A tea picking robot picking device and fresh leaf identification method

Abstract

The present invention discloses a tea picking robot picking device and a fresh leaf identification method, comprising: an upper workbench; a mechanical arm fixed on the upper workbench; a picking mechanism, wherein the picking mechanism comprises an inclined slide, a connecting pipe, an iris cutting mechanism, and a connecting block, wherein the inclined slide is fixed on the mechanical arm, one end of the connecting pipe is fixed to the inclined slide, and the other end is fixed to the iris cutting mechanism, and a mounting hole is provided on the side wall of the connecting pipe, and a vacuum negative pressure type recovery device is connected to the mounting hole. The walking device drives the upper workbench to move to a suitable position, and then the mechanical arm drives the picking mechanism to move, so that the tea leaves to be picked are inserted into the iris cutting mechanism, and after the iris cutting mechanism cuts the tea leaves, the vacuum negative pressure type recovery device extracts the cut tea leaves from the connecting pipe by means of vacuum negative pressure, thereby realizing picking, and picking is very convenient and will not damage the tea leaves.

Description

Technical Field

[0001] The invention belongs to the technical field of tea picking robots, and in particular relates to a tea picking robot picking device and a fresh leaf identification method. Background Art

[0002] Tea picking is a relatively tedious labor, and the efficiency of manual tea picking is low, so more and more machines are replacing manual tea picking.

[0003] The patent with publication number CN116158261A discloses a tea picking robot, which has a mechanical claw joint on the mechanical arm, and a visual system is matched on the mechanical claw joint, and a mechanical claw of a connecting rod mechanical claw, a connecting rod mechanical claw, and a connecting rod mechanical claw multi-link combination can realize the visual action of grabbing tea leaves; the mechanical arm is fixed to the shell of the tea picking machine by bolts; the tea picking machine can put the grabbed tea leaves into the silo inside the silo support frame, and the silo support frame is welded to the shell of the tea picking machine.

[0004] However, the tea leaves are easily damaged by this grabbing method, which affects the quality of the tea leaves, and the speed of placing the picked new tea leaves into the silo is relatively slow. Summary of the invention

[0005] The object of the present invention is to provide a tea picking robot picking device and a fresh leaf identification method, which can avoid damaging the tea leaves during the picking process and can quickly recycle the picked tea leaves.

[0006] To achieve the above object, the present invention provides the following technical solution: a tea picking robot picking device, comprising:

[0007] Go to the workbench;

[0008] A mechanical arm fixed to the upper workbench;

[0009] The picking mechanism includes an inclined slide, a connecting tube and an iris cutting mechanism. The inclined slide is fixed on the mechanical arm. One end of the connecting tube is fixed to the inclined slide, and the other end is fixed to the iris cutting mechanism. A mounting hole is provided on the side wall of the connecting tube, and a vacuum negative pressure recovery device is connected to the mounting hole.

[0010] Furthermore, it also includes a module camera and a picking monocular camera, the picking monocular camera is connected to the outside of the connecting tube through a connecting block, and the module camera is fixed at the top end inside the connecting tube.

[0011] Furthermore, the iris cutting mechanism includes a directional guide ring, a connecting and fixing ring, a first positioning ring, a second positioning ring, a blade and a micro motor, wherein the directional guide ring is fixed on the connecting tube, the first positioning ring is fixed between the directional guide ring and the second positioning ring, the connecting and fixing ring is arranged between the directional guide ring and the second positioning ring, an extension rod is fixed on the connecting and fixing ring, a side extension groove is provided on the side wall of the first positioning ring, a driven arc gear is fixed on one end of the extension rod passing through the extension groove, and the micro motor is fixed on the outer wall of the connecting tube A driving arc gear is fixed on the rotating shaft of the micro motor, and the driving arc gear is meshed with the driven arc gear. A plurality of inclined directional grooves are provided on the connecting and fixing ring, and the plurality of inclined directional grooves are evenly distributed around the axis of the connecting and fixing ring. A blade is connected to each of the inclined directional grooves, and a knife slider plugged into the inclined directional groove is fixed on the blade. A directional milling groove corresponding to the inclined directional groove is provided on the directional guide ring, and a pin key plugged into the directional milling groove is provided on the blade. Two knife bevels are provided on the blade, and adjacent knife bevels fit each other.

[0012] Furthermore, the vacuum negative pressure recovery device includes a vacuum negative pressure machine, a fresh leaf storage box and a negative pressure box, the fresh leaf storage box and the negative pressure box are arranged below the upper workbench, the negative pressure box is provided with a negative pressure cavity, and one side wall of the negative pressure cavity is provided with a plurality of negative pressure holes, the fresh leaf storage box is provided with a fresh leaf cavity, one side of the fresh leaf storage box is provided with a fresh leaf box opening, the fresh leaf box opening faces the negative pressure hole, an air hole connected to the negative pressure cavity is provided on the upper workbench, the air hole and the vacuum negative pressure machine are connected through a first air duct, and a leaf inlet hole located above the fresh leaf storage box is provided on the upper workbench, and the leaf inlet hole is connected to the mounting hole through a leaf duct.

[0013] Furthermore, it also includes a leaf ejecting device and a lower workbench, the lower workbench is arranged below the upper workbench, the leaf ejecting device includes a pushing device and a telescopic rotating device, the pushing device is installed in the fresh leaf cavity, the pushing device is arranged at an end away from the fresh leaf box opening, the telescopic rotating device includes an extending linear electric push rod, a fresh leaf storage box fixing plate and an electric rotating table, the linear electric push rod is fixed on the lower workbench, the fresh leaf storage box fixing plate is fixed on the push rod of the linear electric push rod, the electric rotating table is fixed on the fresh leaf storage box fixing plate, and the fresh leaf storage box is fixed on the electric rotating table.

[0014] Furthermore, it also includes a detection sensor, which is fixed in the fresh leaf cavity and arranged on a side close to the pushing device.

[0015] Furthermore, the pushing device is configured as a linear motor scissor mechanism.

[0016] Furthermore, it also includes a panoramic self-adjusting camera device, which includes a supporting component, a stepper motor is fixed to the upper end of the supporting component, a base connecting block is fixed to the rotating shaft of the stepper motor, a platform fine-tuning mechanism is fixed to the base connecting block, the upper end of the platform fine-tuning mechanism is connected to a servo bracket, a servo motor is fixed to the servo bracket, a servo connecting rod is fixed to the rotating shaft of the servo motor, a cover bottom is fixed to the servo connecting rod, and a binocular camera, a distance sensor, an infrared sensor and two full monocular cameras are fixed to the cover bottom.

