Multi-axis mechanical arm precise insect suction control method based on low-power-consumption intelligent recognition

Abstract

The invention relates to the field of intelligent recognition and mechanical arm control, in particular to a multi-axis mechanical arm precise insect suction control method based on low-power-consumption intelligent recognition, and aims to solve the problems that in the existing multi-axis mechanical arm insect suction control process based on intelligent recognition, a multi-axis mechanical arm is high in power consumption in the operation process; in order to solve the problems of high-speed action execution and large-amplitude shaking at the execution tail end due to wrong frames, the invention provides a multi-axis mechanical arm precise insect suction control method based on low-power-consumption intelligent recognition, and the method comprises the following steps: a long-distance search stage: carrying out long-distance large-range search on a plant target; in the middle distance approaching stage, after the plant leaf target is searched, a tail end self-adaptive anti-shake suppression method is used for controlling the mechanical arm to move to keep aligning with the plant leaf target; in the short-distance adsorption stage, the ID of a single diseased plant leaf target is locked, a tail end self-adaptive anti-shake suppression method is used for controlling a mechanical arm to move, finely adjust and get close to the diseased plant leaf target, and adsorption is executed.

Description

Technical Field

[0001] This invention relates to the fields of intelligent recognition and robotic arm control, specifically to a method for precise insect suction control of a multi-axis robotic arm based on low-power intelligent recognition. Background Technology

[0002] With the continuous development of intelligent robot technology, multi-axis robotic arms have been widely used in industrial automation, medical assistance, and agricultural operations. In recent years, dedicated robotic arm systems for specific task scenarios (such as pest identification and removal) have gradually become a research hotspot. However, existing multi-axis robotic arm systems still have significant shortcomings in terms of low-power operation, real-time intelligent identification, and precise suction control, making it difficult to meet the comprehensive requirements for energy efficiency, response speed, and operational accuracy in field or mobile platforms.

[0003] A search revealed a multi-axis robotic arm collision detection system and method, and a robotic arm, with publication number CN116749196B, published on June 18, 2024. This patent achieves high-precision collision detection and obstacle avoidance control by setting current detection components at the joints of the robotic arm and combining them with electronic skin to obtain collision position and pressure information. However, this solution focuses on safety protection mechanisms and does not involve visual or intelligent recognition capabilities for target objects (such as pests), nor does it integrate a suction actuator, thus failing to achieve integrated "identification-positioning-suction" operations. Furthermore, its reliance on multi-sensor fusion and high-frequency data processing results in high overall power consumption, making it unsuitable for battery-powered or long-term field operations.

[0004] A search revealed a patent, CN120791792B, titled "Intelligent Planning Method for Motion Trajectory of Multi-Axis Robotic Arm," published on December 5, 2025. This patent proposes an intelligent path planning method based on dynamic environment modeling and obstacle trajectory prediction, improving the real-time adaptability of the robotic arm in complex environments. While this method possesses some environmental perception and decision-making capabilities, its core focus is on general trajectory optimization, lacking image recognition, target classification, or low-power inference design for small targets (such as insects). Furthermore, the system does not integrate an end-effector and its control logic, lacking closed-loop control support for "immediately executing the suction action after recognition," making it difficult to directly apply to specific tasks such as automatic removal of agricultural pests.

[0005] The aforementioned problems indicate that while existing technologies have made progress in robotic arm structure, motion control, and collision detection, they generally suffer from deficiencies in specific application scenarios (such as intelligent pest identification and removal), including a lack of low-power intelligent identification modules, the absence of integrated dedicated pest-sucking actuators, and the failure to achieve integrated recognition-control-execution coordination. Therefore, this invention provides a multi-axis robotic arm precision pest-sucking control method based on low-power intelligent identification. This method aims to integrate lightweight target recognition algorithms, low-power embedded processing architecture, and efficient pest-sucking actuators to achieve rapid pest identification, precise positioning, and energy-efficient automatic removal, thereby meeting the practical needs of intelligent and green operations in modern agriculture. Summary of the Invention

[0006] The purpose of this invention is to provide a precise insect-sucking control method for a multi-axis robotic arm based on low-power intelligent recognition, in order to solve the problems of narrow autonomous search range, lack of multi-dimensional spatial position information, and lack of multi-segment path control during the movement of the robotic arm in the existing intelligent recognition-based multi-axis robotic arm insect-sucking control process. These problems result in the robotic arm control process being unable to maintain a wide search range, accurately align with the target, and experience jitter during the movement of the robotic arm and the insect-sucking action.

