate pesticide application operations.

Claims

1. A weeding robot, characterized in that: It includes a main chassis, a shock absorption device, a lifting device, a traveling device and a weeding component. The outer walls at the four corners of the main chassis are respectively provided with a vertical lifting device through the shock absorption device, and the traveling device is arranged at the bottom end of the lifting device; the weeding component includes a rotating base, a three-axis robotic arm, a four-channel nozzle, a camera pan-tilt, a collection module and a radar. The rotating base is fixedly arranged in the middle of the bottom surface of the main chassis. One end of the three-axis robotic arm is connected to the rotating base, and the other end is provided with a four-channel nozzle. There are two camera pan-tilts, and they are symmetrically arranged on the bottom surface of the main chassis on both sides of the rotating base. The collection modules are respectively arranged on the camera pan-tilts. There are two groups of radars, and they are respectively arranged in the middle of the front and rear sides of the main chassis.

2. The weeding robot according to claim 1, wherein: The shock absorption device includes a shock absorption spring, a damper and a support rod. The shock absorption spring is fixedly installed outside the damper through a buckle, and the upper end of the damper is connected to the side wall of the main chassis, and the lower end is connected to the top of the corresponding lifting device; there are two support rods, and the two support rods are arranged in parallel. The two support rods are arranged between the main chassis and the lifting device and are located below the corresponding damper. The two ends of the support rod are respectively connected to the side wall of the main chassis and the corresponding lifting device.

3. The weeding robot according to claim 1 or 2, characterized in that: The lifting device adopts a hydraulic telescopic rod, and the hydraulic telescopic rod is controlled by a hydraulic pump arranged in the inner cavity of the corresponding side main chassis.

4. The weeding robot according to claim 2 or 3, characterized in that: The traveling device includes a steering reduction gearbox housing, a steering reduction motor, a reduction gear group, a traveling wheel protection housing, a traveling wheel and a hub motor. The steering reduction gearbox housing is arranged at the bottom end of the lifting device, and a steering reduction motor and a steering reduction gear group are installed inside it. A Hall angle sensor is arranged inside the steering reduction gearbox housing. The upper end of the traveling wheel protection housing passes through the steering reduction gearbox housing and is connected to the internal output stage of the reduction gear group. The lower end of the traveling wheel protection housing is provided with a traveling rotating shaft, and the traveling wheel is sleeved on the outer wall of the traveling rotating shaft. The hub motor is fixedly installed on the traveling wheel protection housing, and its output end is connected to the traveling rotating shaft.

5. The weeding robot according to claim 4, wherein: The four-channel nozzle includes a nozzle front end, a nozzle rear end and a liquid separation module. One side of the nozzle rear end is connected to the end of the three-axis robotic arm far from the main chassis, and a water inlet channel is opened inside it. The water inlet channel is communicated with a water tank arranged in the inner cavity of the main chassis, and a metering pump is arranged on their connecting pipe; the nozzle front end is connected to the nozzle rear end, and a mixing chamber is opened inside the nozzle front end. The end of the nozzle rear end far from the three-axis robotic arm is located inside the mixing chamber; four liquid separation modules are respectively arranged on the outer wall of the nozzle front end and corresponding to the end of the nozzle rear end. A liquid medicine channel communicated with the mixing chamber is opened in the middle of the liquid separation module, and an electromagnetic valve is arranged on one side of the liquid separation module. The far sides of the liquid separation modules from the nozzle front end are respectively communicated with the corresponding medicine boxes arranged in the main chassis, and metering pumps are respectively arranged between them.

6. The weeding method of a weeding robot according to any one of claims 1 to 5, characterized in that: Including: Step S1: Adjust the overall height of the main chassis according to the plant height of the cotton, and control the weeding robot to travel in the cotton field; Step S2: During the traveling process, detect the weeds in the cotton field in real time and generate an image marked with spectral information; Step S3: Use the image marked with spectral information to perform precise positioning of the robotic arm; Step S4: When the nozzle at the front end of the robotic arm reaches the specified position, control the metering pump of the corresponding classification module to start, and at the same time start the metering pump of the water inlet channel, and open the electromagnetic valve of the corresponding classification module to achieve precise and quantitative spraying of the corresponding liquid medicine.

7. The weeding method of a weeding robot according to claim 6, characterized in that: The specific steps of step S2 are as follows: Use the improved YOLOv10 to detect weeds in the cotton field. The improved YOLOv10 uses a multi-spectral feature fusion detection module and adopts a dual-channel input architecture of the visible light channel and the near-infrared channel. Among them, the visible light channel is used to receive the RGB three-band image, and the near-infrared channel is used to receive the NIR single-band image. The channel attention mechanism is used to achieve feature fusion: In the formula: F fused represents the fused feature; represents the channel attention function; F RGB and F NIR respectively represent the visible light channel feature and the near-infrared channel feature; represents the feature addition operation; represents the backbone network feature extraction function; Then, the classification of cotton plants and weeds is realized through spectral difference discrimination, and the discrimination is carried out by the double thresholds of the vegetation indices of the normalized difference vegetation index and the enhanced difference vegetation index: In the formula: NIR represents the reflectance in the near-infrared band; R represents the reflectance in the red light band; B represents the reflectance in the blue light band; Preset first threshold of normalized difference vegetation index Y NDVI-1 and second threshold Y NDVI-2 , as well as first threshold of enhanced difference vegetation index Y EDVI-1 and second threshold Y EDVI-2 : When NDVI is greater than Y NDVI-1 and EDVI is less than Y EDVI-1 it is determined as a cotton plant; when NDVI is greater than Y NDVI-2 and EDVI is greater than Y EDVI-2 it is determined as a weed; in other cases, it is determined as a background factor or a non-biological obstacle; Generate a four-channel annotation matrix by using the method of spectroscopically annotating images ; For each detection region Ri, if the pixel point , then: Wherein: c i represents a category identifier; W and H represent the spatial resolution of the image; NDVI The value is normalized to [0, 1]; For the weed detection box in the two-dimensional image ( x c , y c , w , h ), its center point is: Then the corresponding three-dimensional space coordinates are: Where: ( c x , c y ) represents the optical center coordinates; f represents the lens focal length; B represents the binocular baseline distance; d represents the corresponding pixel value of the disparity map; represents the compensation coefficient.

8. The weeding method of a weeding robot according to claim 7, characterized in that: The specific steps of step S3 are as follows: Establish a homogeneous transformation matrix from the camera coordinate system to the base coordinate system of the three-axis robotic arm: Where: ( T x , T y , T z ) represents the offset of the camera installation position; represents the deflection angle of the camera installation; ( X cam , Y cam , Z cam ) represents the three-dimensional point coordinates obtained by the camera image; Compensate for the terrain and correct the Z-axis coordinate of the base coordinate system of the three-axis robotic arm: In the formula: represents the ground height difference, which is obtained by a laser rangefinder; Judge the reachability of the three-axis robotic arm: When D is greater than L 1 + L 2, it means that the three-axis robotic arm cannot reach, triggering an alarm and starting the traveling device to move; Among them, L 1 represents the length of the first link of the three-axis robotic arm, L 2 represents the length of the second link of the three-axis robotic arm; conversely, it is determined that the three-axis robotic arm can reach; Obtain the joint angles of the first link and the second link respectively q 1、 q 2: Control the three-axis robotic arm by controlling the joint angles; Meanwhile, the vertical displacement of the four-channel nozzle is .

Images