[0017] Furthermore, the robotic arm includes a six-degree-of-freedom parallel platform and a picking robot device, the six-degree-of-freedom parallel platform includes a fixed platform and a moving platform, three groups of parallel branch chain groups are evenly installed on the upper end surface of the fixed platform along the axial direction, each group of the parallel branch chain groups is composed of two parallel branch chains, the parallel branch chains include a universal joint, a servo drive motor mechanism, a screw rod, two first linear guide rods, two second linear guide rods, a first positioning support block, a second positioning support block, and a fixed ear seat, one end of the universal joint is fixed on the fixed platform, and the other end is rotatably connected to the servo drive motor mechanism, and the rotation of the servo drive motor mechanism The shaft is fixed to the screw rod, one end of the two first linear guide rods is fixedly connected to the servo drive motor mechanism, and the other end is fixedly connected to the second positioning support block, the two first linear guide rods all pass through the first positioning support block, one end of the two second linear guide rods is fixedly connected to the first positioning support block, and the other end is fixed to the fixed ear seat, the screw rod passes through the first positioning support block and is threadedly connected thereto, the top end of the fixed ear seat is fixed to the lower end surface of the moving platform by a rotating pair, the picking robot device is fixed on the moving platform, and the picking mechanism is fixed to the picking robot device.

[0018] Also included is a fresh leaf identification method, based on the above-mentioned tea picking robot picking device, comprising the following steps:

[0019] S1, after collecting fresh leaf image data, annotate the image, obtain the name of the fresh leaf tea type, size and the position information of the fresh leaf in the image in TXT and XML files,

[0020] S2. Divide the collected fresh leaf image data into training set, test set and validation set, construct the labeled image data set into training set, test set and validation set and put them into the improved YOLOV7 algorithm for learning, so that the improved YOLOV7 algorithm finally obtains the feature map of fresh leaves.

[0021] The improved YOLOv7 network structure consists of Input, Backbone, Neck and Head. Its loss function expression is shown in equations (1), (2), (3) and (4):

[0022]

[0023]

[0024]

[0025]

[0026] Where A and B represent the predicted box and the true box respectively. and Respectively represent the coordinates of the upper left corner and lower right corner of the bounding box A, and Respectively represent the coordinates of the upper left corner and lower right corner of the bounding box B;

[0027] S3. The image obtained by the binocular camera transmits the fresh leaf image to the industrial computer in real time. The industrial computer performs noise reduction, filtering, and grayscale image preprocessing on the image transmitted by the camera; then the trained improved YOLOV7 learning model is used to recognize the photos transmitted by the binocular camera.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] (1) The walking device drives the upper workbench to move to a suitable position, and then the mechanical arm drives the picking mechanism to move, so that the tea leaves to be picked are inserted into the iris cutting mechanism. After the iris cutting mechanism cuts the tea leaves, the vacuum negative pressure recovery device extracts the cut tea leaves from the connecting pipe by means of vacuum negative pressure. The tea leaves are recovered quickly, the picking efficiency is high, and the tea leaves will not be damaged;

[0030] (2) After the iris cutting mechanism cuts the tea leaves, the vacuum negative pressure machine generates negative pressure in the negative pressure box, and negative pressure is also generated in the fresh leaf storage box. The tea leaves enter the fresh leaf storage box through the installation hole on the connecting pipe and the leaf pipe, and the storage of the tea leaves is completed;

[0031] (3) A detection sensor is provided to detect whether the tea leaves are full;

[0032] (4) When the fresh leaf storage box is full of fresh leaves, the leaf discharging device is activated, and the tea picking robot resets and moves to the "fresh leaf collection point". The linear electric push rod drives the fresh leaf storage box to move outward, and then the electric rotating table 1 drives the fresh leaf storage box to rotate 180°. Then the pushing device pushes the tea leaves into the "region setting point", completing the automatic fresh leaf transportation to the "region setting point". BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a schematic diagram of the structure of the tea picking robot of the present invention;

[0034] Figure 2 It is a schematic diagram of the structure of the tea picking robot picking device;

[0035] Figure 3 It is a structural diagram of the picking mechanism;

[0036] Figure 4 This is an exploded view of the iris cutting mechanism;

[0037] Figure 5 is a structural schematic diagram of the iris cutting mechanism;

[0038] Figure 6 The figure is a cutting process diagram of the iris cutting mechanism;

[0039] Figure 7 It is a structural schematic diagram of a vacuum negative pressure recovery device I;

[0040] Figure 8 It is the structural schematic diagram II of the vacuum negative pressure recovery device;

[0041] Fig. 9 It is a structural schematic diagram of the leaf ejection device;

[0042] Fig.10 It is a schematic diagram of the structure of the propulsion device;

[0043] Fig.11 It is a structural schematic diagram of a panoramic self-adjusting camera device;

[0044] Fig.12 This is a structural schematic diagram of a six-degree-of-freedom parallel platform.

[0045] In the figure: 1. Upper workbench; 2. Lower workbench; 3. Walking device; 4. Picking manipulator device;

[0046] 100, robotic arm; 101, picking mechanism; 102, tilting slide; 103, connecting pipe; 104, iris cutting mechanism; 107, connecting block; 108, mounting hole; 109, module camera; 110, picking monocular camera; 111, directional guide ring; 112, connecting fixing ring; 113, first positioning ring; 113, second positioning ring; 115, micro motor; 116, extension rod; 117, side extension groove; 118, driven arc gear; 119, driving arc gear; 120, tilted directional groove; 121, blade; 122, knife slider; 123, directional milling groove; 124, pin key; 125, vacuum negative pressure machine; 126, fresh leaf storage box; 127, negative pressure box; 128, negative Pressure cavity; 129, negative pressure hole; 130, fresh leaf cavity; 131, fresh leaf box opening; 132, air hole; 134, leaf inlet hole; 135, pushing device; 136, straight line electric push rod; 137, fresh leaf storage box fixing plate; 138, electric rotating table; 139, pushing plate; 140, knife bevel; 141, panoramic self-adjusting camera device; 142, supporting component; 143, stepping motor; 144, base connecting block; 145, platform fine-tuning mechanism; 146, servo bracket; 147, servo motor; 148, cover bottom; 149, binocular camera; 150, distance sensor; 151, infrared sensor; 152, full monocular camera; 153, servo connecting rod; 154, 2D radar device;

[0047] 200, six-degree-of-freedom parallel platform; 201, fixed platform; 202, moving platform; 203, parallel branch chain; 204, universal joint; 205, servo drive motor mechanism; 206, lead screw; 207, first linear guide rod; 208, second linear guide rod; 209, first positioning support block; 210, second positioning support block; 211, fixed ear seat. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0049] See also Figure 1-Figure 12 The present invention provides a technical solution for a tea picking robot picking device.

