A multi-layer three-dimensional cultivation automatic transplanting robot for a plant factory

Abstract

The present invention relates to the field of agricultural automation technology, and specifically refers to a multi-layer three-dimensional cultivation automatic transplanting robot for a plant factory, which includes a cultivation rack. A seedling plate is arranged on the side of the cultivation rack. The robot includes a control module, a vision module, a positioning module, a moving module, and a clamping module. The present invention integrates a clamping module, including a clamping unit and a clamping seat. The clamping seat is fixedly connected to a moving seat. A plurality of lifting units are arranged below the clamping seat. The clamping unit includes a clamping block detachably connected to the lifting unit. A clamping motor is arranged in the clamping block. A plurality of clamping grooves are formed on the lower end surface of the clamping block. Claw jaws are movably arranged in the clamping grooves. A crank is also arranged in the clamping block. Two ends of the crank are respectively connected to the output end of the clamping motor and the claw jaws, realizing the opening and closing actions of the claw jaws. This design enables the robot to accurately clamp and release the planting basket, improving the accuracy and consistency of transplanting.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural automation, and specifically refers to a multi-layer three-dimensional cultivation automatic transplanting robot for a plant factory. Background Art

[0002] In a plant factory, the multi-layer three-dimensional cultivation mode can effectively improve the space utilization rate and yield. Since the cultivation layer is relatively high, there are problems such as great potential safety hazards, low efficiency, and high labor intensity in the transplanting process using manual transplanting compared with traditional transplanting, and it is difficult to ensure the accuracy and consistency of transplanting.

[0003] Existing automatic transplanting equipment is not suitable for the multi-layer three-dimensional cultivation mode. During the transplanting process, it is difficult to accurately position the planting basket into the cultivation hole, resulting in low efficiency. Therefore, an automatic transplanting robot is needed to improve production efficiency and quality. Summary of the Invention

[0004] The purpose of the present invention is to provide a multi-layer three-dimensional cultivation automatic transplanting robot for a plant factory, which is used to improve the transplanting speed, accuracy, and consistency.

[0005] The present invention is achieved through the following technical solutions:

[0006] A multi-layer three-dimensional cultivation automatic transplanting robot for a plant factory includes a cultivation rack. A seedling-raising plate is arranged on the side of the cultivation rack. The robot includes a control module, a vision module, a positioning module, a moving module, and a clamping module. The moving module includes a moving seat, a vertical frame, and a horizontal frame. The horizontal frame is slidably arranged on the vertical frame and is signal-connected to the control module. The moving seat is slidably arranged on the horizontal frame and is signal-connected to the control module. The vision module is signal-connected to the control module and is used for collecting position information. The positioning module is signal-connected to the control module and is used for determining position information. The clamping module is arranged on the moving seat and clamps and fixes the planting basket in the seedling-raising plate through the control module. Among them, the process of determining position information is as follows: The vision module identifies the position of the seedling for scanning to obtain 3D point cloud data, then processes the 3D point cloud data to extract the position information and shape information of the planting basket and the seedling, and establishes a coordinate system. Finally, the boundary position information of the planting basket is transmitted to the control module.

[0007] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0008] 1. The present invention integrates a clamping module, including a clamping unit and a clamping seat. The clamping seat is fixedly connected to the moving seat, and several lifting units are arranged below the clamping seat. The clamping unit includes a clamping block detachably connected to the lifting unit. A clamping motor is arranged inside the clamping block. Several clamping grooves are formed on the lower end surface of the clamping block, and clamping claws are movably arranged in the clamping grooves. A crank is also arranged inside the clamping block. Two ends of the crank are respectively connected to the output end of the clamping motor and the clamping claws, realizing the opening and closing actions of the clamping claws. This design enables the robot to accurately clamp and release the planting basket, improving the accuracy and consistency of transplantation.

