Visual guidance automatic sorting and deviation rectifying method and system

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a vision-guided automatic sorting and deviation rectifying method and system.The vision-guided automatic sorting and deviation rectifying system comprises a vision parameter processing module, a sorting strategy decision module and a deviation rectifying execution control module, a real-time judgment mechanism of dynamic and static resolution differences is established by setting a resolution collaborative correction end, and the deviation rectifying execution control module is established; a self-adaptive interpolation algorithm is adopted to adjust visual feature data, so that the system can eliminate imaging resolution deviation caused by ambient light and transmission vibration, the acquisition accuracy of material surface texture features is ensured, sorting identification errors caused by projection chromatic aberration are avoided, the environmental adaptability of the visual guidance system is improved, and the visual guidance system is more accurate in image quality. The position and posture coupling analysis end is arranged, the dynamic matching degree of the position and posture offset of the materials and the motion trail of the mechanical arm is monitored in real time, the grabbing angle offset is automatically corrected through a space coordinate compensation mechanism, and it is guaranteed that the relative space position of an end effector and the materials is in the optimal grabbing state all the time.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a vision-guided automatic sorting and deviation correction method and system. Background Art

[0002] Automatic sorting is one of the necessary facilities for advanced distribution centers. It has high sorting efficiency. Automatic sorting machines are a key factor in improving logistics distribution efficiency and are an automatic sorting system widely used in logistics centers. Automatic deviation correction is suitable for correcting the deviation of coils and belts in the production line. Its core components include deviation correction controllers, sensors and actuators. It is divided into two categories: DJ series deviation correction machines and passive hydraulic deviation correction machines. It detects the material position and adjusts the deviation through photoelectric sensors. It can adapt to edge detection and line following modes and is widely used in printing, manufacturing and other fields.

[0003] In the automatic sorting link of existing packaging production lines, the equipped visual inspection system has defects when identifying dynamically transmitted materials: due to the frequent vibration and displacement of materials during the operation of the sorting line and interference from changes in ambient light, the surface feature data of the materials collected by the visual sensor is dissociated from the actual physical properties, resulting in a decrease in the accuracy of material identification; at the same time, the motion trajectory planning of the sorting robot arm relies on a preset parameter library, and is unable to perceive the displacement of the grasping point caused by changes in the material stacking height in real time, resulting in a high grasping failure rate of the vacuum suction cup and gripper; when sorting mixed materials of multiple specifications, the system lacks the ability to adaptively analyze the surface texture and reflective characteristics of the material. The use of unified exposure parameters causes highly reflective materials to be overexposed and dark materials to be undersampled, resulting in a high rate of missed detection of key features; in addition, the traditional correction mechanism relies solely on the closed-loop positioning feedback of the robot arm, and has not established a dynamic coupling model of the material posture and transmission line speed, resulting in delayed correction response and accumulated positioning errors in high-speed sorting scenarios, which is far higher than the process requirements of precision sorting.

[0004] Therefore, a vision-guided automatic sorting and correction method and system are proposed to solve the above problems. Summary of the Invention

[0005] (1) Technical problems solved In view of the shortcomings of the existing technology, the present invention provides a vision-guided automatic sorting and correction method and system, which solves the problems raised in the above background technology.

[0006] (2) Technical solution To achieve the above objectives, the present invention provides the following technical solutions: a vision-guided automatic sorting and deviation correction method and system, the method comprising: S1. Collect visual feature data of target materials and environmental parameter data of sorting scenes; S2. performing adaptive processing of the visual feature data based on the visual feature data and the environmental parameter data; S3, generating adapted visual feature data according to the resolution processing result; S4, combining visual feature data and environmental parameter data to construct fused feature data; S5. Match sorting strategy types based on the fused feature data to generate target sorting strategy identification data; S6. Calling a sorting program according to the target sorting strategy identification data; S7, extracting the position offset parameters of the material; S8, generating the robot arm motion trajectory control instruction to drive the actuator to grab the material; S9, real-time monitoring of position offset during the grasping process and dynamic compensation; S10. Complete accurate material sorting and positioning correction.