[0007] To address the above problems, this invention provides a method for precise insect suction control of a multi-axis robotic arm based on low-power intelligent recognition, the method comprising: S1. Long-distance search phase: Conduct a long-distance, wide-area search for the plant target. S2. In the mid-range approach phase, after the target plant leaf is located, the robotic arm is controlled to keep aligned with the target plant leaf using an end-effector adaptive anti-shake suppression method. S3. In the close-range adsorption stage, the target ID of a single diseased plant leaf is locked, and the robotic arm is controlled by an end-effector adaptive anti-shake suppression method to fine-tune its movement to get close to the target diseased plant leaf and perform adsorption.

[0008] Furthermore, the long-distance search phase includes the following steps: S101. Initialization preparation stage: The multi-axis robotic arm is reset, the camera is started, and the plant image is obtained. S102, Target Search and Coarse Localization Stage: Determine whether a plant target has been detected. If no plant target has been detected, the robotic arm will move along the X-axis and R-axis to expand the search range and continue searching. S103. If a plant target is detected, the position and relative orientation of the plant target in the image are obtained, and the robotic arm moves toward the plant target synchronously. S104. Switching trigger stage: Using camera depth information as the switching trigger condition, during the approach to the plant target, the camera depth information is used to determine whether the current distance meets the preset distance. If the preset distance is met, the approach stops and the target is accurately located.

[0009] Furthermore, the mid-range approach phase includes the following steps: S201, Target Precision Positioning Stage: Based on depth information, the position, distance, and relative direction of the target are continuously output. The robotic arm makes small-range translations along the X-axis to adjust its horizontal position, fine-tunes the R-axis to align the suction cup with the target, and adjusts the Z-axis downwards according to the target distance. S202, Dynamic tracking phase: Based on depth information, the position, distance and relative direction of the target are continuously output to correct the pose of the robotic arm. When the target moves, the X-axis and R-axis of the robotic arm are simultaneously fine-tuned to always be aligned with the target. S203, Switching Trigger Stage: The switching trigger condition is based on the detected target imaging ratio of plant leaves. When the detected target imaging ratio is greater than or equal to a preset value, and the depth information reaches a preset value, the target locking and alignment stage begins.

[0010] Furthermore, the near-field adsorption stage includes the following steps: S301, Target Locking and Alignment Stage: Lock the plant leaf target ID based on image information. After locking the target, ignore other targets. Make fine adjustments to the X and R axes of the robotic arm to keep the center of the suction cup perfectly aligned with the center of the target. Slowly lower the Z axis to a position close to the plant leaf target. S302. During the adsorption phase, the suction cup is activated to generate negative pressure, adsorbing the target. The adsorption action is verified by feedback from the loop detection. S303, Reset and Release Phase: After successful adsorption, the robotic arm rises along the Z-axis, moves along the X-axis to the preset release area, maintains its current posture along the R-axis, closes the suction cup to release the target, and records the X-axis, R-axis, and Z-axis coordinate information of the robotic arm during the current adsorption action. S304, Loopback detection phase: If the locked plant leaf target ID is not detected, proceed to S2. If the target percentage is less than 40%, exit S3 and continue searching in phase S1 based on depth information.

[0011] Furthermore, the end-effector adaptive image stabilization suppression method includes the following steps: S10. Read the acceleration sensor signal and calculate the impact of the trolley disturbance on the robotic arm. : , in, The coupling coefficient between the mobile platform and the robotic arm reflects the proportion of acceleration transmitted from the mobile platform to the end effector of the robotic arm. S11. Read robotic arm data Convert the real-time coordinates of the camera and the real-time coordinates of the suction nozzle. ), calculate positioning error ; S12. Based on the error rate of change Perform blurring processing and adjust the intensity according to the magnitude of the error; S13. Based on the above fuzzy rules, using the formula: , in, Let be the real-time scaling factor, integral factor, and differential factor of the i-th axis at time t. These are the initial proportional, integral, and derivative coefficients of the fuzzy PID controller for the i-th axis, respectively. The PID parameter correction is calculated by fuzzy rule reasoning based on the real-time collected positioning error and error change rate. ,in, This is the mobile platform disturbance compensation term, and the disturbance term is directly proportional to the compensation term; S14. Decoupling of inter-axis coupling of the robotic arm: Using a weighted fusion method, the coupling interference during the movement of the multi-axis robotic arm is eliminated.