[0009] 2. A crank is also arranged inside the clamping block of the present invention. Two ends of the crank are respectively connected to the output end of the clamping motor and the clamping claws. When the clamping motor is started, through the synchronous rotation of the crank, the opening and closing actions of the clamping claws are realized. Such a mechanical transmission design ensures the accuracy and reliability of the clamping action. In order to protect the seedlings from damage, force sensors are arranged on the inner sides of the clamping claws. These force sensors can real-time monitor the acting force of the clamping claws during the grasping process.

[0010] 3. The introduction of the state judgment model in the present invention enables the automatic transplanting robot to evaluate the growth state of the seedlings, which has guiding significance for determining whether to perform the transplanting operation. By obtaining the contour information of the seedling's lateral stem and main stem through OPENCV and combining with the range of the growth state information, the robot can judge whether the growth state of the seedlings is good, thereby avoiding transplanting the seedlings with poor growth state and ensuring that the transplanted seedlings have good growth potential. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, form a part of the present invention, and do not limit the embodiments of the present invention. In the drawings:

[0012] Figure 1 is a schematic structural diagram of the robot of the present invention;

[0013] Figure 2 is Figure 1 an enlarged structural diagram of A in

[0014] Figure 3 is a partial internal structural diagram of the clamping module;

[0015] Figure 4 is an internal structural diagram of the clamping unit;

[0016] Figure 5 is an axonometric structural diagram of a clamping unit of the present invention;

[0017] Figure 6 is another axonometric structural diagram of the clamping unit of the present invention;

[0018] Figure 7 This is a schematic structural diagram of the steering unit of the present invention.

[0019] Markings in the drawings and corresponding component names:

[0020] 1 - Cultivation rack, 2 - Seedling plate, 3 - Moving module, 4 - Clamping module, 5 - Planting basket;

[0021] 31 - Moving seat, 32 - Upright frame, 33 - Cross frame;

[0022] 41 - Clamping unit, 42 - Clamping seat, 43 - Lifting unit, 44 - Steering rod, 45 - Steering motor;

[0023] 411 - Clamping block, 412 - Clamping motor, 413 - Clamping groove, 414 - Claw, 415 - Crank, 416 - Clamping cover, 417 - Opening;

[0024] 431 - Lead screw. Detailed implementation manners

[0025] To make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in combination with embodiments and drawings. The illustrative embodiments of the present invention and their descriptions are only used to explain the present invention and are not intended to limit the present invention. It should be noted that the present invention has been in the actual research and development and use stage.

[0026] Embodiment 1:

[0027] Please refer to the attached Figures 1 to 6 , a multi - layer three - dimensional cultivation automatic transplanting robot for a plant factory, including a cultivation rack 1, a seedling plate 2 is arranged on the side of the cultivation rack 1. It is characterized in that: the robot includes a control module, a vision module, a positioning module, a moving module 3 and a clamping module 4. The moving module 3 includes a moving seat 31, an upright frame 32 and a cross frame 33. The cross frame 33 is slidably arranged on the upright frame 32 and is in signal connection with the control module. The moving seat 31 is slidably arranged on the cross frame 33 and is in signal connection with the control module. The vision module is in signal connection with the control module and is used for collecting position information. The positioning module is in signal connection with the control module and is used for determining position information. The clamping module 4 is arranged on the moving seat 31 and clamps and fixes the planting basket 5 in the seedling plate 2 through the control module; wherein, the process of determining the position information is as follows: the vision module identifies the position of the seedlings for scanning to obtain 3D point cloud data, then processes the 3D point cloud data to extract the position information and shape information of the planting basket 5 and the seedlings, and establishes a coordinate system. Finally, the boundary position information of the planting basket 5 is transmitted to the control module.