[0007] Preferably, the step S1 specifically includes: S11. Collect high-resolution texture, geometric profile, and reflectivity data of the material surface through a multispectral industrial camera to generate visual feature data; S12. Collect the parameters of the sorting line transmission speed, ambient light intensity, and material stacking height through the laser radar and depth sensor to generate environmental parameter data.

[0008] Preferably, the step S2 specifically includes: S21. Importing visual feature data and environmental parameter data into the sorting control platform; S22. Adopt an adaptive search algorithm to retrieve data sampling frequency by resolution keyword, and generate material static resolution data and scene dynamic resolution data, the unit of which is pixels per millimeter.

[0009] Preferably, the step S3 specifically includes: S31. Compare the numerical differences between the static resolution data of the material and the dynamic resolution data of the scene; S32: If the difference is less than a preset threshold, it is determined to be consistent; otherwise, it is determined to be inconsistent.

[0010] Preferably, the step S4 specifically includes: S41. When it is determined that there is inconsistency, a bilinear interpolation algorithm is used to adjust the resolution of the visual feature data to be consistent with the scene dynamic resolution data; S42. Generate material visual feature correction data that retains original texture features.

[0011] Preferably, the step S5 specifically includes: S51, aligning the corrected visual feature data and the environmental parameter data according to spatial coordinates; S52. Construct a fusion feature matrix including material surface features, environmental interference factors and posture parameters.

[0012] Preferably, the step S6 specifically includes: S61. Establish a standard feature library for sorting strategies, including benchmark feature matrices corresponding to nine sorting strategies; S62, using a bionic optimization algorithm to match the fusion feature matrix with the reference feature matrix: S621, initializing the position and number of iterations of the sorting strategy search population; S622, in the exploration phase, simulate the population to randomly search for target features, update the position and retain the better solution; S623. Simulate local fine search of the population in the development phase, calculate feature fitness and replace the suboptimal solution. The update formula is: in For the i-th population in dimension The new location, For the current dimension location, is a random number in the interval [0,1], is the upper limit of the search space, is the lower limit of the search space, is the current iteration number; S624: Output the optimal matching strategy type after reaching the maximum number of iterations; S63: Generate target sorting strategy identification data.

[0013] Preferably, the step S9 specifically includes: S91. Detect the horizontal offset and rotation angle of the material during the grasping process through a real-time visual tracking system; S92. Use the PID control algorithm to generate a correction instruction including the X, Y, and Z axis compensation amounts and the rotation compensation angle. The control formula is: in is the deviation correction control quantity at time t, is the pose error at time t, is the proportional gain coefficient, is the integral gain coefficient, is the differential gain coefficient, For the accumulation of historical errors, is the error change rate.

[0014] Preferably, the step S10 specifically includes: S101, converting the correction instruction into a control signal for the motor of the robot arm joint; S102, performing adaptive grasping by using a vacuum suction cup or a flexible gripper of an end effector; S103. Perform a secondary visual check when placing the material at the target workstation to ensure that the positioning error is less than 0.1 mm.

[0015] Preferably, the system comprises: Visual parameter processing module: includes multispectral imaging unit, environmental perception unit, resolution analysis unit and feature correction unit; Sorting strategy decision module: includes feature fusion unit, strategy feature library storage unit and bionic algorithm matching unit; Execution correction control module: includes trajectory planning unit, real-time correction unit and positioning verification unit.

[0016] (3) Beneficial effects Compared with the prior art, the present invention provides a vision-guided automatic sorting and deviation correction method and system, which has the following beneficial effects: 1. In the present invention, by setting up a resolution collaborative correction terminal, a real-time judgment mechanism for dynamic and static resolution differences is established during the dynamic sorting process of materials, and an adaptive interpolation algorithm is used to adjust the visual feature data, so that the system can eliminate the imaging resolution deviation caused by ambient light and transmission vibration, ensure the accuracy of collecting material surface texture features, avoid sorting and recognition errors caused by projection chromatic aberration, and improve the environmental adaptability of the visual guidance system.

