Cowhide multi-stage sorting and conveying system based on visual identification

Abstract

The invention provides a cow leather multi-stage sorting and conveying system based on visual identification. The cow leather multi-stage sorting and conveying system comprises a conveying belt, a clamping mechanism and a sorting channel. A hyperspectral camera is arranged above the starting end of the conveying belt, and the hyperspectral camera is used for detecting cow leather and outputting first two-dimensional pixel coordinates when material defects are detected; a 3D line laser scanner is arranged at the downstream of the hyperspectral camera, and the 3D line laser scanner is used for generating three-dimensional point cloud coordinates of the cowhide; an industrial RGB camera is arranged at the downstream of the 3D line laser scanner, and the industrial RGB camera is used for detecting cow leather and outputting second two-dimensional pixel coordinates when material defects are detected; the hyperspectral camera, the 3D line laser scanner and the industrial RGB camera are configured with a fusion algorithm, and the fusion algorithm is used for generating target coordinates from the first two-dimensional pixel coordinates, the second two-dimensional pixel coordinates and the three-dimensional point cloud coordinates and generating comprehensive sorting labels according to the first two-dimensional pixel coordinates, the three-dimensional point cloud coordinates and the second two-dimensional pixel coordinates.

Description

Technical Field

[0001] The present invention relates to the technical field of transport devices based on visual recognition, and in particular to a cowhide multi-stage sorting and conveying system based on visual recognition. Background Art

[0002] Compared with traditional sorting and transportation methods, visual recognition-based sorting and transportation has significant advantages in transportation efficiency, flexibility and intelligence level. Because it can surpass traditional sorting methods in transportation efficiency, flexibility, accuracy and long-term cost control through intelligent identification, dynamic path planning and automated control, it is particularly suitable for modern logistics' demand for high-speed, multi-category, and scalable transportation networks, and has become one of the core technologies of smart factories and smart logistics.

[0003] Among them, there is a Chinese patent with application number CN2024101359227, which disclosed a cargo sorting robot and sorting method based on vision guidance on October 8, 2024. Its camera component collects cargo images, the machine vision system determines the orientation through the face sheet recognition algorithm, triggers the clamping and flipping action, and the ranging component assists in size measurement to adapt to the packaging process, thereby completing visual sorting and transportation.

[0004] There is also a Chinese patent with application number CN202010703304X, which disclosed a visual sorting device on July 21, 2020. Its multi-camera linear array captures the position and characteristics of materials in real time, and the control module triggers the air nozzle action through the image processing algorithm, thereby completing the air blowing sorting and transportation.

[0005] However, the existing sorting and conveying system based on visual recognition has limitations in sorting and conveying cowhide. First, the visual system is only for the face sheet and cannot handle the multi-level characteristics of cowhide such as texture, thickness, defects, etc. Second, the rigid clamping mechanism can easily scratch the soft cowhide. Summary of the invention

[0006] The present invention aims to solve the above-mentioned technical problems and provides a multi-level cowhide sorting and conveying system based on visual recognition.

[0007] The technical solution of the present invention is a multi-stage cowhide sorting and conveying system based on visual recognition, comprising a conveyor belt for conveying cowhide, an adsorption clamping mechanism, and a sorting channel; A hyperspectral camera is provided above the starting end of the conveyor belt, and the hyperspectral camera is used to detect cowhide and output a first two-dimensional pixel coordinate when a material defect is detected; a 3D line laser scanner is provided downstream of the hyperspectral camera, and the 3D line laser scanner scans the surface of the conveyor belt at a 45° angle, and the 3D line laser scanner is used to generate three-dimensional point cloud coordinates of the cowhide, and mark the thickness distribution and surface contour; an industrial RGB camera is provided downstream of the 3D line laser scanner, and the industrial RGB cameras are symmetrically arranged on both sides of the conveyor belt, and each of them forms an angle of 30° with the conveyor belt, and the industrial RGB camera is used to detect cowhide and output a second two-dimensional pixel coordinate when a material defect is detected; The hyperspectral camera, the 3D line laser scanner, and the industrial RGB camera are configured with a fusion algorithm, which maps the first two-dimensional pixel coordinates and the second two-dimensional pixel coordinates into a three-dimensional point cloud coordinate system to obtain corresponding three-dimensional physical coordinates, and generates target coordinates when the hyperspectral camera and the 3D line laser scanner both detect abnormalities in the same area, and generates a comprehensive sorting label based on the material grade reflected by the first two-dimensional pixel coordinates, the thickness tolerance reflected by the 3D line laser scanner, and the texture uniformity reflected by the second two-dimensional pixel coordinates.