[0028] It should be noted that in a plant factory, the multi-layered three-dimensional cultivation mode can effectively improve the space utilization rate and yield. Since the cultivation layer is relatively high, there are relatively large potential safety hazards, low efficiency, high labor intensity, etc. in the transplanting process of this mode if manual transplanting is used, and it is difficult to ensure the accuracy and consistency of transplanting. The existing automated transplanting equipment is not suitable for the multi-layered three-dimensional cultivation mode, and it is difficult to accurately position the planting in the cultivation holes during the transplanting process, resulting in low efficiency. Therefore, an automated transplanting robot is needed to improve production efficiency and quality.

[0029] Based on the above problems, the applicant has proposed a multi-layered three-dimensional cultivation automated transplanting robot for a plant factory. By integrating a clamping module 4, which includes a clamping unit 41 and a clamping seat 42, the clamping seat 42 is fixedly connected to the moving seat 31. A number of lifting units 43 are arranged below the clamping seat 42. The clamping unit 41 includes a clamping block 411 detachably connected to the lifting unit 43. A clamping motor 412 is arranged inside the clamping block 411. A number of clamping grooves 413 are formed on the lower end surface of the clamping block 411. Claw 414 is movably arranged in the clamping groove 413. A crank 415 is also arranged inside the clamping block 411. The two ends of the crank 415 are respectively connected to the output end of the clamping motor 412 and the claw 414, realizing the opening and closing action of the claw 414. This design enables the robot to accurately clamp and release the planting basket 5, improving the accuracy and consistency of transplanting.

[0030] As shown in the appendix Figure 6 For the lifting unit 43, referring to the appendix Figure 3 , the lifting unit 43 includes a lead screw 431. The end of the lead screw 431 is connected to a lifting motor (not shown in the figure). The lifting motor is signal-connected to the control module. It should also be noted that a clamping cover 416 is arranged at the upper end of the clamping block 411. An opening 417 matching the lead screw 431 is arranged in the middle of the clamping cover 416.

[0031] More preferably, the moving module 3 includes an X-axis moving unit, a Z-axis moving unit, and a steering unit. The X-axis moving unit includes a cross frame 33, and the Z-axis moving unit includes a vertical frame 32. In this embodiment, the moving seat 31 is preferably slidably arranged on the cross frame 33 through an electric slide rail structure. Similarly, the cross frame 33 is slidably arranged on the vertical frame 32 through an electric slide rail. The clamping seat 42 is movably connected to the moving seat 31 through the steering unit. For the steering unit, it includes a steering rod 44 and a steering motor. One end of the steering rod 44 is fixedly connected to the clamping seat 42. The steering motor is arranged inside the moving seat 31 and its output end is connected to the other end of the steering rod 44. When the steering motor is started, the relative rotation of the clamping seat 42 and the moving seat 31 is realized through the steering unit. As shown in the appendix Figure 7As shown, the upper end surface of the moving seat 31 has a U-shaped opening, and the steering rod 44 is rotatably arranged in the U-shaped opening. A cavity is provided inside the moving seat 31, and the steering motor 45 is fixedly arranged in the cavity, and the output end of the steering motor 45 is fixedly connected to the other end of the steering rod 44. Another preferred embodiment in this embodiment is that an annular limiting groove is opened on the upper and lower end surfaces inside the U-shaped opening, and a pin matching the annular limiting groove is arranged on the steering rod 44 to prevent the center of gravity of the clamping unit 41 from being unbalanced during rotation. It can be understood that the extending direction of both ends of the horizontal frame 33 is the X-axis direction, the extending direction of both ends of the vertical frame 32 is the Z-axis direction, and the extending direction of the steering rod 44 when perpendicular to the vertical frame 32 is the Y-axis direction.