[0050] A tea picking robot picking device, comprising:

[0051] An upper workbench 1, which is mounted on a walking device 3;

[0052] A mechanical arm 100 fixed on the upper workbench 1;

[0053] The picking mechanism 101 includes an inclined slide 102, a connecting tube 103 and an iris cutting mechanism 104. The inclined slide 102 is fixed on the robot arm 100. One end of the connecting tube 103 is fixed to the inclined slide 102, and the other end is fixed to the iris cutting mechanism 104. A mounting hole 108 is provided on the side wall of the connecting tube 103, and a vacuum negative pressure recovery device is connected to the mounting hole 108.

[0054] The walking device 3 drives the upper workbench 1 to move to a suitable position, and then the robotic arm 100 drives the picking mechanism 101 to move, so that the tea leaves to be picked are inserted into the iris cutting mechanism 104. After the iris cutting mechanism 104 cuts the tea leaves, the vacuum negative pressure recovery device extracts the cut tea leaves from the connecting tube 103 by means of vacuum negative pressure, thereby realizing picking. Picking is very convenient and will not damage the tea leaves.

[0055] like Figure 3 As shown, the picking device also includes a module camera 109 and a picking monocular camera 110 which are connected to the outside of the connecting tube 103 through a connecting block 107. The module camera 109 is fixed to the top of the inside of the connecting tube 103. The lens of the picking monocular camera 110 is facing downward and is coplanar with the coaxial part of the fresh leaf picking mechanism 101.

[0056] like Figure 1 and Fig.11 As shown, the picking device also includes a panoramic self-adjusting camera device 141, which includes a supporting component 142, a stepper motor 143 is fixed to the upper end of the supporting component 142, a base connecting block 144 is fixed to the rotating shaft of the stepper motor 143, a platform fine-tuning mechanism 145 is fixed to the base connecting block 144, a servo bracket 146 is connected to the upper end of the platform fine-tuning mechanism 145, a servo motor 147 is fixed to the servo bracket 146, a servo connecting rod 153 is fixed to the rotating shaft of the servo motor 147, a cover bottom 148 is fixed to the servo connecting rod 153, a binocular camera 149, a distance sensor 150, an infrared sensor 151 and two full monocular cameras 152 are fixed to the cover bottom 148.

[0057] The rotation of the stepper motor 143 can make the panoramic self-adjusting camera device 141 rotate 360° in the XY axis plane, and the rotation of the servo motor 147 can make the binocular camera 149 rotate 270°.

[0058] like Figure 8As shown, the picking device also includes a 2D radar device 154, which is installed on the front side of the walking device 3. The 2D radar device 154 is equipped with 4 radar probes. The first and second radar probes on the right side monitor the obstacles on the right front side of the tea picking robot and the obstacles or pits directly under the right crawler suspension assembly and transmit feedback to the control system. The GPS locator is used to send and receive the position and direction information of the robot; the control system controls and determines the plateau mountain picking block of the robot according to the received sensor data.

[0059] Furthermore, when the tea-picking robot is performing tea-picking operations in the wild on plateau mountains, the panoramic self-adjusting camera device 141, under the coordinated rotation of the stepper motor 143 and the DS3225 servo motor 147, can realize the recognition and shooting of the panoramic area of ​​the tea-picking robot's surrounding environment.

[0060] Its binocular camera works:

[0061] 1. The driving path planning of the tea picking robot and the captured images of the tea ridges of large-leaf Pu'er tea and fresh leaves of high-end ancient Pu'er tea trees are transmitted to the image processor of the visual device. The image processor performs target recognition on the fresh leaves of large-leaf Pu'er tea in the images of tea ridges and ancient tea trees captured by the binocular camera, and transmits the position information of the identified fresh leaves of large-leaf Pu'er tea to the control system of the robot, so as to form an optimal solution picking path planning for the picking path of the fresh leaves in the regional image. The control system controls the robot arm 100 through the optimal solution picking path planning for the fresh leaves of large-leaf Pu'er tea, so that the robot arm 100 extends the Pu'er tea fresh leaf picking mechanism 101 to the side of the fresh tea leaves for preparatory picking;

[0062] 2. The driving path planning of the picking robot includes the path planning captured by the binocular camera and the 2D radar device 154. When the signals fed back to the control system are both "YES", the walking device 3 can be driven to perform the driving path planning movement.

[0063] The monocular camera 152 works as follows: 1. The monocular camera captures the Pu'er tea fresh leaf picking mechanism 101 of the robot arm 100 when it is placed next to the fresh tea leaves, so that its control system controls the robot arm 100 to be placed just above the axis of the large-leaf Pu'er tea fresh leaves, the monocular camera on the left captures the left robot arm, and the monocular camera on the right captures the right robot arm.

[0064] The distance sensor 150 and infrared sensor 151 are used for: 1. Collecting environmental information around the robot and transmitting it to the robot's control system. When the signals of the two sensors exceed the obstacle distance setting value, the tea picking robot directly turns to reduce computing power.

[0065] like Figure 3-Figure 6As shown, the iris cutting mechanism 104 includes a directional guide ring 111, a connecting and fixing ring 112, a first positioning ring 113, a second positioning ring 113, a blade 121 and a micro motor 115. The directional guide ring 111 is fixed on the connecting tube 103, the first positioning ring 113 is fixed between the directional guide ring 111 and the second positioning ring 113, the connecting and fixing ring 112 is arranged between the directional guide ring 111 and the second positioning ring 113, an extension rod 116 is fixed on the connecting and fixing ring 112, a side extension groove 117 is provided on the side wall of the first positioning ring 113, a driven arc gear 118 is fixed on one end of the extension rod 116 passing through the extension groove, and the micro motor 115 is fixed on the outer side of the connecting tube 103. On the wall, a driving arc gear 119 is fixed on the rotating shaft of the micro motor 115, and the driving arc gear 119 is meshed with the driven arc gear 118. A plurality of inclined directional grooves 120 are provided on the connecting fixed ring 112, and the plurality of inclined directional grooves 120 are evenly distributed around the axis of the connecting fixed ring 112. A blade 121 is connected to each inclined directional groove 120, and a knife slider 122 plugged into the inclined directional groove 120 is fixed on the blade 121. A directional milling groove 123 corresponding to the inclined directional groove 120 is provided on the directional guide ring 111, and a pin key 124 plugged into the directional milling groove is provided on the blade 121. Two knife bevels 140 are provided on the blade 121, and adjacent knife bevels 140 fit each other.

[0066] like Figure 3-Figure 6 As shown, when picking, the fresh leaves enter the connecting tube 103, the micro motor 115 drives the driving arc gear 119, the driving arc gear 119 drives the driven arc gear 118 to rotate, thereby driving the connecting fixing ring 112 to rotate, and the rotation of the connecting fixing ring 112 can drive the following Figure 6 The way of rotation, Figure 6 The blade 121 gradually closes from left to right to complete the picking of fresh leaves.