[0012] Furthermore, the weighted fusion method includes the following steps: S20. Calculate the PID control output torque of one axis in a multi-axis robotic arm. : , S21. Set the coupling weights for the remaining two axes of the multi-axis robotic arm; S22, According to the formula : , in, For the other axes of the multi-axis robotic arm, the calculations are as follows: It is converted into a motor drive signal to adjust the posture of the robotic arm.

[0013] Furthermore, the end-effector adaptive anti-shake suppression method is used in S2, the mid-distance approach phase and S3, the close-range adsorption phase to eliminate end-effector vibration and positioning deviation of the multi-axis robotic arm.

[0014] Furthermore, a multi-axis robotic arm precision insect-sucking control system based on low-power intelligent recognition includes: The decision control unit receives image data from the image sensing unit and calculates and outputs an electrical control signal; The motor drive unit receives the electrical control signal output by the decision control unit, converts the electrical control signal into a robotic arm drive signal, and controls the movement of the robotic arm. The PSO anti-shake monitoring unit receives drive signals and uses the control model to output control parameters adapted to the scene, thereby offsetting external motion interference and decoupling the coupling disturbances caused by the multi-axis robotic arm motion. An image sensing unit is used to acquire image information and depth information of plant leaf targets; The insect-sucking execution unit is an end effector that removes pests from plant leaves through suction action. The loopback monitoring unit is used to monitor whether the end-effector action is completed and to feed back the execution result to the decision control unit.

[0015] This invention provides a multi-axis robotic arm precision insect-sucking control method based on low-power intelligent recognition. This method enables autonomous searching and identification of defective leaves on target plants from multiple angles and over a wide range. It reduces computational load and provides energy-efficient, fuzzy observation of target locations, improving the accuracy of identifying defective leaves. Furthermore, it smoothly filters and processes data noise related to the target spatial information and imaging proportion of defective leaves, resulting in smoother multi-axis robotic arm movement and improved tracking stability and accuracy. Consequently, it stabilizes the execution speed of the robotic arm during insect sucking, resolving the issues of high power consumption, high-speed execution, and significant jitter at the end of the execution process caused by frame errors in multi-axis robotic arms. Attached Figure Description

[0016] Figure 1 Flowchart of a multi-axis robotic arm insect control method based on low-power intelligent recognition; Figure 2 Schematic diagram of the long-distance search phase method; Figure 3 Flowchart of the approach phase method Figure 4 Schematic diagram of the close-range adsorption stage method; Figure 5 Flowchart of the end-effector adaptive image stabilization suppression method; Figure 6 Flowchart of weighted fusion method; Figure 7 Schematic diagram of a multi-axis robotic arm insect-sucking control system based on low-power intelligent recognition. Detailed Implementation

[0017] To better understand the purpose, structure, and function of this invention, the following detailed description, in conjunction with the accompanying drawings, provides a precise insect-sucking control method for a multi-axis robotic arm based on low-power intelligent recognition.

[0018] like Figure 1The present invention provides a method for precise insect suction control of a multi-axis robotic arm based on low-power intelligent recognition. This method includes steps S1 to S3, which are described in detail below: Step S1, long-distance search stage: conduct a long-distance, large-scale search for the plant target.

[0019] It should be noted that in step S1, when the distance to the plant target is greater than the preset value during the long-distance search stage, the camera parameters are set to low frame rate and low resolution mode. Only the distance from the center point of the plant recognition box to the camera is used to calculate the depth and search for the plant target carrier.

[0020] In specific implementation, such as Figure 2 As shown, step S1 includes steps S101 to S104, which are described in detail below: S101. Initialization preparation stage: The multi-axis robotic arm is reset, the camera is started, and the plant image is obtained. Specifically, the multi-axis robotic arm resets to its initial pose with the X origin, Z safe height, and R initial angle, and the suction cup is closed; the camera starts and quickly scans the environment at low resolution and low frame rate to detect plant targets.

[0021] S102, Target Search and Coarse Localization Stage: Determine whether a plant target has been detected. If no plant target has been detected, the robotic arm will move along the X-axis and R-axis to expand the search range and continue searching. S103. If a plant target is detected, the position and relative orientation of the plant target in the image are obtained, and the robotic arm moves toward the plant target synchronously. S104. Switching trigger stage: Using camera depth information as the switching trigger condition, during the approach to the plant target, the camera depth information is used to determine whether the current distance meets the preset distance. If the preset distance is met, the approach stops and the target is accurately located.

[0022] In practice, the multi-axis robotic arm expands the search range by translating a large range of X-axis and rotating at multiple angles of R-axis, while maintaining a basic height on the Z-axis to assist scanning. The camera detects the target simultaneously. If no plant target is found, the multi-axis robotic arm continues to expand the search range. If a target is detected, the robotic arm gradually moves closer until the camera determines that the distance to the plant target is less than the preset value in step S1, triggering the switching condition.