[0032] For the cultivation process:

[0033] The clamping module 4 receives instructions through the control module, the clamping motor 412 is started, and the crank 415 is driven to rotate synchronously, thereby controlling the opening and closing of the clamping jaw 414. When it is necessary to clamp the planting basket 5, the clamping jaw 414 is opened wide enough to accommodate the size of the planting basket 5, and then the motor drives the lead screw 431 to descend to the plane of the target planting basket 5. The annular clamping jaw 414 is closed, and the planting basket 5 is clamped inside the ring. Then, the control module controls the lifting motor to rotate so that the internal thread of the opening 417 matches the external thread of the lead screw 431, and the planting basket 5 is lifted from the seedling plate 2. Among them, the maximum lifting amount is the depth of the opening 417. Subsequently, the control module changes the height of the planting basket 5 through the Z-axis moving unit. When the planting basket 5 reaches the predetermined height, the control module changes the angle of the planting basket 5 relative to the cultivation rack 1 through the steering unit, that is, rotates the side close to the seedling plate 2 to the side away from the seedling plate 2, and detects whether it has moved to the predetermined position through the vision module, and detects the vacant cultivation holes on the cultivation rack 1, and repeats the above process to transplant the seedlings on the seedling plate 2 into the cultivation holes on the cultivation rack 1.

[0034] For the vision module, in this embodiment, its preferred structure includes a binocular vision camera and a multi-degree-of-freedom pan-tilt structure. More preferably, the camera is a FL3-U3-13S2C model camera. The vision module can be set below the clamping seat 42, on one side of the cross frame 33 close to the seedling plate 2, or on one side of the vertical frame 32 close to the seedling plate 2. In this embodiment, its preferred setting position is on one side of the cross frame 33 close to the seedling plate 2. The initial position of the clamping module 4 within the vision range of the vision module is when the vertical frame 32 is at the lowest point and the cross frame 33 is at the rightmost side (in terms of facing the clamping cultivation rack 1). For the positioning of the target position information, it uses 3D point cloud data calculation, which will not be elaborated here. For the control module, it is a high-performance microcontroller or embedded computer system. In this embodiment, it is preferably an ARM Cortex-A series processor, which has sufficient computing power to process data from the vision module and real-time control each module of the robot. This control module should have good interface scalability and be able to perform efficient data exchange and instruction transmission with the vision module, positioning module, mobile module 3, and clamping module 4. For the positioning module, it is preferably a lidar or a stereo vision system, which can provide high-precision 3D point cloud data for accurately positioning the planting basket 5 and the seedlings. In this embodiment, the Velodyne series of lidar is selected, which can generate an accurate three-dimensional map of the surrounding environment in real time with its high precision and fast scanning ability. For the mobile module 3, the preferred structure is an electric slide rail system, including a moving seat 31, a vertical frame 32, and a cross frame 33, which can achieve precise movement on the X and Z axes. The moving seat 31 and the cross frame 33 are driven by servo motors to ensure the accuracy and stability of the movement. The vertical frame 32 is made of high-strength materials to ensure the stability of the structure and the bearing capacity.

[0035] Embodiment 2:

[0036] This embodiment only describes the parts different from Embodiment 1. Specifically: The clamping module 4 includes a clamping unit 41 and a clamping seat 42. The clamping seat 42 is movably connected to the moving seat 31. Below the clamping seat 42, several lifting units 43 are provided. The clamping unit 41 includes a clamping block 411 detachably connected to the lifting unit 43. A clamping motor 412 is arranged inside the clamping block 411. A plurality of clamping grooves 413 are formed on the lower end surface of the clamping block 411. Claw 414 is movably arranged in the clamping groove 413. A crank 415 is also arranged inside the clamping block 411. The two ends of the crank 415 are respectively connected to the output end of the clamping motor 412 and the claw 414. When the clamping motor 412 is started, the opening and closing of the claw 414 are realized through the synchronous rotation of the crank 415. A force sensor is arranged inside the claw 414. The force sensor is signal-connected to the control module. A force feedback model is loaded in the control module. The force feedback model is used to monitor the acting force during the grasping process of the claw 414.