[0017] 2. In the present invention, by setting up a posture coupling analysis terminal, when the actuator grasps the material, the dynamic matching degree between the posture offset of the material and the motion trajectory of the robot arm is monitored in real time, and the grasping angle offset is automatically corrected through the spatial coordinate compensation mechanism, so that the system can prevent the positioning offset of the material during high-speed transmission, and trigger an immediate correction program when a posture abnormality is detected, to ensure that the relative spatial position of the end effector and the material is always in the optimal grasping state.

[0018] 3. In the present invention, by setting a sorting strategy optimization end, when sorting mixed materials of multiple specifications, the surface features of the materials are intelligently matched with the preset strategy library through the bionic optimization algorithm, and the material categories with different textures and reflective characteristics are automatically identified and partition decision instructions are generated, so that the system can realize differentiated hierarchical processing of materials with complex surfaces, avoid sorting path planning errors caused by feature mismatching, and enhance the system's robustness in processing special-shaped materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a flow chart of the vision-guided automatic sorting and deviation correction method of the present invention.

[0020] Figure 2This is the architecture diagram of the vision-guided automatic sorting and correction system of the present invention. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0022] See also Figure 1-Figure 2 The vision-guided automatic sorting and deviation correction method and system include: S1. Collect visual feature data of target materials and environmental parameter data of sorting scenes; S2. performing adaptive processing of the visual feature data based on the visual feature data and the environmental parameter data; S3, generating adapted visual feature data according to the resolution processing result; S4, combining visual feature data and environmental parameter data to construct fused feature data; S5. Match sorting strategy types based on the fused feature data to generate target sorting strategy identification data; S6. Calling a sorting program according to the target sorting strategy identification data; S7, extracting the position offset parameters of the material; S8, generating the robot arm motion trajectory control instruction to drive the actuator to grab the material; S9, real-time monitoring of position offset during the grasping process and dynamic compensation; S10. Complete accurate material sorting and positioning correction.

[0023] Step S1 specifically includes: S11. Collect high-resolution texture, geometric profile, and reflectivity data of the material surface through a multispectral industrial camera to generate visual feature data; S12. Collect the parameters of the sorting line transmission speed, ambient light intensity, and material stacking height through the laser radar and depth sensor to generate environmental parameter data.

[0024] Step S2 specifically includes: S21. Importing visual feature data and environmental parameter data into the sorting control platform; S22. Adopt an adaptive search algorithm to retrieve data sampling frequency by resolution keyword, and generate material static resolution data and scene dynamic resolution data, the unit of which is pixels per millimeter.

[0025] Step S3 specifically includes: S31. Compare the numerical differences between the static resolution data of the material and the dynamic resolution data of the scene; S32: If the difference is less than a preset threshold, it is determined to be consistent; otherwise, it is determined to be inconsistent.

[0026] Step S4 specifically includes: S41. When it is determined that there is inconsistency, a bilinear interpolation algorithm is used to adjust the resolution of the visual feature data to be consistent with the scene dynamic resolution data; S42. Generate material visual feature correction data that retains original texture features.

[0027] Step S5 specifically includes: S51, aligning the corrected visual feature data and the environmental parameter data according to spatial coordinates; S52. Construct a fusion feature matrix including material surface features, environmental interference factors and posture parameters.

[0028] Step S6 specifically includes: S61. Establish a standard feature library for sorting strategies, including benchmark feature matrices corresponding to nine sorting strategies; S62, using a bionic optimization algorithm to match the fusion feature matrix with the reference feature matrix: S621, initializing the position and number of iterations of the sorting strategy search population; S622, in the exploration phase, simulate the population to randomly search for target features, update the position and retain the better solution; S623. Simulate local fine search of the population in the development phase, calculate feature fitness and replace the suboptimal solution. The update formula is: in For the i-th population in dimension The new location, For the current dimension location, is a random number in the interval [0,1], is the upper limit of the search space, is the lower limit of the search space, is the current iteration number; S624: Output the optimal matching strategy type after reaching the maximum number of iterations; S63: Generate target sorting strategy identification data.