[0008] As an implementation mode, the clamping mechanism is used to flexibly transport cowhide, and the clamping mechanism includes independently controlled flexible array suction cups arranged in a honeycomb matrix, the flexible array suction cups locate the gripping points according to the XY axis data reflected by the target coordinates, and dynamically adjust the adsorption force of each unit suction cup according to the Z axis data reflected by the target coordinates; The sorting channel is provided with a high-speed swing arm, a pneumatic push rod, and a robot grabbing arm which perform actions according to the comprehensive sorting label. The high-speed swing arm is used to guide the seriously defective cowhide to the waste channel on one side of the sorting channel. The pneumatic push rod is used to push the cowhide of different thicknesses into the thick material channel and the thin material channel on both sides of the sorting channel according to the thickness threshold. The robot grabbing arm is used to deliver the cowhide with high texture grade into the finishing buffer zone.

[0009] As an implementation manner, the installation height of the hyperspectral camera is 1.0 to 1.5 m, and the spectral detection range is 400 to 2500 nm.

[0010] As an implementation manner, the line scanning frequency of the 3D line laser scanner is greater than 2000 Hz, the Z-axis resolution is less than 0.05 mm, and the scanning width covers 105% to 110% of the width of the conveyor belt.

[0011] As an implementation manner, the industrial RGB camera is configured with a circularly polarized light source, and the imaging overlap rate of the industrial RGB cameras on both sides is greater than 85%.

[0012] As an implementation mode, the diameter of the unit suction cup of the flexible array suction cup is 5 to 15 mm, the center distance between adjacent unit suction cups is 1.2 to 1.5 times the diameter of the unit suction cup, and each unit suction cup is equipped with an independent vacuum pressure sensor and proportional valve.

[0013] As an implementation manner, the adsorption force of the flexible array suction cup corresponding to the thin area is 0.2 bar, and the adsorption force of the flexible array suction cup corresponding to the thick area is 0.5 bar.

[0014] As an implementation manner, the maximum swing speed of the high-speed swing arm is greater than 1.5 m / s.

[0015] As an implementation manner, the thickness threshold is 0.8 to 1.2 mm.

[0016] As an embodiment, the end effector of the robot grasping arm is a bionic flexible gripper, the gripping force of the bionic flexible gripper is controlled according to the grade of cowhide texture, and the gripping surface of the bionic flexible gripper is covered with a polyurethane anti-slip layer.

[0017] Compared with the prior art, the beneficial effect of the present invention is that the cowhide multi-stage sorting and conveying system based on visual recognition is composed of three parts, namely, a conveyor belt, an adsorption clamping mechanism, and a sorting channel, arranged in sequence according to the direction of the assembly line. By deeply coupling visual recognition with the sorting mechanism (adsorption clamping mechanism and sorting channel), not only can the cowhide multi-parameter high-precision sorting be achieved, but also the dynamic optimization mechanism can be used to adapt to complex working conditions, thus meeting the requirements of zero damage and high consistency in the sorting and conveying of cowhide.