[0037] Based on the above mechanism, it can be understood that the technical problem addressed in this embodiment is how to achieve precise and safe clamping and transplanting operations for the planting basket 5, while ensuring the protection of seedlings during the transplanting process and improving the efficiency and accuracy of the operation. Through the design of the clamping module 4, this embodiment provides a solution with a complex structure but precise operation. More specifically, the clamping module 4 is composed of a clamping unit 41 and a clamping seat 42. The clamping seat 42 is fixedly connected to the moving seat 31, and several lifting units 43 are arranged below the clamping seat 42. The above structural design enables the clamping module 4 to not only move horizontally but also make precise position adjustments vertically; the clamping unit 41 is connected to the lifting unit 43 in a detachable manner, increasing the flexibility of the device and the convenience of maintenance. A clamping motor 412 is arranged inside the clamping block 411, which is the core power source for realizing the clamping action. A plurality of clamping grooves 413 are formed on the lower end surface of the clamping block 411, and clamping claws 414 are movably arranged in the clamping grooves 413. These clamping claws 414 are responsible for directly contacting the planting basket 5 to achieve the clamping function.

[0038] In addition, a crank 415 is also arranged inside the clamping block 411. The two ends of the crank 415 are respectively connected to the output end of the clamping motor 412 and the clamping claws 414. When the clamping motor 412 is started, the opening and closing actions of the clamping claws 414 are realized through the synchronous rotation of the crank 415. Such a mechanical transmission design ensures the precision and reliability of the clamping action; in order to protect the seedlings from damage, force sensors are arranged on the inner sides of the clamping claws 414. These force sensors can monitor the acting force of the clamping claws 414 during the grasping process in real time and transmit the data to the control module.

[0039] Embodiment 3:

[0040] This embodiment only describes the parts different from Embodiment 1. Specifically:

[0041] The force feedback model satisfies:

[0042] ;

[0043] ;

[0044] Wherein, is the control signal;

[0045] is the proportional gain, used to adjust the immediate response of the error;

[0046] is the integral gain, used to adjust the cumulative effect of the error;

[0047] is the differential gain, which is used to adjust the rate of change of the error;

[0048] is the target force, that is, the force applied by the gripper 414 set by the control module;

[0049] is the measured force, that is, the force actually applied by the gripper 414, which is measured by a force sensor;

[0050] is the error, that is, the difference between the target force and the measured force;

[0051] is the integral of the error, which represents the accumulation of the error from the start to the current moment;

[0052] is the rate of change of the error, which represents the rate of change of the error over time.

[0053] A safety threshold is set in the force sensor. When the measured force exceeds the safety threshold, the control module generates a feedback signal to reduce the actual force applied by the gripper 414.

[0054] It should be noted that a force feedback model is carried in the control module, which is a key technical component. It calculates the difference between the current force applied by the gripper 414 and the preset target force through an algorithm based on the data provided by the force sensor, and adjusts the clamping force of the gripper 414 accordingly. This force feedback control mechanism can ensure that the force applied by the gripper 414 is always within a safe range during the process of grasping and releasing the planting basket 5, avoiding damage to the seedlings. The working principle of the force feedback model is based on the PID control algorithm in control theory, and through the adjustment of three parameters: proportional (P), integral (I), and differential (D), the fine control of the force of the gripper 414 is achieved.

[0055] Specifically, the proportional gain is used to adjust the immediate response of the error, the integral gain is used to adjust the cumulative effect of the error, and the differential gain is used to adjust the rate of change of the error. The control module sets a target force value, and the difference between the actual force value measured by the force sensor and the target force value is the error. The control module calculates a control signal based on the error and adjusts the drive signal of the gripper 414 to maintain the target force. If the measured force exceeds the set safety threshold, the control module will generate a feedback signal to reduce the actual force applied by the gripper 414 to protect the plant from damage.

[0056] Embodiment 3:

[0057] This embodiment only describes the parts different from Embodiment 1. Specifically: the control module generates a movement signal through the vision module, and the movement signal satisfies:

[0058] ;

[0059] Among them, is the immediate position coordinate of the planting basket;

[0060] is the target position coordinate of the planting basket.