[0029] Step S9 specifically includes: S91. Detect the horizontal offset and rotation angle of the material during the grasping process through a real-time visual tracking system; S92. Use the PID control algorithm to generate a correction instruction including the X, Y, and Z axis compensation amounts and the rotation compensation angle. The control formula is: in is the deviation correction control quantity at time t, is the pose error at time t, is the proportional gain coefficient, is the integral gain coefficient, is the differential gain coefficient, Historical error accumulation, is the error change rate.

[0030] Step S10 specifically includes: S101, converting the correction instruction into a control signal for the motor of the robot arm joint; S102, performing adaptive grasping by using a vacuum suction cup or a flexible gripper of an end effector; S103. Perform a secondary visual check when placing the material at the target workstation to ensure that the positioning error is less than 0.1 mm.

[0031] The system includes: Visual parameter processing module: includes multispectral imaging unit, environmental perception unit, resolution analysis unit and feature correction unit; Sorting strategy decision module: includes feature fusion unit, strategy feature library storage unit and bionic algorithm matching unit; Execution correction control module: includes trajectory planning unit, real-time correction unit and positioning verification unit.

[0032] Example 1: Precision medical packaging sorting On the automated sorting production line for medical syringe packaging boxes, after the system is started, a high-resolution multispectral camera captures images of the packaging boxes on the conveyor belt. The camera captures the drug information, batch number, and anti-counterfeiting logo printed on the surface of the box at a rate of 60 frames per second. At the same time, a lidar scanner monitors the running speed and vibration frequency of the conveyor belt in real time.

[0033] When the ambient light sensor detects that the top LED light source has a color temperature offset due to voltage fluctuations, the visual parameter processing module immediately starts the resolution collaborative correction program: first, it compares the numerical difference between the static imaging resolution and the dynamic environment resolution. When it is determined that the difference exceeds the preset safety threshold, the adaptive bilinear interpolation algorithm is used to reconstruct the original image.

[0034] While preserving the sharpness of printed text, this algorithm adjusts the sampling frequency of color channels pixel by pixel to eliminate text edge blurring caused by chromatic aberration. The feature fusion unit then constructs a three-dimensional spatial feature matrix from the reconstructed image, along with vibration parameters and material position coordinates. The bionic optimization engine within the sorting strategy decision module activates the fish-schooling algorithm, searching within a pre-existing library of nine medical packaging sorting strategies. After 12 iterations, it identified the "Precision Medicine Box Grasping Strategy Type IV."

[0035] The execution correction control module calls the strategy parameters to drive the six-axis robotic arm. When the vacuum suction cup is 15 mm away from the surface of the box, the real-time visual tracking system detects a 2.5° angular offset of the box caused by the vibration of the conveyor belt through the principle of triangulation. The PID control unit immediately generates a composite correction instruction including X-axis reverse displacement compensation, Z-axis height fine-tuning and counterclockwise rotation torque. The flexible gripper completes the posture adjustment before contacting the box, and finally accurately transfers the packaging box to the sterilization area. In the secondary verification link before blanking, the box is 3D modeled using a laser profilometer to confirm that the positioning error does not exceed the medical packaging industry standard of 0.08 mm.

[0036] Example 2: Sorting of Multi-Material Electronic Components When the electronic component sorting center handles mixed transmission of circuit boards and metal casings, the system faces multiple challenges: the green solder mask layer of the circuit board and the mirror reflective properties of the metal casing cause the spectral reflectivity to differ by more than 70%. The multispectral camera simultaneously obtains surface feature data of the two types of materials in a single scan. The environmental perception unit detects the high-light overexposure phenomenon of the metal casing caused by strong light exposure. The resolution collaborative correction end activates a dual-channel processing mechanism: adaptive histogram equalization is used to enhance the pad details of the circuit board, and a light suppression algorithm based on the Fresnel reflection model is implemented on the metal casing.

[0037] The feature fusion unit constructs the processed image data, material stacking height parameters, and transmission line speed into a four-dimensional feature tensor. The sorting strategy decision module uses a genetic algorithm to search in parallel in the strategy space. After nine generations of evolutionary calculations, it matches the "low contact pressure strategy type III" for the circuit board and the "electromagnetic adsorption strategy type VII" for the metal shell.