[0023] Step S2, the mid-range approach stage: After the target plant leaf is located, the robotic arm is controlled to keep aligned with the target plant leaf using the end-effector adaptive anti-shake suppression method.

[0024] It should be noted that in step S2, when the target depth information of the plant leaf is less than the preset distance, the camera parameters are adjusted to high frame rate and high resolution mode; the average distance from the 5*5 pixel position of the center area of ​​the target box to the camera is calculated using median filtering to calculate the average distance between 30% and 70% of the target depth; based on the average distance between 30% and 70%, an exponential smoothing method is introduced; and the depth value of the previous frame is superimposed on 30% of the current average depth value to obtain the current frame rate depth value; during the movement of the multi-axis robotic arm, the end-effector adaptive image stabilization suppression method is called to stabilize the imaging and depth information of the end-effector camera of the multi-axis robotic arm.

[0025] In specific implementation, such as Figure 3 As shown, step S2 includes steps S201 to S203, which are described in detail below: Step S201, Target Precision Positioning Stage: Based on the depth information, the position, distance and relative direction of the target are continuously output. The robot arm moves slightly along the X-axis to adjust its horizontal position, and makes minor adjustments along the R-axis to align the suction cup with the target. The Z-axis is adjusted downwards according to the target distance. Step S202, Dynamic Tracking Stage: Based on depth information, the position, distance and relative direction of the target are continuously output to correct the pose of the robotic arm. When the target moves, the X-axis and R-axis of the robotic arm are simultaneously fine-tuned to always be aligned with the target. Step S203: Switching trigger stage. The target proportion of the detected plant leaves is used as the switching trigger condition. When the detected target proportion is greater than or equal to the preset value, and the depth information reaches the preset value, the target locking and alignment stage is entered.

[0026] In practice, the multi-axis robotic arm switches to the mid-range approach stage, and the X-axis and the rotational axis R-axis are finely adjusted to align with the target. The Z-axis adjusts the height synchronously. The camera measures the depth information and the imaging ratio in real time. If the target distance is greater than the preset distance, it returns to the long-range search stage. If the imaging ratio is less than the preset target value, it continues in the mid-range approach stage. When the imaging ratio of the plant leaf target is greater than or equal to the preset target value, it switches to the next stage.

[0027] S3. In the close-range adsorption stage, the target ID of a single plant leaf is locked, and the robotic arm is controlled by an end-effector adaptive anti-shake suppression method to fine-tune its movement to get close to the target plant leaf and perform adsorption.

[0028] It should be noted that in step S3, during the close-range adsorption stage, when the proportion of the plant leaf target image is greater than or equal to the preset proportion value, the camera parameters are set to the highest detection frame rate and the highest resolution. In the close-range adsorption stage, target locking and tracking are introduced. During the target locking and tracking process, only the plant leaf target that enters for the first time with a proportion greater than the preset proportion value is processed. The single plant leaf target ID is locked instead of the category ID. This process can effectively reduce the system's computational load and prevent interference from other targets during the movement of the multi-axis robotic arm.

[0029] Furthermore, in the close-range adsorption stage, the average distance from the center area of ​​the plant leaf target frame (i.e., the 5*5 pixel area) to the camera is first calculated, and then the average distance between 30% and 70% is calculated using median filtering.

[0030] Furthermore, after calculating the average distance between 30% and 70% using median filtering, a sliding window filter is introduced to cache the depth information of the most recent 5 frames, and the average depth of the most recent 5 frames is calculated. Furthermore, based on the average depth of the most recent 5 frames, an exponential smoothing method is introduced to calculate 30% of the current average depth, and then 70% of the previous frame image is superimposed to obtain the current frame rate depth value. The end-effector adaptive stabilization suppression method is called to suppress the end-effector jitter interference caused by frequent movement during the approach of the multi-axis robotic arm and prevent motion jumps caused by frequent stage switching.

[0031] Specifically, the exponential smoothing method uses quadratic exponential smoothing to capture the linear trend of the target distance, which is constantly approaching. Quadratic exponential smoothing can capture the linear trend of distance, filter noise, and follow the target's movement in real time, making it more suitable for dynamic distance tracking than single-stage smoothing. This is based on the single-stage exponential smoothing formula. : , in, No. The first smoothing value, The smoothing coefficient ranges from 0 to 1. An example parameter selection is provided. =0.3, achieving trend tracking while satisfying smoothness. No. The raw distance value obtained by the sensor during the second sampling.