[0067] like Figure 7-Figure 10 As shown, the vacuum negative pressure recovery device includes a vacuum negative pressure machine 125, a fresh leaf storage box 126 and a negative pressure box 127. The fresh leaf storage box 126 and the negative pressure box 127 are arranged below the upper workbench 1. A negative pressure cavity 128 is arranged in the negative pressure box 127. A side wall of the negative pressure cavity 128 is provided with a plurality of negative pressure holes 129. A fresh leaf cavity 130 is arranged in the fresh leaf storage box 126. A fresh leaf box opening 131 is arranged on one side of the fresh leaf storage box 126. The fresh leaf box opening 131 faces the negative pressure hole 129. An air hole 132 is arranged on the upper workbench 1. The air hole 132 is connected with the negative pressure cavity 128. The air hole 132 and the vacuum negative pressure machine 125 are connected through a first air pipeline. A leaf inlet hole 134 is arranged on the upper workbench 1 above the fresh leaf storage box 126. The leaf inlet hole 134 is connected with the fresh leaf cavity, and the leaf inlet hole 134 is connected with the mounting hole 108 through a leaf pipeline.

[0068] When the iris cutting mechanism 104 cuts the fresh leaves, the vacuum negative pressure machine 125 generates negative pressure in the negative pressure box 127. At this time, negative pressure is also generated in the fresh leaf storage box 126. The tea leaves enter the fresh leaf storage box 126 through the installation hole 108 on the connecting tube 103 and the leaf duct, completing the storage of the tea leaves.

[0069] like Figure 7-Figure 10 As shown, the picking device also includes a leaf discharging device and a lower workbench 2. The lower workbench 2 is arranged below the upper workbench 1. The leaf discharging device includes a pushing device 135 and a telescopic rotating device. The pushing device 135 is installed in the fresh leaf cavity 130. The pushing device 135 is arranged at an end away from the fresh leaf box opening 131. A pushing plate 139 is fixed on the pushing device 135. The telescopic rotating device includes an extending linear electric push rod 136, a fresh leaf storage box fixing plate 137 and an electric rotating table 138. The linear electric push rod is fixed on the lower workbench 2, the fresh leaf storage box fixing plate 137 is fixed on the push rod of the linear electric push rod, the electric rotating table 138 is fixed on the fresh leaf storage box fixing plate 137, and the fresh leaf storage box 126 is fixed on the electric rotating table 138.

[0070] The picking device also includes a detection sensor, which is fixed in the fresh leaf cavity 130. The detection sensor is arranged on a side close to the pushing device 135 and uses red light detection to detect whether the tea leaves are full.

[0071] Furthermore, when the fresh leaves in the fresh leaf storage box 126 reach the red light measuring point, the signal is transmitted to the control system, and the tea picking robot resets and drives to the "fresh leaf collection point", and the leaf discharge device is activated. The linear electric push rod drives the fresh leaf storage box 126 to move outward, and then the electric rotating table 138 drives the fresh leaf storage box 126 to rotate 180 degrees, and then the pushing device 135 pushes the tea leaves into the "region setting point".

[0072] like Fig.10 As shown, the pushing device 135 is configured as a linear motor scissor mechanism, which includes a pushing plate 139 .

[0073] like Figure 2 As shown, the robot arm 100 includes a six-degree-of-freedom robot arm, which includes a one-axis motor base, a two-axis robot arm, a three-axis robot arm, a four-axis robot arm and a five-axis robot arm joint, which are connected in sequence. The one-axis motor base is fixed to the center of the fixed large plate, and the first motor is arranged therein. The two-axis robot arm is connected to the driving arm clamping plate arranged on the one-axis motor base, and the driving arm clamping plate is equipped with a robot arm control circuit board. There are steering gears, reducers and potentiometers at the joints where the robot arms are connected.

[0074] The following is another embodiment of the robot arm 100: Figure 1 and Fig.12As shown, the robot arm 100 includes a six-degree-of-freedom parallel platform 200 and a picking robot device 4, the six-degree-of-freedom parallel platform 200 includes a fixed platform 201 and a moving platform 202, three groups of parallel branch chains 203 are evenly installed on the upper end surface of the fixed platform 201 along the axial direction, each group of parallel branch chains 203 is composed of two parallel branch chains 203, the parallel branch chains 203 include a universal joint 204, a servo drive motor mechanism 205, a screw rod 206, two first linear guide rods 207, two second linear guide rods 208, a first positioning support block 209, a second positioning support block 210, and a fixed ear seat 211, one end of the universal joint 204 is fixed on the fixed platform 201, and the other end is rotationally connected to the servo drive motor mechanism 205 Then, the rotating shaft of the servo drive motor mechanism 205 is fixed to the screw rod 206, one end of the two first linear guide rods 207 is fixedly connected to the servo drive motor mechanism 205, and the other end is fixedly connected to the second positioning support block 210, the two first linear guide rods 207 both pass through the first positioning support block 209, one end of the two second linear guide rods 208 is fixedly connected to the first positioning support block 209, and the other end is fixed to the fixed ear seat 211, the screw rod 206 passes through the first positioning support block 209 and is threadedly connected thereto, the top end of the fixed ear seat 211 is fixed to the lower end surface of the moving platform 202 by a rotating pair, the picking robot device 4 is fixed on the moving platform 202, and the picking mechanism 101 is fixed on the picking robot device 4.

[0075] Three groups of parallel branch chains 203 are arranged at 120° azimuth angles on the fixed platform 201, and three groups of points where the three groups of parallel branch chains 203 and their adjacent groups of parallel branch chains 203 are connected to the lower end surface of the moving platform 202 are arranged at 120° azimuth angles on the lower end surface of the moving platform 202, so that the lower connection point of the parallel branch chains 203 forms an azimuth angle of 60° in the top view.

[0076] The present invention controls the servo motor 147 by connecting the controller with the control wire to the servo drive motor mechanism 205 at the outer side of the housing to rotate the screw rod 206, so that the first positioning support block 209 moves up and down along the second linear guide rod 208, thereby driving the fixed ear seat 211 to move.

[0077] When the six parallel branches 203 of the three parallel branches 203 groups and the moving platform 202 form a 6-RPC parallel mechanism and adopt a two-stage distributed guide rod directional transmission form, the moving platform 202 has the advantages of good dynamic response, good stability, high rigidity, large load-bearing capacity and high motion accuracy under the control of the controller signal.