[0038] When the parallel robotic arms perform synchronous grasping, the posture coupling analysis terminal detects the 3.2mm horizontal offset of the metal casing caused by electrostatic adsorption through the TOF depth sensor, and generates a real-time correction plan that includes Z-axis lift compensation and static eliminator activation instructions. The actuator completes the coordinated operation of the suction cup and electromagnetic clamp within 100ms: the vacuum suction cup picks up the circuit board with a contact pressure of 0.15N, and the electromagnetic clamp attracts the metal casing with a 12mT magnetic flux after static elimination. The two types of materials are finally transferred to the SMT patch area and electroplating processing line respectively. The entire process is carried out without material collision or surface scratches. After 8 hours of continuous operation, the system has processed a total of 23,000 components, and the sorting error rate has remained stable at an industry-leading level.

[0039] Example 3: Sorting of Heavy-Duty Auto Parts Under the extreme working conditions of the automobile wheel sorting center, the system needs to deal with three concurrent anomalies: intermittent strong light interference caused by the welding robot operation, 0.8mm amplitude mechanical vibration caused by the resonance of the transmission chain plate, and abnormal stacking posture due to residual magnetism of the components. The visual parameter processing module activates the multimodal anti-interference protocol: the multispectral camera switches to infrared-visible light fusion mode to collect the surface features of the wheel hub, and the environmental perception unit fuses the gyroscope angular velocity data with the laser ranging value to generate vibration compensation parameters.

[0040] When the resolution consistency judgment module detects geometric distortion of the bolt hole in the dynamic imaging, the feature correction unit uses the phase recovery algorithm to reconstruct the three-dimensional point cloud model. The sorting strategy decision module locks the "heavy load grasping strategy V type" in the standard strategy library. However, the bionic algorithm matching unit finds through fitness calculation that the current feature deviates by 32% from the standard strategy. The engine then initiates a deep exploration mechanism: it redeploys the Osprey optimization algorithm in the nine-dimensional strategy space, and after 22 iterative explorations, it matches the "electromagnetic field adaptation strategy type IX."

[0041] The execution correction control module calls the magnetic interference compensation parameters of this strategy. When the heavy robotic arm approaches the wheel hub, the real-time monitoring system detects the 7° angular offset in the XY plane caused by the strong magnetic field attracting the component through the Hall sensor. The dynamic compensation algorithm integrates the electromagnetic intensity meter data to generate a reverse compensation field and adjusts the contact trajectory of the hydraulic fixture. In the final grasping stage, the six-dimensional force sensor provides real-time feedback of the contact pressure data to ensure that the 12kN clamping force is evenly distributed on the contact surface of the wheel hub flange. After the wheel hub is accurately transferred to the inspection station and reviewed by the laser scanner, the positioning error is controlled within 0.15mm, meeting the strict standards of the automobile assembly process.

[0042] Example 4: Long-cycle operation of food packaging During the 216-hour continuous operation of the eight-station sorting system for quick-frozen food packaging bags, the system processed a total of 540,000 pieces of three types of packaging: PE film, aluminum foil composite film, and degradable paper film. The visual parameter processing module established a dynamic learning database: recording the transmittance change curves of different materials in low-temperature environments, storing historical data on light intensity at each station, and learning the correlation model between conveyor belt speed and vibration amplitude. When processing new pearlescent film packaging, the stepped color difference end automatically separates the color difference detection thresholds of the printed pattern area, heat-sealed edge area, and batch coding area to avoid color interference of reflective materials on the main pattern.

[0043] The sorting strategy decision module incorporates the optical characteristics of pearlescent film into the strategy library through the online learning mechanism and creates the "High-Reflective Packaging Strategy X-Type". During the execution process, the sudden warping of the heat-sealed edge causes the grasping point to shift. The posture coupling analysis end detects a height anomaly of 2.3mm through 3D structured light scanning and immediately triggers a three-level response mechanism: the first level adjusts the adsorption point distribution of the vacuum suction cup array, the second level replans the obstacle avoidance path of the robotic arm to bypass the warped area, and the third level controls the contact pressure within the safety threshold through the piezoelectric sensor.