[0032] According to the quadratic exponential smoothing formula : , , in, No. Secondary smoothing value No. The distance after smoothing is used as the final output value.

[0033] In specific implementation, such as Figure 4 As shown, step S3 includes steps S301 to S304, which are described in detail below: S301, Target Locking and Alignment Stage: Lock the plant leaf target ID based on image information. After locking the target, ignore other targets. Make fine adjustments to the X and R axes of the robotic arm to keep the center of the suction cup perfectly aligned with the center of the target. Slowly lower the Z axis to a position close to the plant leaf target. S302. During the adsorption phase, the suction cup is activated to generate negative pressure, adsorbing the target. The adsorption action is verified by feedback from the loop detection. S303, Reset and Release Phase: After successful adsorption, the robotic arm rises along the Z-axis, moves along the X-axis to the preset release area, maintains its current posture along the R-axis, and closes the suction cup to release the target. S304, loop closure detection phase: If the locked plant leaf target ID is not detected, proceed to S2. If the target percentage is less than 40%, exit S3 and continue the search in phase S1 based on the depth information.

[0034] In practice, the multi-axis robotic arm fine-tunes the X-axis and rotation axis to align with the target, and the Z-axis continuously approaches the plant leaf target. When the plant leaf target image coverage reaches 60%, it indicates successful verification. The position coefficients of the multi-axis robotic arm's X-axis, Z-axis, and R-axis at this time are recorded in the search list. Furthermore, after recording is complete, turn on the insect suction device at the end of the multi-axis robotic arm, start a 10-second timer, and turn off the suction cup after the suction action ends.

[0035] Furthermore, after completing the adsorption action, the multi-axis robotic arm moves to a distance greater than 30cm from the target, while simultaneously accumulating the count of the trematodes; if the target disappears during the adsorption action or the imaging ratio does not meet 40%, it will return to the long-distance search stage or the medium-distance approach stage based on the depth information.

[0036] Furthermore, after the count is updated, if the count is less than the preset number, the system returns to the long-distance search stage to continue searching the current plant leaves. During this process, targets that have already undergone adsorption operations will no longer be searched. If the count is greater than the preset number, the system switches to the next plant, and at the same time, the robotic arm and camera are reset, and the count is cleared to zero.

[0037] like Figure 5 The diagram shows that an end-effector adaptive anti-shake suppression method is used to eliminate end-effector vibration and positioning deviation in the mid-range approach and close-range adsorption phases of a multi-axis robotic arm.

[0038] It is necessary to explain the end-effector adaptive anti-shake suppression method as follows: the fitness function optimization objective is defined as follows: combining the characteristics of the trematode scenario, a load change penalty term is introduced to ensure that the parameters take into account positioning accuracy, control stability and load adaptability. , in, This is a fitness value, in the context of trematodes. A smaller value is better, as it indicates better positioning accuracy and control stability. According to the formula : , in, , , Used as reference coordinates for diseased leaves. , , This provides the real-time coordinates of the suction port position. (t) represents the vibration acceleration at the nozzle tip, used to reflect the vibration of the robotic arm itself. (t) represents the acceleration of the trolley, reflecting the disturbance caused by the moving base; For example, =0.3 is the coupling coefficient of the robotic arm. The coupling coefficient of the robotic arm is an assumed value, which reflects the proportion of the acceleration of the moving base transmitted to the end of the robotic arm. For example, =0.5、 =0.3、 =0.2, which is the weighting coefficient to suppress end jitter while prioritizing positioning accuracy; T=10s is the optimization period. The optimization period is an assumed value and covers all scenarios of start-stop, constant speed, bumping and multi-axis robotic arm movement of the mobile base.

[0039] Furthermore, the particle position vector : ; Specifically, The PSO method randomly generates a fuzzy PID quantization scale within a preset range of 10 to 100, and finally optimizes it into a fuzzy PID quantization scale through the PSO method. In actual control, first use Calculate the error quantization value. Calculate the quantized value of the error rate of change, and determine the current error level based on the quantized value of the error and the quantized value of the error rate of change. Furthermore, the PID parameter correction is output based on the fuzzy rules. Responsible for calibrating the correction amount to an appropriate size and the baseline value given by the PSO. The real-time PID control force is obtained by superimposing the values ​​to stabilize the robotic arm.