[0078] The present invention uses the six-degree-of-freedom parallel platform 200 to enable the picking manipulator device 4 on the moving platform 202 to operate the robot in the field of the Yunnan plateau mountain and to pick the fresh leaves of the ancient tea trees of the large-leaf Pu'er tea in the field, which has the following advantages:

[0079] The picking manipulator device 4 platform is in the mountainous terrain of the Yunnan Plateau. During the operation of the robot, its path planning activities must pass through a large slope road (at least 15° to a maximum of 60°). When a six-degree-of-freedom parallel platform 200 is used, its fixed platform 201 is fixed on the workbench, and the controller controls the six parallel branches 203 to keep its moving platform 202 horizontal, so that its influence on the servo motor 147 of the picking manipulator device 4 for high-precision picking of high-value wild large-leaf Pu'er tea fresh leaves is reduced to zero, and it maintains adaptive long-term work in an unmanned environment in the wild. The robot does not cause damage to the mechanical arm 100 due to the continuous movement of the robot, affecting its high-quality fresh leaf picking.

[0080] Wild large-leafed Pu'er ancient tea trees are distributed in the wild mountains of high-altitude plateaus in southwest Yunnan. Ancient tea trees over 100 years old are often higher than 2.1 meters to 5.2 meters, or even higher. Its fresh leaves sprout in spring every year and are not distributed in a three-dimensional shape on the crown of the tea tree. Its fresh leaves are large-leafed and sun-oriented. They are not vertically upward on the 1.2m tea ridge like the fresh leaves of Longjing tea, but are distributed in all directions and irregularly. Therefore, the six-degree-of-freedom parallel platform 200 of this device can keep the picking manipulator device 4 horizontal while the robot is moving, so that the picking manipulator device 4 can pick more efficiently, and the driving amount of each manipulator 100 of the six-degree-of-freedom serial manipulator 100 is minimized when planning to pick each fresh leaf, and the corresponding driver capacity is reduced to, which is divided into two parts of distributed control.

[0081] The picking manipulator device 4 is arranged on the moving platform 202, and the picking manipulator device 4 will not tilt with the tea picking robot due to the bumpy terrain or climbing. The picking manipulator device 4 includes a high six-degree-of-freedom mechanical arm 100 and a Pu'er tea fresh leaf picking mechanism 101. The picking manipulator device 4 is similar to the mechanical structure of the picking manipulator device 4, but the motor and the mechanical arm 100 cooperate with each other more accurately and flexibly.

[0082] Tea picking steps:

[0083] 1. When working in the field, the distance sensor 150 and the infrared sensor 151 are used to collect environmental information around the robot and feed it back to the control system in the industrial computer of the electrical control layer of the robot. When the signals of the two sensors exceed the obstacle distance setting value, the tea-picking robot directly turns to reduce the computing power. Then, the panoramic self-adjusting camera device 141 can realize the recognition and shooting of the panoramic area of ​​the tea-picking robot's surrounding environment under the joint rotation of the stepper motor 143 and the DS3225 servo motor 147. When the path planning shot by the binocular camera and the signal fed back to the control system by the 2D radar device 154 are both "YES", the crawler chassis can be driven to perform the driving path planning movement. The 2D radar device 154 is equipped with 4 radar probes. The right first and right second radar probes respectively monitor the obstacles on the right front side of the tea-picking robot and the obstacles or pits directly under the right crawler suspension assembly and feed them back to the control system. When the size setting value of the obstacle determines to pass or bypass. The GPS locator is used to send and receive the position and direction information of the robot; the control system controls and determines the plateau mountain picking area of ​​the robot according to the received sensor data.

[0084] 2. When the robot arrives near the terraced Pu'er tea ridge, the picking manipulator device rotates to face the tea ridge surface on one side, and if the other side is an intercropped ancient tea tree or a terraced Pu'er tea ridge, the picking manipulator device 4 platform can rotate to face the crown of the ancient tea tree and the tea ridge surface. Its binocular camera works to transmit the captured images of the large-leaf Pu'er tea ridge and the fresh leaves of the high-end ancient Pu'er tea tree to the image processor of the visual device. The image processor performs target recognition on the large-leaf Pu'er tea fresh leaves in the tea ridge and ancient tea tree images captured by the binocular camera, and transmits the recognized large-leaf Pu'er tea fresh leaf position information to the robot's control system, forming an optimal solution picking path planning for the picking path of the fresh leaf position in the regional image, and the picking manipulator device 4 picks from the "starting end" of the optimal solution picking path of the captured regional image, and the picking manipulator device 4 picks from the "end" of the optimal solution picking path, and the picking manipulator device 4 platform picks from the crown of the ancient tea tree or tea ridge on the other side. The control system controls the robot arm 100 through the optimal picking path planning of large-leaf Pu'er tea fresh leaves, so that the robot arm 100 extends the end effector to the tea fresh leaves for preparation for picking. The monocular camera captures the image when the end effector of the robot arm 100 is placed next to the tea fresh leaves, so that the control system controls the robot arm 100 to place the Pu'er tea fresh leaf picking mechanism 101 just above the axis of the large-leaf Pu'er tea fresh leaves.

[0085] 3. The Pu'er tea fresh leaf picking mechanism 101 starts to operate. The Pu'er tea fresh leaf picking mechanism 101 is suitable for picking large-leaf Pu'er tea fresh leaves from terraced tea and ancient tea trees with one bud and one leaf or one bud and two leaves. (Note: The shape of large-leaf Pu'er tea leaves is very different from that of famous and high-quality tea (such as Longjing tea leaves). The length of one bud and one leaf and one bud and two leaves of famous and high-quality tea leaves does not exceed 2.8cm, the radius of the cylinder from the bud core to the second leaf does not exceed 1.7cm, and the maximum width of the leaf does not exceed 0.5cm, showing a short, upright cylindrical body facing the sun. The length of one bud and one leaf and one bud and two leaves of large-leaf Pu'er tea leaves is 5cm-7.5cm, the radius of the cylinder from the bud core to the first leaf exceeds 2.5cm, the radius of the cylinder from the bud core to the second leaf exceeds 4.5cm, and the maximum width of the leaf exceeds 4.2cm. And because the size of Pu'er tea leaves is large, the fresh leaves of the plant face the sun, and the fresh leaves are bent towards the sun, which makes it impossible to use mechanized picking of bulk low-end tea. There is currently no patented design method to achieve picking.)