[0044] While maintaining the production line's high-speed operation at 30m / min, the system's peak processing capacity reached 125 pieces per minute. After a cumulative operation of 500 hours, the self-diagnosis module issued a maintenance warning: the gain attenuation of the multispectral camera's blue channel reached the calibrated value, prompting a white balance calibration. After preventive maintenance, the system continued to operate stably for three production cycles, setting a new industry record for mean time between failures, and the sorting accuracy continued to maintain the highest level required for industrial mass production.

[0045] Example 5: Precision Electronic Component Sorting In the micro-sensor component sorting scenario, the system processes MEMS gyroscope chips that are only 0.8mm×0.5mm in size. The multispectral camera switches to microscopic imaging mode and uses a 200x optical zoom lens to capture the gold wire bonding status of the chip pads. The environmental perception unit monitors the 0.3m / s airflow disturbance generated by the clean room laminar flow system, and the visual parameter processing module initiates nanometer-level resolution collaborative correction: first, the subtle differences between the static imaging resolution and the dynamic environment resolution are compared. When the difference is determined to reach the critical threshold, the Gaussian-Laplacian pyramid algorithm is used for multi-scale image fusion.

[0046] This technology eliminates edge chattering caused by airflow disturbances through four-layer decomposition and reconstruction, retaining the geometric features of the bonding point with a diameter of 2μm. The feature fusion unit constructs a five-dimensional feature matrix using the reconstructed image, air pressure sensor data, and electrostatic monitoring values. The sorting strategy decision module activates the deep reinforcement learning engine and performs Monte Carlo tree search in a knowledge base containing 12 types of precision electronic sorting strategies. After 35 rounds of strategy evaluation, it locks in the "Microcomponent Non-Destructive Grasping Strategy Type XII."

[0047] The execution correction control module calls this strategy parameter to drive the piezoelectric ceramic manipulator. When the tungsten steel probe approaches the chip surface within 5μm, the posture coupling analysis end detects a 0.12μm displacement caused by electrostatic adsorption through the laser interferometer. The dynamic compensation algorithm immediately generates a composite instruction: 1. Activate the ion air gun to neutralize 500V surface static electricity; 2. Adjust the probe contact angle to 87°; 3. Control the contact pressure at the critical value of 0.02mN; During the final picking stage, the six-axis nanopositioning platform transfers the chip to the test station with a repeatability of 0.004μm. A triple protection mechanism is adopted throughout the process: the first level verifies in real time through a scanning electron microscope that there is no collision risk in the grasping path; the second level uses a quartz crystal microbalance to monitor contact stress fluctuations; the third level starts an atomic force microscope to perform a three-dimensional morphology re-inspection after the material is dropped. After the system continuously processed 18,000 chips, statistics showed: the bonding gold wire zero damage rate was 100%, and the position repeatability error was less than 0.08μm, setting a new record for accuracy in the field of microelectronics sorting. The self-diagnosis module simultaneously generates an equipment loss report, indicating that the ceramic suction nozzle requires plasma cleaning maintenance after a cumulative operation of 500,000 times.

[0048] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0049] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A vision-guided automatic sorting and deviation correction method, characterized by: The method comprises: S1. Collect visual feature data of target materials and environmental parameter data of sorting scenes; S2. performing adaptive processing of the visual feature data based on the visual feature data and the environmental parameter data; S3, generating adapted visual feature data according to the resolution processing result; S4, combining visual feature data and environmental parameter data to construct fused feature data; S5. Match sorting strategy types based on the fused feature data to generate target sorting strategy identification data; S6. Calling a sorting program according to the target sorting strategy identification data; S7, extracting the position offset parameters of the material; S8, generating the robot arm motion trajectory control instruction to drive the actuator to grab the material; S9, real-time monitoring of position offset during the grasping process and dynamic compensation; S10. Complete accurate material sorting and positioning correction.