[0018] A hyperspectral camera, a 3D line laser scanner, and an industrial RGB camera are set above the conveyor belt. The hyperspectral camera, the 3D line laser scanner, and the industrial RGB camera are equipped with a fusion algorithm. That is, the target coordinates and comprehensive sorting labels of the visual system are generated based on the hyperspectral camera, the 3D line laser scanner, and the industrial RGB camera. A multimodal cross-validation is formed, which greatly reduces the missed detection rate. Then, the adsorption clamping mechanism is responsible for non-destructively grabbing the cowhide from the conveyor belt to the sorting channel. The flexible array suction cup of the clamping mechanism locates the grabbing point according to the XY axis data reflected by the target coordinates, thereby avoiding the texture-dense area. The adsorption force of each unit suction cup is also dynamically adjusted according to the Z axis data reflected by the target coordinates to avoid local deformation. Finally, the high-speed swing arm, pneumatic push rod, and robot grabbing arm installed on the sorting channel perform actions according to the comprehensive sorting label, making the sorting channel a multi-level sorting channel. The first-level sorting is defect removal, which is performed by the high-speed swing arm, the second-level sorting is thickness grading, which is performed by the pneumatic push rod, and the third-level sorting is texture optimization, which is performed by the robot grabbing arm. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A schematic diagram of the hardware layout of a cowhide multi-stage sorting and conveying system based on visual recognition provided by an embodiment of the present invention; Figure 2 A flow chart of a fusion algorithm provided for an embodiment of the present invention.

[0020] In the figure: 1. Conveyor belt; 2. Clamping mechanism; 3. Sorting channel; 4. Hyperspectral camera; 5. 3D line laser scanner; 6. Industrial RGB camera; 7. Flexible array suction cup; 8. High-speed swing arm; 9. Pneumatic push rod; 10. Robot grabbing arm; 11. Waste channel; 12. Thick material channel; 13. Thin material channel; 14. Finishing buffer zone. DETAILED DESCRIPTION

[0021] The above and other embodiments and advantages of the present invention are described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments.

[0022] In one embodiment, Figure 1 to Figure 2 shown.

[0023] The cowhide multi-stage sorting and conveying system based on visual recognition provided in this embodiment includes a conveyor belt 1 for conveying cowhide, an adsorption clamping mechanism 2, and a sorting channel 3; a hyperspectral camera 4 is provided above the starting end of the conveyor belt 1, and the hyperspectral camera 4 is used to detect cowhide and output a first two-dimensional pixel coordinate when a material defect is detected; a 3D line laser scanner 5 is provided downstream of the hyperspectral camera 4, and the 3D line laser scanner 5 scans the surface of the conveyor belt 1 at a 45° angle, and the 3D line laser scanner 5 is used to generate three-dimensional point cloud coordinates of the cowhide, and mark the thickness distribution and surface contour; an industrial RGB camera 6 is provided downstream of the 3D line laser scanner 5, and the industrial RGB camera 6 is symmetrically arranged on both sides of the conveyor belt 1, and each is at an angle of 30° to the conveyor belt 1, and the industrial RGB camera 6 is used to detect cowhide and output a second two-dimensional pixel coordinate when a material defect is detected; the hyperspectral camera 4, the 3D line laser scanner 5, and the industrial RGB camera 6 are configured with a fusion algorithm, and the fusion algorithm is to map the first two-dimensional pixel coordinates and the second two-dimensional pixel coordinates to a three-dimensional point cloud coordinate system to obtain a coordinate system. The three-dimensional physical coordinates corresponding to the first two-dimensional pixel coordinates, and when the hyperspectral camera 4 and the 3D line laser scanner 5 both detect anomalies in the same area, the target coordinates are generated, and a comprehensive sorting label is generated according to the material grade reflected by the first two-dimensional pixel coordinates, the thickness tolerance reflected by the 3D line laser scanner 5, and the texture uniformity reflected by the second two-dimensional pixel coordinates; the clamping mechanism 2 is used to flexibly transport the cowhide, and the clamping mechanism 2 includes independently controlled flexible array suction cups 7 arranged in a honeycomb matrix, and the flexible array suction cups 7 are positioned and grasped according to the XY axis data reflected by the target coordinates. The sorting channel 3 is provided with a high-speed swing arm 8, a pneumatic push rod 9, and a robot grabbing arm 10 which perform actions according to the comprehensive sorting label. The high-speed swing arm 8 is used to guide the seriously defective cowhide to the waste channel 11 on one side of the sorting channel 3. The pneumatic push rod 9 is used to push the cowhide of different thicknesses into the thick material channel 12 and the thin material channel 13 on both sides of the sorting channel 3 according to the thickness threshold. The robot grabbing arm 10 is used to send the cowhide with high texture grade into the finishing buffer 14.