[0086] The one-bud-one-leaf mode of picking large-leaf Pu'er tea fresh leaves of the present invention is as follows: the mechanical arm 100 continues to move the Pu'er tea fresh leaf picking mechanism 101 closer to the target one-bud-one-leaf, and the S900 industrial module camera 109 identifies: the distance between the tip dot of the bud core and the tip dot of the first leaf; the distance between the tip dot of the bud core and the tip dot of the second leaf; the azimuth of the first leaf and the second leaf in the one bud. The distance between the tip dot of the bud core and the tip dot of the first leaf is compared to the control system through the image processor to see if the radius size of the cutter of the iris cutting mechanism 104 is less than. If "YES", command 1 is executed: the picking monocular camera 110 directly identifies the picking point 5mm below a leaf, and controls the robotic arm 100 to place the blade 121 of the iris cutting mechanism 104 of the Pu'er tea fresh leaf picking mechanism 101 at the picking point (the picking monocular camera 110 identifies the feedback to make the cutting blade of the iris cutting mechanism 104 perpendicular to the tender stem). At this time, the fresh leaves with one bud and one leaf are placed in the Pu'er tea fresh leaf picking mechanism 101, and the iris cutting mechanism 104 is driven by the micro motor 115 to close the blade 121 to achieve the cutting of one bud and one leaf. At this time, the fresh leaves are adsorbed into the fresh leaf storage box 126 through the flexible telescopic tube by the vacuum negative pressure recovery device. If "NO", execute command 2: The picking monocular camera 110 first identifies 2mm below a bud, and the control system drives the robot arm 100 to make the end face of the directional guide ring 111 of the Pu'er tea fresh leaf picking mechanism 101 and the iris cutting mechanism 104 coplanar with 2mm below a bud. During this process, the picking monocular camera 110 continues to shoot and identify. Then the control system controls the robot arm through the azimuth angle of the tip of the bud core of a leaf identified by the S900 industrial module camera 109 and links the sixth axis rotation motor to achieve the azimuth angle. Then the control system controls the GFX40 tilting slide 102 to rotate the slide surface (rotation angle ±20°), and the iris cutting mechanism 104 also rotates accordingly. At this time, the leaves 2mm below a bud within the Pu'er tea fresh leaf picking mechanism 101 also bend in the same direction, while the tender tendons below 2mm will not bend. During the rotation of the slide, when the S900 industrial module camera 109 recognizes that the distance from the tip dot of the bud core to the tip dot of a leaf is less than the radius size of the cutter of the iris cutting mechanism 104, the control system controls the robot arm 100 downward to 5mm below a leaf, and then performs the step of the control system command one.

[0087] The one-bud-two-leaf mode of picking large-leaf Pu'er tea fresh leaves of the present invention is as follows: the action command of one bud and one leaf in the one-bud-two-leaf picking mode is the same as the above-mentioned "one bud-one-leaf" action command, and the iris cutting mechanism 104 does not perform cutting at the 5mm above the one-bud-one-leaf picking point within the Pu'er tea fresh leaf picking mechanism 101 (the action form at this stage is that the central axis of the Pu'er tea fresh leaf picking mechanism 101 is colinear with the central axis of one bud and two leaves). Then the industrial computer control system uses the S900 industrial module camera 109 to capture and identify the azimuth of two leaves in one bud, and executes command three through the control system:

[0088] During this process, the monocular camera 538 continues to shoot and identify, and then the control system controls the robot and links the sixth-axis rotation motor. Then the control system controls the GFX40 tilt slide 102 to rotate the slide surface (rotation angle ±20°). At this time, the iris cutting mechanism 104 also rotates together. At this time, the one bud and one leaf within the Pu'er tea fresh leaf picking mechanism 101 5mm below the one leaf also bend in the same direction, and the angle bending 5mm below the one leaf will not cause the slight stem under the two leaves to bend in the same direction. During the rotation of the slide, when the S900 industrial module camera 109 recognizes that the distance between the tip dot of the bud core and the tip dot of the two leaves is less than the radius size of the cutter of the iris cutting mechanism 104, the control system controls the robot arm 100 downward to the one bud and two leaves picking point 8mm below the two leaves, and then performs the step of control system command one.

[0089] The one-bud-two-leaf mode of picking large-leaf Pu'er tea fresh leaves of the present invention is as follows: the action command of one bud and one leaf in the one-bud-two-leaf picking mode is the same as the above-mentioned "one bud-one-leaf" action command, and the iris cutting mechanism 104 does not perform cutting at the 5mm above the one-bud-one-leaf picking point within the Pu'er tea fresh leaf picking mechanism 101 (the action form at this stage is that the central axis of the Pu'er tea fresh leaf picking mechanism 101 is colinear with the central axis of one bud and two leaves). Then the industrial computer control system uses the S900 industrial module camera 109 to capture and identify the azimuth of two leaves in one bud, and executes command three through the control system:

[0090] During this process, the picking monocular camera 110 continues to shoot and identify, and then the control system controls the manipulator and links the sixth axis rotation motor. Then the control system controls the GFX40 tilt slide 102 to rotate the slide surface (rotation angle ±20°). At this time, the iris cutting mechanism 104 also rotates together. At this time, the one bud and one leaf within the Pu'er tea fresh leaf picking mechanism 101 5mm below the one leaf also bend in the same direction, and the angle bending 5mm below the one leaf will not cause the slight stem under the two leaves to bend in the same direction. During the rotation of the slide, when the S900 industrial module camera 109 recognizes that the distance between the tip dot of the bud core and the tip dot of the two leaves is less than the radius size of the cutter of the iris cutting mechanism 104, the control system controls the manipulator 100 to move it downward to the one bud and two leaves picking point 8mm below the two leaves, and then performs the step of control system command one.

[0091] 4. When the ancient tea trees or tea ridges on both sides of the binocular project are found to have fresh tea leaves that can no longer be picked, the robot crawler uses various sensors to detect that the surrounding environment is safe, and then feeds back to the control system to drive the crawler chassis to pick the next optimal solution path for the images taken on both sides, repeating steps 2, 3, and 4.

[0092] 5. When the fresh leaves in the fresh leaf storage box 126 reach the red light measuring point, the signal is transmitted to the control system, and the tea picking robot resets and drives to the "region setting point". Then the control system drives the fresh leaf storage and discharging device to discharge the fresh leaves.

[0093] A fresh leaf identification method is also disclosed, based on the above tea picking robot picking device, comprising the following steps:

[0094] S1. Take photos of fresh leaves under different backgrounds and light intensities, and collect various fresh leaf images as the initial data set;

[0095] After collecting the fresh leaf image data, the image is annotated. The result of the annotation is to use a rectangular frame to frame the fresh leaves, which is also the object of the fresh leaf picking mechanism. The name, size and position information of the fresh leaf tea type in the image are obtained in TXT and XML files.

[0096] S2. Divide the collected fresh leaf image data into training set, test set and validation set, and construct the labeled image data set into training set, test set and validation set in a ratio of 6:2:2 and put them into the improved YOLOV7 algorithm for learning, so that the improved YOLOV7 algorithm finally obtains the Feature Maps of fresh leaves.

