Fresh flower harvesting trolley based on stalk adaptive cutting and vibration compensation

Abstract

The invention provides a fresh flower harvesting trolley based on stalk adaptive cutting and vibration compensation, and relates to the field of agricultural automatic picking, and the fresh flower harvesting trolley comprises a rack (10), a moving mechanism (20), a harvesting self-balancing mechanism, a platform (40), a guide rail sliding table mechanism, a retractable picking mechanical arm mechanism, a tail end cutting mechanism, a storage mechanism and a binocular camera, the bottom ends of the four corners of the rack (10) are each provided with a set of moving mechanism (20). The four harvesting self-balancing mechanisms are sequentially arranged at the four corners of the top end of the rack (10), and the top ends of the harvesting self-balancing mechanisms are connected with the bottom face of the platform (40). The guide rail sliding table mechanism is arranged on the end face of the platform (40) and located in the middle of the platform (40) and provided with the retractable picking mechanical arm mechanism, and the tail end cutting mechanism is arranged at the bottom end of the retractable picking mechanical arm mechanism. According to the harvesting trolley, high-precision positioning and high-success-rate cutting of various fresh flowers can be achieved, and the picking efficiency and the quality of the picked fresh flowers are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of agricultural automation picking, in particular to a fresh flower harvesting trolley based on stem adaptation cutting and vibration compensation. BACKGROUND

[0002] Fresh flower harvesting is a key link from field picking to final presentation as a commodity or ornamental. Current fresh flower harvesting mainly relies on manual picking and mechanical harvesting. Manual picking can ensure the integrity of fresh flowers and avoid damage to fresh flowers during picking, but it is low in efficiency, high in labor cost, and the maturity of picked flowers mainly depends on manual judgment, which is high in human factors and prone to picking unripe flowers, resulting in uneven quality of picked flowers. Mechanical harvesting can effectively reduce the cost of manual picking and improve harvesting efficiency to some extent, but the characteristics of fresh flower stems vary significantly (for example, rose stems are thick and thorny, lily stems are thick and hollow, and petunia stems are thin and brittle). The existing electromagnetic cutting mechanism uses "fixed magnetic force" to drive the blade, which can only adapt to a single stem thickness. For thick stems, the cutting is incomplete due to insufficient magnetic force, and for thin stems, the stem is broken due to excessive magnetic force. In addition, thorny stems are easy to get stuck in the blade sliding groove, resulting in cutting mechanism jamming. At the same time, there are height differences in the ridges and furrows of the fresh flower planting area, and the chassis of the mechanized harvesting trolley will produce non-periodic vibration when moving. This vibration will be transmitted to the harvesting platform through the rack, even if the existing self-balancing mechanism can maintain the level of the platform, but the camera will still be shaken, resulting in focusing deviation of the camera, positioning error of the harvesting robot arm, and recognition error of the fresh flower picking point, which will cause the harvesting arm to fail to cover the flower stem, the flower stem to deviate from the barrel opening when put into the flower storage barrel, and other problems, such as damage to the flower stem during picking, and even failure to achieve effective picking. SUMMARY

[0003] In view of the above problems existing in the prior art, the purpose of the present application is to provide a fresh flower harvesting trolley based on stem adaptation cutting and vibration compensation. The harvesting trolley realizes high-precision positioning and high-success-rate cutting of various fresh flowers through dynamic matching of stem characteristics and cutting force and vibration feedforward compensation self-balancing, thereby improving picking efficiency and postharvest fresh flower quality.

[0004] The purpose of the present application is achieved by the following technical solutions: A flower harvesting trolley based on stem-adaptive cutting and vibration compensation includes a frame, a moving mechanism, a harvesting self-balancing mechanism, a platform, a guide rail slide mechanism, a retractable harvesting robotic arm mechanism, an end-cutting mechanism, a storage mechanism, and a binocular camera. The frame has a portal frame structure with a set of moving mechanisms installed at each of its four corners to control the overall movement of the harvesting trolley. There are four sets of harvesting self-balancing mechanisms, which are sequentially located at the four corners of the top of the frame. The top of the harvesting self-balancing mechanism (i.e., the end furthest from the frame) is connected to the bottom of the platform, and the platform has a rectangular frame structure in its transverse cross-section. The guide rail slide mechanism is located on the end face of the platform, and the retractable harvesting robotic arm mechanism is located in the middle of the platform frame. The end-cutting mechanism is located at the bottom of the retractable robotic arm mechanism. There are two sets of storage mechanisms, which are symmetrically located on the inner walls of both sides of the frame. The binocular camera is located on the bottom of the platform, corresponding to the retractable harvesting robotic arm mechanism, and its viewing direction is with respect to the direction of travel of the harvesting trolley.

[0005] Based on further optimization of the above scheme, the moving mechanism includes a rubber wheel, a servo motor, a servo motor connector, a motor fixing component, and a brushless motor. The top of the servo motor is connected to the bottom of the frame, and the bottom of the servo motor is equipped with a motor fixing component through the servo motor connector. The brushless motor is installed inside the motor fixing component, and the rubber wheel is fixedly sleeved on the outer wall of the output shaft of the brushless motor. The harvesting self-balancing mechanism includes an electric telescopic rod and a single-rod universal ball joint bearing. One end of the electric telescopic rod is connected to the top of the frame, and the other end is connected to the bottom surface of the platform through the single-rod universal ball joint bearing.