[0024] In this embodiment, the cowhide multi-level sorting and conveying system based on visual recognition is aimed at the traditional cowhide sorting and conveying work, which requires a large number of processes to be manually participated, and the current sorting and conveying system based on visual recognition is not suitable for the sorting and conveying of cowhide, so an innovative technical solution is proposed. Therefore, the technical problem to be solved by the cowhide multi-level sorting and conveying system based on visual recognition is that the current visual system cannot handle the multi-level characteristics of cowhide such as texture, thickness, defects, etc., and the rigid clamping mechanism 2 is easy to scratch the soft cowhide. The cowhide multi-level sorting and conveying system based on visual recognition is composed of three parts: a conveyor belt 1, an adsorption clamping mechanism 2, and a sorting channel 3 arranged in sequence according to the direction of the assembly line. By deeply coupling visual recognition with the sorting mechanism (adsorption clamping mechanism 2 and sorting channel 3), not only can the high-precision sorting of cowhide multi-parameters be achieved, but also the dynamic optimization mechanism can adapt to complex working conditions, and meet the requirements of zero damage and high consistency in the sorting and conveying of cowhide.

[0025] In this embodiment, a hyperspectral camera 4, a 3D line laser scanner 5, and an industrial RGB camera 6 are arranged above the conveyor belt 1, and the hyperspectral camera 4, the 3D line laser scanner 5, and the industrial RGB camera 6 are arranged in sequence according to the direction of the assembly line. The hyperspectral camera 4 is located 1.2m above the conveyor belt 1, vertically facing the center of the conveyor belt 1, and covers a width of about 1.5m. The function of the hyperspectral camera 4 is to detect material uniformity, hidden defects, and surface contamination. When the hyperspectral camera 4 detects material defects, it outputs the first two-dimensional pixel coordinates. The 3D line laser scanner 5 is located 0.5m downstream of the hyperspectral camera 4, with a lateral span of 1.8m, and scans the surface of the conveyor belt 1 at a 45° oblique angle. The function of the 3D line laser scanner 5 is to generate three-dimensional point cloud coordinates of cowhide, calculate thickness distribution (avoiding manual thickness measurement errors), and edge warping (sorting leather with high flatness requirements). One of the functions of placing the 3D line laser scanner 5 after the hyperspectral camera 4 is to combine the defect coordinates of the hyperspectral image to perform secondary thickness verification on abnormal areas (such as bulges and depressions). The industrial RGB camera 6 is located 0.3m downstream of the 3D line laser scanner 5, and is arranged symmetrically on both sides (at an angle of 30° with the conveyor belt 1), covering the entire surface of the cowhide. The function of the industrial RGB camera 6 is to extract texture clarity (pore density), surface scratches (micron level), and color uniformity (color difference ΔE<1.5). The industrial RGB camera 6 eliminates reflective interference through multi-angle imaging (30° angle on both sides) and assists in correcting the surface coordinate accuracy of the texture area.