[0061] It should be noted that in this embodiment, due to the presence of the steering unit, when the height reaches the predetermined position, the steering action of the steering unit replaces the movement on the Y-axis. Therefore, there is no coordinate movement in the sense of the Y-axis. The essence of the movement signal formula is the Pythagorean theorem. That is, in the established three-axis coordinate system, in the plane formed by the X-axis and the Z-axis, when the immediate position coordinate and the target position are determined, the distance calculated by the movement signal is the straight-line distance in this plane. Since the X-axis and the Z-axis are necessarily perpendicular, therefore, the shortest distance of its movement on the X-axis and the Z-axis can also be directly calculated through the Pythagorean theorem.

[0062] Embodiment 4:

[0063] This embodiment only describes the parts different from Embodiment 1. Specifically:

[0064] The visual module is equipped with a target detection model. The target detection model extracts features through a feature extraction network to obtain a feature map; and inputs the feature map into a region proposal network to generate multiple candidate regions of the seedling stem targets; then inputs the feature map and the candidate regions into the RoIAlign layer together, so that each target candidate region is normalized to the same scale, and the pixels in the corresponding feature map are completely aligned with the pixels in the original image; finally, target features are extracted from the feature map, and then output to the fully connected layer and the fully convolutional network respectively for target classification and instance segmentation to generate categories, bounding boxes and binary images.

[0065] The visual module is also equipped with a state judgment model. The state judgment model is used to judge the growth state of the seedlings. Among them, the judgment process includes: using OPENCV to obtain the contour information of the seedling branch stems and the main stem; obtaining the centroid coordinates of the seedlings and determining the main stem; determining the center point of the branch stem bifurcation position; positioning the main stem state operation point; judging the range of the main stem state operation point.

[0066] The range quantity of the growth state information is input into the state judgment model. When the stem state operation point belongs to the range quantity, it is judged that the growth state of the seedling is good. When the stem state operation point does not belong to the range quantity, it is judged that the growth state of the seedling is abnormal. The feature extraction network is a ResNet50 structure and an FPN feature pyramid network.

[0067] It should be noted that the working principle of the object detection model is based on deep learning. First, through a feature extraction network, such as the ResNet50 structure and the FPN feature pyramid network, key features are extracted from the input image to form a feature map. This step can capture important information in the image, such as shape, texture, and color, etc., laying a foundation for subsequent object recognition. Further, the feature map is input into the Region Proposal Network (RPN), which is a network specifically designed to generate potential object regions (i.e., candidate regions). The RPN can identify regions in the image that may contain objects and generate multiple candidate regions for these regions. These candidate regions are then input together with the features Figure 1 into the RoIAlign layer. The function of this layer is to normalize each candidate region to the same scale and ensure that the pixels in the feature map are exactly aligned with the pixels in the original image. This process is crucial for improving the accuracy of object detection because it ensures that objects at different scales and positions can be accurately recognized.

[0068] After feature extraction and alignment, the model further extracts object features from the feature map and outputs them to the fully connected layer and the fully convolutional network respectively. The fully connected layer is responsible for object classification to determine the object category in the candidate region; while the fully convolutional network is responsible for instance segmentation to generate the precise bounding box and binary image of the object. In the binary image, the object region is marked white and the background is black, which provides precise object position information for subsequent transplanting operations.

[0069] The state judgment model is used to evaluate the growth state of the seedlings. Its working process includes using the OPENCV library to obtain the contour information of the seedling's lateral stem and main stem, determining the centroid coordinates and the position of the main stem of the seedling, identifying the center point of the lateral stem bifurcation position, locating the working point of the main stem state, and judging whether the working point of the main stem state is within the preset growth state information range. This model judges whether the growth state of the seedling is good or abnormal through the input growth state information range, thereby providing a decision basis for the automated transplanting robot on whether to perform transplanting operations.