[0040] in, Positioning error of multi-axis robotic arm axis x fuzzy domain quantization coefficients; : Rate of change of error on axis x : , Quantization coefficients of the fuzzy universe of discourse with a sampling period of 10ms. Δ is the PID parameter correction amount for the x-axis of the multi-axis robotic arm. , △ , △ Quantization coefficients of the fuzzy universe of discourse The initial proportional, integral, and derivative coefficients of the x-axis fuzzy PID controller.

[0041] Furthermore, referring to the above embodiments, the multi-axis robotic arm independently optimizes each axis and outputs the initial proportional, integral, and derivative coefficients of the fuzzy PID.

[0042] According to the particle velocity update formula When it is necessary to speed up the convergence, k should be reduced appropriately; when it is necessary to enhance the global optimization capability, k should be increased appropriately.

[0043] , in, , For particles exist , The velocity vector at any given moment; This is the convergence factor, used to avoid excessively rapid convergence. When faster convergence is needed: appropriately reduce k; when global optimization capability is needed: appropriately increase k. ; Furthermore, inertia weight ; For example, =0.8, =0.4, This represents the total number of iterations. : The optimal position of particle i.

[0044] For particles The optimal position of the population, and the particle position update formula. : , Furthermore, based on the local optimal avoidance rule, according to the formula... : , Set minimum speed threshold If | |less than| After three iterations, if the particle is determined to be trapped in a local optimum, the particle is restarted. : , After iterating 50 times using the PSO method, a fuzzy rule table for the multi-axis robotic arm is output to facilitate subsequent online control calls.

[0045] In practical implementation, the end-effector adaptive image stabilization suppression method mainly includes steps S10 to S14: S10. Read the acceleration sensor signal and calculate the impact of the trolley disturbance on the robotic arm. : , in, The coupling coefficient between the mobile platform and the robotic arm reflects the proportion of acceleration transmitted from the mobile platform to the end effector of the robotic arm. S11. Read robotic arm data Convert the real-time coordinates of the camera and the real-time coordinates of the suction nozzle. Calculate positioning error ; S12. Based on the error rate of change : , Will The blurring process is performed, and the intensity is adjusted according to the magnitude of the error.

[0046] Specifically, if the error is large and the rate of change of the error is large: increase the proportional coefficient P, decrease the integral coefficient I, and increase the derivative coefficient D; quickly reduce the error and suppress jitter; if the error is moderate and the rate of change of the error is moderate, maintain the proportional coefficient P, fine-tune the integral coefficient I, and maintain the derivative coefficient D; smoothly adjust the attitude; if the error is small and the rate of change of the error is small, decrease the proportional coefficient P, increase the integral coefficient I, and decrease the derivative coefficient D to eliminate residual error and avoid overshoot; if a disturbance of the trolley is detected... The scale factor P is increased to offset the effects of base disturbance.

[0047] S13. Based on the above fuzzy rules, using the formula: , in, Let be the real-time scaling factor, integral factor, and differential factor of the i-th axis at time t. These are the initial proportional, integral, and derivative coefficients of the fuzzy PID controller for the i-th axis, respectively. The PID parameter correction is calculated by fuzzy rule reasoning based on the real-time collected positioning error and error change rate. ; These represent the i-th axis (i=1,2,3), corresponding to the three motion axes of the multi-axis robotic arm. Real-time proportional coefficient, integral coefficient, and differential coefficient at time t; where The sampling time for the PLC is 10ms.

[0048] The initial proportional, integral, and derivative coefficients of the i-th axis fuzzy PID controller are obtained through simulation optimization using the improved PSO algorithm. These are the proportional, integral, and differential coefficient corrections for the i-th axis at time t, respectively, obtained through fuzzy rule reasoning based on the real-time collected positioning error and error change rate.

[0049] Specifically, the reasoning logic is that the larger the error, the stronger the proportional correction; the faster the rate of change of the error, the stronger the differential correction. This is the disturbance compensation term for the car; the more severe the disturbance, the greater the compensation.

[0050] S14. Decoupling of inter-axis coupling of the robotic arm: Using a weighted fusion method, the coupling interference during the movement of the multi-axis robotic arm is eliminated.

[0051] Specifically, inter-axis coupling decoupling refers to the elimination of coupling interference that exists in the axis movement of a multi-axis robotic arm through weighted fusion.