[0026] In this embodiment, the hyperspectral camera 4, 3D line laser scanner 5, and industrial RGB camera 6 are not independent of each other and are not related to each other, but are configured with a fusion algorithm. That is, based on the hyperspectral camera 4, 3D line laser scanner 5, and industrial RGB camera 6, the target coordinates and comprehensive sorting labels of the visual system are generated. The 3D line laser scanner 5 is the core sensor for spatial positioning, which is responsible for providing the three-dimensional point cloud coordinates of the cowhide and marking the thickness distribution, surface contour and other spatial information. When the hyperspectral camera 4 detects material defects (such as subcutaneous blood spots), it outputs two-dimensional pixel coordinates, but it needs to be matched with the 3D point cloud data to convert into three-dimensional physical coordinates. The industrial RGB camera 6 detects surface scratches and marks another set of two-dimensional pixel coordinates of texture defects. The 3D line laser scanner 5 maps the two-dimensional pixel coordinates of the hyperspectral camera 4 and the industrial RGB camera 6 into the three-dimensional point cloud coordinate system to obtain the corresponding physical coordinates. Then, the multi-sensor coordinate system is aligned through calibration parameters (sensor external parameters) to ensure spatial consistency. If the hyperspectral camera 4 and the 3D line laser scanner 5 both detect an abnormality in the same area (such as a material defect accompanied by a local thickness mutation), it is determined to be a real defect and the final target coordinates are generated. Correspondingly, if only a single sensor alarms (such as hyperspectral misdetection of dust), the interference is eliminated through texture verification of the RGB camera. The multimodal cross-validation formed by the hyperspectral camera 4, the 3D line laser scanner 5, and the industrial RGB camera 6 greatly reduces the missed detection rate. For example, the hyperspectral camera 4 alone cannot identify tiny scratches. Then, the target coordinates finally generated are the three-dimensional physical coordinates after multimodal data fusion, with the 3D line laser scanner 5 providing the spatial reference, and the hyperspectral and RGB cameras providing the defect semantic information. In addition, a comprehensive sorting label is generated based on the texture uniformity (measured by the industrial RGB camera 6), the thickness tolerance (measured by the 3D line laser scanner 5), and the material grade (measured by the hyperspectral camera 4). The so-called comprehensive sorting label distinguishes cowhide with serious defects, cowhide of different thicknesses (divided by thickness threshold), and cowhide with high texture grade. The generation of comprehensive sorting labels is not based solely on parameters such as texture and thickness, but is generated through a preset texture-thickness-material association model (for example, a thinner thickness is allowed in a texture-dense area), combined with material grade, thickness tolerance, and texture uniformity. Therefore, the sorting logic is more in line with process requirements.

[0027] In this embodiment, the suction clamping mechanism 2 is responsible for grabbing the cowhide from the conveyor belt 1 to the sorting channel 3 without damage. The clamping mechanism 2 includes a flexible array suction cup 7 arranged in a honeycomb matrix, and each unit suction cup in the flexible array suction cup 7 is independently controlled to adjust the suction force of each unit suction cup. The flexible array suction cup 7 locates the grabbing point according to the XY axis data reflected by the target coordinates, thereby avoiding the texture-dense area. The flexible array suction cup 7 dynamically adjusts the suction force of each unit suction cup according to the Z axis data (thickness data) reflected by the target coordinates (low pressure 0.2 bar in the thin area, high pressure 0.5 bar in the thick area) to avoid local deformation.

[0028] In this embodiment, a waste channel 11, two opposite thick material channels 12 and thin material channels 13, and a finishing buffer 14 are sequentially arranged on the sorting channel 3. A high-speed swing arm 8 is arranged at the waste channel 11, a pneumatic push rod 9 is arranged between the opposite thick material channel 12 and thin material channel 13, and a robot grab arm 10 is arranged at the finishing buffer 14. The high-speed swing arm 8 is used to guide the cowhide with serious defects to the waste channel 11 on one side of the sorting channel 3, the pneumatic push rod 9 is used to push cowhide of different thicknesses into the thick material channel 12 and thin material channel 13 on both sides of the sorting channel 3 according to the thickness threshold, and the robot grab arm 10 is used to send the cowhide with high texture grade into the finishing buffer 14. The high-speed swing arm 8, the pneumatic push rod 9, and the robot grabbing arm 10 perform actions according to the comprehensive sorting label, so that the sorting channel 3 becomes a multi-level sorting channel 3. The first-level sorting is defect removal, which is performed by the high-speed swing arm 8, the second-level sorting is thickness grading, which is performed by the pneumatic push rod 9, and the third-level sorting is texture optimization, which is performed by the robot grabbing arm 10.

[0029] In one embodiment, the multi-stage cowhide sorting and conveying system based on visual recognition has a hyperspectral camera 4 installed at a height of 1.0 to 1.5 m, and a spectral detection range of 400 to 2500 nm.

[0030] In this embodiment, the installation height of the hyperspectral camera 4 is set to 1.0-1.5m to ensure that the maximum size of the cowhide (about 1.2m×1.8m) is covered and lens distortion is avoided. The spectral detection range is 400-2500nm, which can cover visible light to short-wave infrared, and is used to detect subcutaneous blood spots (characteristic absorption peak at 1450nm) and mildew (abnormal reflectivity at 1700-2100nm).