[0070] It should also be noted that during the training process, the optimal number of layers required for the neural network structure is unknown, and its depth depends on the complexity of the dataset. Therefore, by adding skip connections to the network structure, it is allowed to skip the training of useless layer parameters when training the network structure and ignore the parameters of the skipped layers of the network structure, effectively reducing the number of parameters and improving the effectiveness of the model. So the ResNet network structure has dynamics during training and optimally adjusts the number of layers during the training process. In the detection of seedlings, its categories are divided into two types: with bifurcation and without bifurcation, and the detection categories are simple. Therefore, in this embodiment, ResNet50 is preferentially selected as the backbone neural network for model training.

[0071] In addition, due to the large number of seedlings on the seedling-raising board, there is also the problem of multi-scale object detection. FPN adopts a top-down hierarchical structure with lateral connections, from single-scale input to constructing a network feature pyramid, fusing different feature maps, so that each layer of the fused network has both deep and shallow features.

[0072] For the preprocessing of images, preprocessing is carried out in the early stage of training. The pictures are uniformly processed into a size with a pixel size not exceeding 800×800. The data augmentation method of adding Gaussian white noise, random cropping and hue transformation is used to expand the dataset, and the picture data is expanded to at least 3000 pictures.

[0073] For the object detection process: by reading the binary image respectively, the contour extraction function in OPENCV is called to retrieve the contours in the binary image, and the contour information of the branch stem and the main stem is obtained respectively.

[0074] After obtaining the contour information of the branch stem and the main stem, the main stem and the branch stem are determined by obtaining the centroid coordinates of each contour, so as to carry out the processing operation. First, determine the coordinate of point a of the centroid of the branch stem, and obtain the coordinates of points b and c of the mass points of the two nearest main stems adjacent to it. By comparing the vertical coordinates of points b and c, the main stem above the branch stem among the adjacent main stems is determined. Finally, the branch stem a and the main stem b to be processed are determined.

[0075] The method of contour fitting a straight line is adopted to fit a straight line to the branch stem and the main stem respectively and obtain the straight line equation, and the intersection point of the two straight lines is obtained. This point is the central position of the fork of the seedling stem.

[0076] In the two-dimensional image, assume that the width of the recognized main stem segmentation area is a constant. According to the contour information of the main stem and the branch stem, the centroid point A of the main stem and the centroid point B of the branch stem can be obtained, and the included angles between the fitted main stem center line and the branch stem center line and the horizontal axis respectively. From this, the straight line equation of the fitted center line and the coordinates of the intersection point (branch stem center point) with the center line can be determined. Through the relationship of the branch stem center point, the whole branch operation point can be obtained. The introduction of the state judgment model enables the automatic transplanting robot to evaluate the growth state of the seedlings, which has guiding significance for deciding whether to carry out the transplanting operation. By obtaining the contour information of the seedling branch stem and the main stem through OPENCV and combining the range of the growth state information, the robot can judge whether the growth state of the seedlings is good, so as to avoid transplanting the seedlings with poor growth state and ensure that the transplanted seedlings have good growth potential.