[0052] In practice, the weighted fusion method mainly includes steps S20 to S14: S20. Calculate the PID control output torque of one axis in a multi-axis robotic arm. ; For example, taking a multi-axis robotic arm with two axes as an example, according to the formula... : , in, The coupling weights of multi-axis robot arm axis 1 and multi-axis robot arm axis 3 to multi-axis robot arm axis 2 are as follows: =0.7、 =0.2、 =0.1; S21. Set the coupling weights for the remaining two axes of the multi-axis robotic arm; S22, According to the formula : , Calculate the independent PID output torque of a multi-axis robotic arm ,Will (t), (t), (t) is converted into a servo motor drive signal to drive the multi-axis robotic arm to adjust its posture.

[0053] When the nozzle positioning error Less than or equal to a preset threshold and end jitter acceleration When (t) is less than or equal to the preset threshold and stabilizes for a preset time, the PLC triggers the suction nozzle negative pressure switch to complete the suction of insects.

[0054] In addition to the above-described method embodiments, this invention also provides a multi-axis robotic arm precision insect suction control system based on low-power intelligent recognition. The system includes: a decision control unit 901, which receives image data from an image sensing unit 905 and calculates and outputs an electrical control signal. Specifically, the decision control unit 901 is connected to the motor drive unit 902 via a data line, and the decision control unit is deployed on a PLC controller or other controller with execution logic.

[0055] The motor drive unit 902 receives the electrical control signal output by the decision control unit 901, converts the electrical control signal into a robotic arm drive signal, and controls the movement of the robotic arm. Specifically, the motor drive unit 902 is connected to the PSO anti-shake monitoring unit 903 via a data bus. The motor drive unit 902 is mainly used to control the movement of the multi-axis robotic arm motor.

[0056] PSO anti-shake monitoring unit 903 receives drive signals and uses the control model to output control parameters adapted to the scene, thereby offsetting external motion interference and decoupling coupling disturbances caused by the motion of the multi-axis robotic arm. Specifically, the PSO anti-shake monitoring unit 903 integrates an end-effector adaptive anti-shake suppression method to monitor the execution jitter of the multi-axis robotic arm in real time; Image sensing unit 905, the image sensing unit is used to acquire image information and depth information of plant leaf target; Specifically, the image sensing unit 905 is located at the end of the multi-axis robotic arm, equipped with a camera and connected to the decision control unit 901 via a data bus to transmit image information in real time.

[0057] The insect-sucking execution unit 906 is an end effector that removes pests from plant leaves through suction action. Specifically, the insect-absorbing execution unit 906 is located at the end of the multi-axis robotic arm and uses an insect-absorbing device to absorb pests from plant leaves.

[0058] The loopback monitoring unit 907 is used to monitor whether the end-effector action is completed and to feed back the execution result to the decision control unit 901.

[0059] Specifically, the loop monitoring unit is connected to the data bus of the trematode execution unit 906 to provide feedback on the completion status of the execution action.

[0060] It is understood that the present invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1. A method for precise insect suction control of a multi-axis robotic arm based on low-power intelligent recognition, characterized in that, The method includes: S1. Long-distance search phase: Conduct a long-distance, wide-area search for the plant target. S2. In the mid-range approach phase, after the target plant leaf is located, the robotic arm is controlled to keep aligned with the target plant leaf using an end-effector adaptive anti-shake suppression method. S3. In the close-range adsorption stage, the target ID of a single diseased plant leaf is locked, and the robotic arm is controlled by an end-effector adaptive anti-shake suppression method to fine-tune its movement to get close to the target diseased plant leaf and perform adsorption.

2. The method for precise insect suction control of a multi-axis robotic arm based on low-power intelligent recognition according to claim 1, characterized in that, The long-distance search phase includes the following steps: S101. Initialization preparation stage: The multi-axis robotic arm is reset, the camera is started, and the plant image is obtained. S102, Target Search and Coarse Localization Stage: Determine whether a plant target has been detected. If no plant target has been detected, the robotic arm will move along the X-axis and R-axis to expand the search range and continue searching. S103. If a plant target is detected, the position and relative orientation of the plant target in the image are obtained, and the robotic arm moves toward the plant target synchronously. S104. Switching trigger stage: Using camera depth information as the switching trigger condition, during the approach to the plant target, the camera depth information is used to determine whether the current distance meets the preset distance. If the preset distance is met, the approach stops and the target is accurately located.