2. The vision-guided automatic sorting and deviation correction method according to claim 1, characterized in that: The step S1 specifically includes: S11. Collect high-resolution texture, geometric profile, and reflectivity data of the material surface through a multispectral industrial camera to generate visual feature data; S12. Collect the parameters of the sorting line transmission speed, ambient light intensity, and material stacking height through the laser radar and depth sensor to generate environmental parameter data.

3. The vision-guided automatic sorting and deviation correction method according to claim 1, characterized in that: The step S2 specifically includes: S21. Importing visual feature data and environmental parameter data into the sorting control platform; S22. Adopt an adaptive search algorithm to retrieve data sampling frequency by resolution keyword, and generate material static resolution data and scene dynamic resolution data, the unit of which is pixels per millimeter.

4. The vision-guided automatic sorting and deviation correction method according to claim 1, characterized in that: The step S3 specifically includes: S31. Compare the numerical differences between the static resolution data of the material and the dynamic resolution data of the scene; S32: If the difference is less than a preset threshold, it is determined to be consistent; otherwise, it is determined to be inconsistent.

5. The vision-guided automatic sorting and deviation correction method according to claim 1, characterized in that: The step S4 specifically includes: S41. When it is determined that there is inconsistency, a bilinear interpolation algorithm is used to adjust the resolution of the visual feature data to be consistent with the scene dynamic resolution data; S42. Generate material visual feature correction data that retains original texture features.

6. The vision-guided automatic sorting and deviation correction method according to claim 1, characterized in that: The step S5 specifically includes: S51, aligning the corrected visual feature data and the environmental parameter data according to spatial coordinates; S52. Construct a fusion feature matrix including material surface features, environmental interference factors and posture parameters.

7. The vision-guided automatic sorting and deviation correction method according to claim 1, characterized in that: The step S6 specifically includes: S61. Establish a standard feature library for sorting strategies, including benchmark feature matrices corresponding to nine sorting strategies; S62, using a bionic optimization algorithm to match the fusion feature matrix with the reference feature matrix: S621, initializing the position and number of iterations of the sorting strategy search population; S622, in the exploration phase, simulate the population to randomly search for target features, update the position and retain the better solution; S623. Simulate local fine search of the population in the development phase, calculate feature fitness and replace the suboptimal solution. The update formula is: in For the i-th population in dimension The new location, For the current dimension location, is a random number in the interval [0,1], is the upper limit of the search space, is the lower limit of the search space, is the current iteration number; S624: Output the optimal matching strategy type after reaching the maximum number of iterations; S63: Generate target sorting strategy identification data.

8. The vision-guided automatic sorting and deviation correction method according to claim 1, characterized in that: The step S9 specifically includes: S91. Detect the horizontal offset and rotation angle of the material during the grasping process through a real-time visual tracking system; S92. Use the PID control algorithm to generate a correction instruction including the X, Y, and Z axis compensation amounts and the rotation compensation angle. The control formula is: in is the deviation correction control quantity at time t, is the pose error at time t, is the proportional gain coefficient, is the integral gain coefficient, is the differential gain coefficient, Historical error accumulation, is the error change rate.

9. The vision-guided automatic sorting and deviation correction method according to claim 1, characterized in that: The step S10 specifically includes: S101, converting the correction instruction into a control signal for the motor of the robot arm joint; S102, performing adaptive grasping by using a vacuum suction cup or a flexible gripper of an end effector; S103. Perform a secondary visual check when placing the material at the target workstation to ensure that the positioning error is less than 0.1 mm.

10. A vision-guided automatic sorting and deviation-correcting system for implementing the vision-guided automatic sorting and deviation-correcting method according to any one of claims 1 to 9, characterized in that: The system comprises: Visual parameter processing module: includes multispectral imaging unit, environmental perception unit, resolution analysis unit and feature correction unit; Sorting strategy decision module: includes feature fusion unit, strategy feature library storage unit and bionic algorithm matching unit; Execution correction control module: includes trajectory planning unit, real-time correction unit and positioning verification unit.

Images