[0031] In one embodiment, the cowhide multi-level sorting and conveying system based on visual recognition has a line scanning frequency of the 3D line laser scanner 5 greater than 2000 Hz, a Z-axis resolution less than 0.05 mm, and a scanning width covering 105% to 110% of the width of the conveyor belt 1.

[0032] In this embodiment, the line scanning frequency of the 3D line laser scanner 5 is selected to be greater than 2000Hz, which can match the conveyor belt 1 running at a high speed (1.2m / s) to ensure that the entire surface is scanned without missing. The Z-axis resolution is selected to be less than 0.05mm, which can accurately measure the thickness distribution. The coverage reaches 105% to 110% of the width of the conveyor belt 1, which can eliminate the edge blind area.

[0033] In one embodiment, the cowhide multi-level sorting and conveying system based on visual recognition has its industrial RGB camera 6 configured with a circularly polarized light source, and the imaging overlap rate of the industrial RGB cameras 6 on both sides is greater than 85%.

[0034] In this embodiment, by configuring a circular polarized light source, the specular reflection on the cowhide surface can be eliminated, and the texture recognition accuracy can be improved. The imaging overlap rate is greater than 85%, and the surface micro scratches (minimum detection size 0.1mm) can be reconstructed through binocular stereo vision.

[0035] In one embodiment, the cowhide multi-stage sorting and conveying system based on visual recognition has a unit suction cup diameter of 5 to 15 mm, a center distance between adjacent unit suction cups of 1.2 to 1.5 times the diameter of the unit suction cup, and each unit suction cup is equipped with an independent vacuum pressure sensor and a proportional valve. Preferably, the adsorption force of the flexible array suction cup 7 corresponding to the thin area is 0.2 bar, and the adsorption force of the flexible array suction cup 7 corresponding to the thick area is 0.5 bar.

[0036] In this embodiment, the diameter is 5 to 15 mm, and the center distance is 1.2 to 1.5 times the diameter, thereby forming a uniform negative pressure distribution. Each suction cup is equipped with a vacuum pressure sensor and a proportional valve to dynamically adjust the adsorption force according to the thickness data (Z axis). The adsorption force of the flexible array suction cup 7 corresponding to the thin area is 0.2 bar to prevent excessive stretching. The adsorption force of the flexible array suction cup 7 corresponding to the thick area is 0.5 bar to ensure the grasping stability.

[0037] In one embodiment, the maximum swing speed of the high-speed swing arm 8 of the cowhide multi-level sorting and conveying system based on visual recognition is greater than 1.5m / s. Its thickness threshold is 0.8-1.2mm. The end effector of its robot grabbing arm 10 is a bionic flexible clamp, the clamping force of the bionic flexible clamp is controlled according to the cowhide texture level, and the clamping surface of the bionic flexible clamp is covered with a polyurethane anti-slip layer.

[0038] In this embodiment, the maximum speed of the high-speed swing arm 8 is >1.5m / s, the response time is <100ms, and the leather with serious defects is preferentially removed. The pneumatic push rod 9 is graded according to the process requirements based on the thickness threshold of 0.8-1.2mm. The clamping surface of the bionic flexible clamp is covered with a polyurethane anti-slip layer to increase the friction coefficient, and the clamping force is divided into three levels (5N / 10N / 15N) to match the texture level.