[0097] The improved YOLOv7 network structure consists of Input, Backbone, Neck and Head, where Input is mainly used for data preprocessing, including data enhancement, image size scaling, and predefined candidate box size calculation; Backbone is the backbone network of deep learning, which is used to realize the feature extraction function of the target; Neck is the neck network that connects feature layers of different scales and performs feature fusion; Head is the head network, which calculates the regression loss value through the loss function. The present invention adopts the latest MPDIoU loss function to improve the original loss function. In order to be applicable to the regression of overlapping and non-overlapping bounding boxes at the same time, and to take into account the center point distance and the deviation of width and height, MPDIOU adopts the bounding box similarity metric based on the minimum point distance, which not only simplifies the calculation process to a certain extent, improves the model convergence speed, but also makes the regression result more accurate. Its loss function expression is shown in formulas (1), (2), (3) and (4):

[0098]

[0099]

[0100]

[0101]

[0102] Where A and B represent the predicted box and the true box respectively. and Respectively represent the coordinates of the upper left corner and lower right corner of the bounding box A, and Respectively represent the coordinates of the upper left corner and lower right corner of the bounding box B;

[0103] S3, the image obtained by the binocular camera 149 transmits the fresh leaf image to the industrial computer in real time, and the industrial computer performs noise reduction, filtering, and grayscale image preprocessing on the image transmitted by the camera; then the trained improved YOLOV7 learning model is used to recognize the photos transmitted by the camera.

[0104] The image obtained by the binocular camera can be divided into two images, left and right. After the camera calibration is completed, the two images need to be merged into one image. Spatial and Channel reconstruction Convolution is used to effectively suppress the redundant features of the feature map through the two components SRU (Spatial Reconstruction Unit) and CRU (Channel Reconstruction Unit), thereby reducing the complexity and computational cost of the model. Spatial and Channel reconstruction Convolution performs stereo matching to obtain the disparity map of the large-leaf Pu'er tea fresh leaves. The image taken by the binocular camera contains the position information (u, v) of the large-leaf Pu'er tea fresh leaves in the image, and after segmenting the disparity map, the depth value d of the large-leaf Pu'er tea fresh leaves can be obtained. Based on this information, a three-dimensional point cloud map of the large-leaf Pu'er tea fresh leaves is established, and then the point cloud map is subjected to denoising processing by direct filtering, statistical filtering and radius filtering, and a three-dimensional point cloud of large-leaf Pu'er tea fresh leaves with less noise and smoothness can be obtained.

[0105] The separation operation of SRU mainly uses the scaling factor of GN (Group Normalization) to evaluate the information content of the feature map in order to better separate the feature maps with rich information and those with less information. Its expression is shown in formula (5). The reconstruction operation combines the features with rich information with the features with less information through cross reconstruction and strengthens the information flow between the two to generate richer information features and save space. Its expression is shown in formula (6).

[0106]

[0107]

[0108] in, It means element-wise multiplication. represents element-by-element addition, U represents concatenation, μ and σ are the mean and standard deviation of X, ε is a small positive number added to stabilize division, γ and β are trainable affine transformations, W is the weight value of the feature map, W1 is the weight with richer information, and W2 is the weight with less rich information.

[0109] The Split operation of CRU is to divide the spatial refinement features generated by SRU into two parts, Xup and Xlow, and compress them respectively using 1x1 convolution to improve the computational efficiency of the model. The Transform operation is to extract features from Xup and Xlow obtained by the split operation using different convolutions, thereby obtaining two sets of feature maps with different information richness. The expressions are shown in equations (7) and (8). The Fusion operation extracts the spatial channel information of feature maps Y1 and Y2 by Pooling, merges features Y1 and Y2 in the form of channels, and generates Channel-Refined FeafureY. The expressions are shown in equations (9), (10), and (11).

[0110]

[0111]

[0112]

[0113]

[0114] Y=β1Y1+β2Y2 (11)

[0115] Among them, MG, MP1 and MP2 are the learnable weight matrices in the convolution operation, and β1 and β2 are the feature importance vectors.

[0116] After the fresh leaf image is divided into S×S non-overlapping regions, the feature vector contained in each region is Where H is the height of the original image, W is the width of the original image, and Q, K, and V are obtained by linearly mapping the feature vector. Their expressions are shown in formula (12), where XT represents the segmented input image, Wq, Wk, and Wv represent the weighted projections of query, key, and value, respectively. The region-level features are calculated by average pooling, and the average value of each region is calculated. Calculate the adjacency matrix of the inter-region correlation of Qr and Kr, which is expressed as formula (13)

[0117] As shown in the figure, Ar represents the adjacency matrix of correlation, Qr represents the query at the regional level, Kr represents the key at the regional level, and T represents the transposition operation. The coarse-grained regional-level routing calculation uses the routing index matrix The indexes of the first K links are saved row by row so that only the first K connections in each region are used when pruning the correlation graph. The expression is shown in formula (14). The public key normalization operation is to aggregate the key and value tensors. The aggregation formulas are shown in formulas (15) and (16). Among them, Kg represents the tensor after key aggregation, K represents the key, Ir represents the routing index matrix, Vg represents the tensor after value aggregation, and V represents the value. Collecting the scattered key-value pairs is to use the attention operation on the aggregated KV pairs to perform fine-grained label-to-label attention calculation. The expression is shown in formula (17), where O represents fine-grained label-to-label attention and LCE(V) represents the local context enhancement term.

[0118] Q=X r W q , K = X r W k , V = X r W, (12)

[0119] A r =Q r (K r ) T (13)

[0120] I r =topkIndex(A r ) (14)

[0121] K g =gather(K,I r ) (15)

[0122] V g =gather(V,I r ) (16)

[0123] O = Attention(Q, K g , V g )+LCE(V) (17)

[0124] In order to avoid the network overfitting caused by insufficient number of training images, the visual recognition of the present invention uses image enhancement technology to expand the original data. The original data is expanded by 11 times by means of Cropping, Rotate, Local enlargement, Change exposure and Adding Gaussian noise. The above-mentioned images are the data sets used for visual recognition of the present invention. Then the photos are input into the computer for image preprocessing, including image denoising and filtering, and then the color difference of the image is enhanced. In the present invention, the visual recognition uses the Labelimg tool to annotate the images in the data set after the processed images, and generates TXT and XML files after the annotation is completed.