3. The method for precise insect suction control of a multi-axis robotic arm based on low-power intelligent recognition according to claim 1, characterized in that, The mid-range approach phase includes the following steps: S201, Target Precision Positioning Stage: Based on depth information, the position, distance, and relative direction of the target are continuously output. The robotic arm's X-axis is slightly translated to adjust its horizontal position, the R-axis is finely adjusted to align the suction cup with the target, and the Z-axis is adjusted downwards according to the target distance. S202, Dynamic tracking phase: Based on depth information, the position, distance and relative direction of the target are continuously output to correct the pose of the robotic arm. When the target moves, the X-axis and R-axis of the robotic arm are simultaneously fine-tuned to always be aligned with the target. S203, Switching Trigger Stage: The switching trigger condition is based on the detected target imaging ratio of plant leaves. When the detected target imaging ratio is greater than or equal to a preset value, and the depth information reaches a preset value, the target locking and alignment stage begins.

4. The method for precise insect suction control of a multi-axis robotic arm based on low-power intelligent recognition according to claim 1, characterized in that, The near-field adsorption stage includes the following steps: S301, Target Locking and Alignment Stage: Lock the plant leaf target ID based on image information. After locking the target, ignore other targets. Make fine adjustments to the X and R axes of the robotic arm to keep the center of the suction cup completely aligned with the center of the target. Slowly descend the Z axis to a position close to the plant leaf target. S302. During the adsorption phase, the suction cup is activated to generate negative pressure, adsorbing the target. The adsorption action is verified by feedback from the loop detection. S303, Reset and Release Phase: After successful adsorption, the robotic arm rises along the Z-axis, moves along the X-axis to the preset release area, maintains its current posture along the R-axis, closes the suction cup to release the target, and records the X-axis, R-axis, and Z-axis coordinate information of the robotic arm during the current adsorption action. S304, Loopback detection phase: If the locked plant leaf target ID is not detected, proceed to S2. If the target percentage is less than 40%, exit S3 and continue searching in phase S1 based on depth information.

5. The method for precise insect suction control of a multi-axis robotic arm based on low-power intelligent recognition according to claim 1, characterized in that, The end-effector adaptive image stabilization suppression method includes the following steps: S10. Read the acceleration sensor signal and calculate the impact of the trolley disturbance on the robotic arm. : ， in, The coupling coefficient between the mobile platform and the robotic arm reflects the proportion of acceleration transmitted from the mobile platform to the end effector of the robotic arm. S11. Read robotic arm data Convert the real-time coordinates of the camera and the real-time coordinates of the suction nozzle. Calculate positioning error ; S12. Based on the error rate of change Perform blurring processing and adjust the intensity according to the magnitude of the error; S13. Based on the above fuzzy rules, using the formula: ， in, Let be the real-time scaling factor, integral factor, and differential factor of the i-th axis at time t. These are the initial proportional, integral, and derivative coefficients of the fuzzy PID controller for the i-th axis, respectively. The PID parameter correction is calculated by fuzzy rule reasoning based on the real-time collected positioning error and error change rate. ,in, This is the mobile platform disturbance compensation term, and the disturbance term is directly proportional to the compensation term; S14. Decoupling of inter-axis coupling of the robotic arm: Using a weighted fusion method, the coupling interference during the movement of the multi-axis robotic arm is eliminated.

6. The method for precise insect suction control of a multi-axis robotic arm based on low-power intelligent recognition according to claim 5, characterized in that, The weighted fusion method includes the following steps: S20. Calculate the PID control output torque of one axis in a multi-axis robotic arm. : ， S21. Set the coupling weights for the remaining two axes of the multi-axis robotic arm; S22, According to the formula : ， in, For the other axes of the multi-axis robotic arm, the calculations are as follows: It is converted into a motor drive signal to adjust the posture of the robotic arm.

7. The method for precise insect suction control of a multi-axis robotic arm based on low-power intelligent recognition according to claim 5, characterized in that, The end-effector adaptive anti-shake suppression method is used in the S2, mid-distance approach phase and the S3, close-range adsorption phase to eliminate end-effector vibration and positioning deviation of the multi-axis robotic arm.

8. A multi-axis robotic arm precision insect suction control system based on low-power intelligent recognition, comprising: The decision control unit receives image data from the image sensing unit and calculates and outputs an electrical control signal; The motor drive unit receives the electrical control signal output by the decision control unit, converts the electrical control signal into a robotic arm drive signal, and controls the movement of the robotic arm. The PSO anti-shake monitoring unit receives drive signals and uses the control model to output control parameters adapted to the scene, thereby offsetting external motion interference and decoupling the coupling disturbances caused by the multi-axis robotic arm motion. An image sensing unit is used to acquire image information and depth information of plant leaf targets; The insect-sucking execution unit is an end effector that removes pests from plant leaves through suction action. The loopback monitoring unit is used to monitor whether the end-effector action is completed and to feed back the execution result to the decision control unit.

Images