[0077] The specific embodiments described above further elaborate on the object, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A multi-layered three-dimensional cultivation automatic transplanting robot for a plant factory, comprising a cultivation rack, and a seedling plate is arranged on the side of the cultivation rack, and is characterized in that: The robot includes a control module, a vision module, a positioning module, a moving module, and a clamping module. The moving module includes a moving base, a vertical frame, and a horizontal frame. The horizontal frame is slidably disposed on the vertical frame and is in signal connection with the control module. The moving base is slidably disposed on the horizontal frame and is in signal connection with the control module. The vision module is in signal connection with the control module and is used for collecting position information. The positioning module is in signal connection with the control module and is used for determining position information. The clamping module is disposed on the moving base and is used for clamping and fixing the planting basket in the seedling plate through the control module; Among them, the process of determining the position information is as follows: The position of the seedling is recognized by the vision module to obtain 3D point cloud data through scanning. Then, the 3D point cloud data is processed to extract the position information and shape information of the planting basket and the seedling, and a coordinate system is established. Finally, the boundary position information of the planting basket is transmitted to the control module; The control module generates a moving signal through the vision module, and the moving signal satisfies: ； Among them, is the immediate position coordinate of the planting basket; is the target position coordinate of the planting basket; A target detection model is carried in the vision module. The target detection model extracts features through a feature extraction network to obtain a feature map; and the feature map is input into a region proposal network to generate multiple candidate regions of the seedling stem target; then the feature map and the candidate regions are jointly input into the RoIAlign layer, so that each target candidate region is normalized to the same scale, and the pixels in the corresponding feature map are completely aligned with the pixels in the original image; finally, target features are extracted from the feature map, and then are respectively output to a fully connected layer and a fully convolutional network for target classification and instance segmentation to generate categories, bounding boxes, and binary images; A state judgment model is also carried in the vision module. The state judgment model is used to judge the growth state of the seedling. Among them, the judgment process includes: obtaining the contour information of the seedling branch stem and the main stem by using OPENCV; obtaining the centroid coordinates of the seedling and determining the main stem; determining the center point of the branch stem bifurcation position; positioning the working point of the main stem state; and judging the range of the working point of the main stem state.

2. The multi-layer three-dimensional cultivation automatic transplanting robot for a plant factory according to claim 1, characterized in that: The clamping module includes a clamping unit and a clamping seat. The clamping seat is movably connected to the moving base. A plurality of lifting units are arranged below the clamping seat. The clamping unit includes a clamping block detachably connected to the lifting unit. A clamping motor is arranged in the clamping block. A plurality of clamping grooves are formed in the lower end surface of the clamping block. Claw is movably arranged in the clamping groove. A crank is also arranged in the clamping block. Two ends of the crank are respectively connected to the output end of the clamping motor and the claw. When the clamping motor is started, the opening and closing of the claw are realized through the synchronous rotation of the crank.

3. The multi-layer three-dimensional cultivation automatic transplanting robot for a plant factory according to claim 2, characterized in that: A force sensor is arranged on the inner side of the claw. The force sensor is in signal connection with the control module. A force feedback model is carried in the control module. The force feedback model is used for monitoring the acting force during the grasping process of the claw.

4. The multi-layer three-dimensional cultivation automatic transplanting robot for a plant factory according to claim 3, characterized in that: The force feedback model satisfies: ； ； Among them, is a control signal; is the proportional gain, which is used to adjust the immediate response to the error; is the integral gain, which is used to adjust the cumulative effect of the error; is the differential gain and is used to adjust the rate of change of the error; is the target force, that is, the force applied by the gripper set by the control module; To measure the force, i.e., the force actually applied by the gripper, it is measured by a force sensor; is the error, i.e., the difference between the target force and the measured force; is the integral of the error, representing the accumulation of the error from the start to the current moment; is the rate of change of the error, representing the speed at which the error changes over time.

5. The multi-layer three-dimensional cultivation automatic transplanting robot for a plant factory according to claim 4, characterized in that: A safety threshold is set in the force sensor. When the measured force exceeds the safety threshold, the control module generates a feedback signal to reduce the actual acting force of the claw.

6. The multi-layer three-dimensional cultivation automatic transplanting robot for a plant factory according to claim 1, characterized in that: The range of the amount of growth state information is input into the state judgment model. When the operating point of the stem state belongs to the range, it is judged that the growth state of the seedling is good. When the operating point of the stem state does not belong to the range, it is judged that the growth state of the seedling is abnormal.

7. The multi-layered three-dimensional cultivation automatic transplanting robot for a plant factory according to claim 1, characterized in that: The feature extraction network is a ResNet50 structure and a FPN feature pyramid network.

Images