[0125] Although 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 the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A tea picking robot picking device, characterized in that: include: Go to the workbench; A mechanical arm fixed to the upper workbench; A picking mechanism, the picking mechanism comprising an inclined slide, a connecting tube and an iris cutting mechanism, the inclined slide is fixed on the mechanical arm, one end of the connecting tube is fixed to the inclined slide, and the other end is fixed to the iris cutting mechanism, a mounting hole is provided on the side wall of the connecting tube, and a vacuum negative pressure recovery device is connected to the mounting hole; The vacuum negative pressure recovery device comprises a vacuum negative pressure machine, a fresh leaf storage box and a negative pressure box, the fresh leaf storage box and the negative pressure box are arranged below the upper workbench, a negative pressure cavity is arranged in the negative pressure box, a side wall of the negative pressure cavity is arranged with a plurality of negative pressure holes, a fresh leaf cavity is arranged in the fresh leaf storage box, a fresh leaf box opening is arranged on one side of the fresh leaf storage box, the fresh leaf box opening faces the negative pressure hole, an air hole connected to the negative pressure cavity is arranged on the upper workbench, the air hole and the vacuum negative pressure machine are connected through a first air pipeline, a leaf inlet hole is arranged on the upper workbench above the fresh leaf storage box, and the leaf inlet hole is connected to the mounting hole through a leaf pipeline; It also includes a leaf ejecting device and a lower workbench, the lower workbench is arranged below the upper workbench, the leaf ejecting device includes a pushing device and a telescopic rotating device, the pushing device is installed in the fresh leaf cavity, the pushing device is arranged at an end away from the opening of the fresh leaf box, the telescopic rotating device includes an extending linear electric push rod, a fresh leaf storage box fixing plate and an electric rotating table, the linear electric push rod is fixed on the lower workbench, the fresh leaf storage box fixing plate is fixed on the push rod of the linear electric push rod, the electric rotating table is fixed on the fresh leaf storage box fixing plate, and the fresh leaf storage box is fixed on the electric rotating table.

2. A tea picking robot picking device according to claim 1, characterized in that: It also includes a module camera and a picking monocular camera. The picking monocular camera is connected to the outside of the connecting tube through a connecting block, and the module camera is fixed at the top end inside the connecting tube.

3. A tea picking robot picking device according to claim 1, characterized in that: The iris cutting mechanism includes a directional guide ring, a connecting and fixing ring, a first positioning ring, a second positioning ring, a blade and a micro motor, wherein the directional guide ring is fixed on the connecting tube, the first positioning ring is fixed between the directional guide ring and the second positioning ring, the connecting and fixing ring is arranged between the directional guide ring and the second positioning ring, an extension rod is fixed on the connecting and fixing ring, a side extension groove is provided on the side wall of the first positioning ring, a driven arc gear is fixed on one end of the extension rod passing through the side extension groove, the micro motor is fixed on the outer wall of the connecting tube, A driving arc gear is fixed on the rotating shaft of the micro motor, and the driving arc gear is meshed with the driven arc gear. A plurality of inclined directional grooves are provided on the connecting and fixing ring, and the plurality of inclined directional grooves are evenly distributed around the axis of the connecting and fixing ring. A blade is connected to each of the inclined directional grooves, and a knife slider plugged into the inclined directional groove is fixed on the blade. A directional milling groove corresponding to the inclined directional groove is provided on the directional guide ring, and a pin key plugged into the directional milling groove is provided on the blade. Two knife bevels are provided on the blade, and adjacent knife bevels fit each other.

4. A tea picking robot picking device according to claim 1, characterized in that: It also includes a detection sensor, which is fixed in the fresh leaf cavity and arranged on a side close to the pushing device.

5. The tea picking robot picking device according to claim 1, characterized in that: The pushing device is configured as a linear motor scissor-type mechanism.

6. A tea picking robot picking device according to claim 2, characterized in that: It also includes a panoramic self-adjusting camera device, which includes a supporting component, a stepper motor is fixed to the upper end of the supporting component, a base connecting block is fixed to the rotating shaft of the stepper motor, a platform fine-tuning mechanism is fixed to the base connecting block, the upper end of the platform fine-tuning mechanism is connected to a servo bracket, a servo motor is fixed to the servo bracket, a servo connecting rod is fixed to the rotating shaft of the servo motor, a cover bottom is fixed to the servo connecting rod, and a binocular camera, a distance sensor, an infrared sensor and two full monocular cameras are fixed to the cover bottom.

7. The tea picking robot picking device according to claim 1, characterized in that: The robotic arm includes a six-degree-of-freedom parallel platform and a picking robot device, the six-degree-of-freedom parallel platform includes a fixed platform and a moving platform, three groups of parallel branch chain groups are evenly installed on the upper end surface of the fixed platform along the axial direction, each group of the parallel branch chain groups is composed of two parallel branch chains, the parallel branch chains include a universal joint, a servo drive motor mechanism, a screw, two first linear guide rods, two second linear guide rods, a first positioning support block, a second positioning support block, and a fixed ear seat, one end of the universal joint is fixed on the fixed platform, and the other end is rotatably connected to the servo drive motor mechanism, and the rotating shaft of the servo drive motor mechanism is connected to the The screw is fixed, one end of the two first linear guide rods is fixedly connected to the servo drive motor mechanism, and the other end is fixedly connected to the second positioning support block, the two first linear guide rods all pass through the first positioning support block, one end of the two second linear guide rods is fixedly connected to the first positioning support block, and the other end is fixed to the fixed ear seat, the screw passes through the first positioning support block and is threadedly connected thereto, the top end of the fixed ear seat is fixed to the lower end surface of the moving platform by a rotating pair, the picking robot device is fixed on the moving platform, and the picking mechanism is fixed to the picking robot device.

8. A method for identifying fresh leaves, based on a tea picking robot picking device according to any one of claims 1 to 7, characterized in that: The following steps are involved: S1, after collecting fresh leaf image data, annotate the image, obtain the name of the fresh leaf tea type, size and the position information of the fresh leaf in the image in TXT and XML files, S2. Divide the collected fresh leaf image data into training set, test set and validation set, construct the labeled image data set into training set, test set and validation set and put them into the improved YOLOV7 algorithm for learning, so that the improved YOLOV7 algorithm finally obtains the feature map of fresh leaves. The improved YOLOv7 network structure consists of Input, Backbone, Neck and Head. Its loss function expression is shown in equations (1), (2), (3) and (4): Where A and B represent the predicted box and the true box respectively. and Respectively represent the coordinates of the upper left corner and lower right corner of the bounding box A, and Respectively represent the coordinates of the upper left corner and lower right corner of the bounding box B; S3. The image obtained by the binocular camera transmits the fresh leaf image to the industrial computer in real time. The industrial computer performs noise reduction, filtering, and grayscale image preprocessing on the image transmitted by the camera; then the trained improved YOLOV7 learning model is used to recognize the photos transmitted by the binocular camera.

Images