[0039] The specific implementation methods described above further describe the invention purpose, technical solutions, and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the protection scope of the present invention. It is particularly pointed out that for those skilled in the art, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. The multi-level sorting and conveying system for cowhide based on visual recognition is characterized by: It includes a conveyor belt for conveying cowhide, an adsorption clamping mechanism, and a sorting channel; A hyperspectral camera is provided above the starting end of the conveyor belt, and the hyperspectral camera is used to detect cowhide and output a first two-dimensional pixel coordinate when a material defect is detected; a 3D line laser scanner is provided downstream of the hyperspectral camera, and the 3D line laser scanner scans the surface of the conveyor belt at a 45° angle, and the 3D line laser scanner is used to generate three-dimensional point cloud coordinates of the cowhide, and mark the thickness distribution and surface contour; an industrial RGB camera is provided downstream of the 3D line laser scanner, and the industrial RGB cameras are symmetrically arranged on both sides of the conveyor belt, and each of them forms an angle of 30° with the conveyor belt, and the industrial RGB camera is used to detect cowhide and output a second two-dimensional pixel coordinate when a material defect is detected; The hyperspectral camera, the 3D line laser scanner, and the industrial RGB camera are configured with a fusion algorithm, which maps the first two-dimensional pixel coordinates and the second two-dimensional pixel coordinates into a three-dimensional point cloud coordinate system to obtain corresponding three-dimensional physical coordinates, and generates target coordinates when the hyperspectral camera and the 3D line laser scanner both detect abnormalities in the same area, and generates a comprehensive sorting label based on the material grade reflected by the first two-dimensional pixel coordinates, the thickness tolerance reflected by the 3D line laser scanner, and the texture uniformity reflected by the second two-dimensional pixel coordinates.

2. The multi-level cowhide sorting and conveying system based on visual recognition according to claim 1 is characterized in that: The clamping mechanism is used to flexibly transport the cowhide, and the clamping mechanism includes independently controlled flexible array suction cups arranged in a honeycomb matrix, and the flexible array suction cups locate the grasping points according to the XY axis data reflected by the target coordinates, and dynamically adjust the adsorption force of each unit suction cup according to the Z axis data reflected by the target coordinates; The sorting channel is provided with a high-speed swing arm, a pneumatic push rod, and a robot grabbing arm which perform actions according to the comprehensive sorting label. The high-speed swing arm is used to guide the seriously defective cowhide to the waste channel on one side of the sorting channel. The pneumatic push rod is used to push the cowhide of different thicknesses into the thick material channel and the thin material channel on both sides of the sorting channel according to the thickness threshold. The robot grabbing arm is used to deliver the cowhide with high texture grade into the finishing buffer zone.

3. The multi-level cowhide sorting and conveying system based on visual recognition according to claim 1 is characterized in that: The installation height of the hyperspectral camera is 1.0-1.5m, and the spectrum detection range is 400-2500nm.

4. The multi-level cowhide sorting and conveying system based on visual recognition according to claim 1 is characterized in that: The line scanning frequency of the 3D line laser scanner is greater than 2000 Hz, the Z-axis resolution is less than 0.05 mm, and the scanning width covers 105% to 110% of the width of the conveyor belt.

5. The multi-level cowhide sorting and conveying system based on visual recognition according to claim 1 is characterized in that: The industrial RGB camera is equipped with a circular polarized light source, and the imaging overlap rate of the industrial RGB cameras on both sides is greater than 85%.

6. The multi-level cowhide sorting and conveying system based on visual recognition according to claim 2 is characterized in that: The diameter of the unit suction cup of the flexible array suction cup is 5-15 mm, the center distance between adjacent unit suction cups is 1.2-1.5 times the diameter of the unit suction cup, and each unit suction cup is equipped with an independent vacuum pressure sensor and a proportional valve.

7. The multi-level cowhide sorting and conveying system based on visual recognition according to claim 6 is characterized in that: The adsorption force of the flexible array suction cup corresponding to the thin area is 0.2 bar, and the adsorption force of the flexible array suction cup corresponding to the thick area is 0.5 bar.

8. The multi-level cowhide sorting and conveying system based on visual recognition according to claim 2 is characterized in that: The maximum swing speed of the high-speed swing arm is greater than 1.5 m / s.

9. The multi-level cowhide sorting and conveying system based on visual recognition according to claim 2 is characterized in that: The thickness threshold is 0.8-1.2 mm.

10. The multi-level cowhide sorting and conveying system based on visual recognition according to claim 2 is characterized in that: The end effector of the robot grabbing arm is a bionic flexible clamp, the clamping force of the bionic flexible clamp is graded and controlled according to the cowhide texture level, and the clamping surface of the bionic flexible clamp is covered with a polyurethane anti-slip layer.

Images