[0006] Based on further optimization of the above scheme, the guide rail slide mechanism includes a first lead screw, a first stepper motor, a sliding crossbar, a second lead screw, a second stepper motor, and a slider. There are two first lead screws, which are respectively set on the end face of the platform corresponding to the long rods of the platform. The first lead screws are driven by the first stepper motor. The two ends of the sliding crossbar are slidably connected to the end face of the long rods of the platform. The two ends of the sliding crossbar are threaded through and connected by the corresponding first lead screws. The end face of the sliding crossbar is provided with a second lead screw, which is driven by the second stepper motor. The slider is slidably set on the end face of the sliding crossbar, and the middle of the slider is threaded through and connected by the second lead screw.

[0007] Based on further optimization of the above scheme, the retractable harvesting robotic arm mechanism includes a connector, a telescopic mechanism, a flexible hose, a barrier cover, a cutting support, and an inertial sensor. The bottom surface of the connector is fixedly connected to the end face of the slider, and telescopic mechanisms are respectively set at both ends of the connector. The two telescopic mechanisms are located at the two ends of a diagonal line of the connector, and a ring-shaped cutting support is set at the bottom of the two telescopic mechanisms. A barrier cover is set on the top surface of the cutting support through multiple evenly distributed connecting rods, and a flexible hose is set at the top of the barrier cover. An end-effector camera is set inside the barrier cover, and an inertial sensor is set on the end face of the connector; the bottom of the barrier cover... The end is designed with a sloping structure. The end cutting mechanism is located on the upper side of the cutting support corresponding to the bottom of the barrier cover. It includes a blade holder, a cutting opening, a return spring disc, a barrier cloth, a blade, an electromagnet, and a strong magnet. The outer side wall of the blade holder is penetrated and fixedly connected by multiple connecting rods. A cutting opening is opened in the middle of the blade holder, and electromagnets are respectively installed on the two side walls of the cutting opening. The blade is slidably installed in the cutting opening, and a strong magnet is fixedly installed on the end face of the blade. A return spring disc is installed at the top of the sloping surface of the blade holder. One end of the barrier cloth is fixedly connected to the outer wall of the return spring disc and wrapped around the outer wall of the return spring disc, and the other end is fixedly connected to the end face of the strong magnet.

[0008] Based on further optimization of the above scheme, the storage mechanism includes an outer shell and flower storage tubes, the outer shell is fixedly installed on the inner wall of the frame and the flower storage tubes are distributed in an array inside the outer shell.

[0009] Based on further optimization of the above scheme, the specific steps for the harvesting cart to harvest flowers are as follows: Step S1: First, activate the self-balancing harvesting mechanism to keep the platform level and avoid the effects of displacement and vibration caused by ground elevation differences; then, use the moving mechanism to control the harvesting trolley to move above the flower ridge. Step S2: Use a binocular camera to transmit images of flowers below the platform, match the flower image position with the actual position to obtain the position that the retractable picking robotic arm needs to reach (i.e., the harvesting position), and control the retractable picking robotic arm to move to the harvesting position through the guide rail slide mechanism. Step S3: The flower is inserted through the telescopic mechanism, and the flower image is identified by the end camera to determine the cutting position of the flower stem. The flower stem is then cut by the end cutting mechanism. Step S4: Place the cut flowers into the corresponding flower storage tube to collect the flowers; then, the retractable picking robotic arm mechanism resets and waits for the next harvesting operation.

[0010] Based on further optimization of the above scheme, step S1 specifically includes: Step S11, Platform Self-Balancing Control: The sampling frequency of the gyroscope of the inertial sensor is... f , per 1 / f Output angular velocity data wHigh-frequency noise is removed through adaptive window filtering; window size... N Dynamically adjusted by the rate of change of the signal: ; In the formula: w t Indicates the current angular velocity. w t-1 This represents the angular velocity at the previous moment; w max Indicates the maximum value of angle SU; k This represents the adjustment coefficient (obtained from a large amount of experimental data, typically 10). w fil Indicates the angular velocity after filtering; Then, a Kalman filter model is constructed to fuse attitude data from the gyroscope and accelerometer (within the inertial sensor). Through Kalman filter prediction and update steps, the optimal attitude estimate is obtained. : Equations of state: ; Observation equation: ; In the formula: X k This represents the matrix composed of attitude angles and attitude angular velocities, i.e. A represents the state transition matrix, i.e. B represents the control input matrix, i.e. ; U k This indicates a control input (e.g., the output of an accelerometer). W k Indicates process noise; Z k This represents the attitude observation matrix of the gyroscope and accelerometer, i.e. ; H Represents the observation matrix, i.e. ; V k Indicates observation noise; Because gyroscope sensors have good instantaneous dynamic response but can produce angular drift, while accelerometers have good static accuracy but can fluctuate during motion or when subjected to disturbances; therefore, the simply filtered angular velocity and acceleration data are weighted and fused. ; In the formula: C gyro , C acc These represent the confidence levels of the gyroscope and accelerometer, respectively, calculated based on the gyroscope drift and the accelerometer noise level. These represent the fusion coefficients of the gyroscope and accelerometer, respectively; ; In the formula: This indicates the gyroscope angle at the previous moment. Indicates the gyroscope is in The change in angle during the integral over time. Indicates the angle of the accelerometer; According to the fusion and Construct an attitude gradient differential formula that considers both the current attitude deviation and the attitude change trend to achieve more agile balance adjustment: ; In the formula: Indicates the first i The elongation of the root telescopic mechanism; k 1. k 2 represents the attitude angle adjustment coefficient and the attitude angle change rate adjustment coefficient, respectively (obtained from a large amount of experimental data); Step S12: Controlling the forward direction of the harvester: The control system transmits the movement signal to the harvester through the SBUS communication protocol, and then controls the servo motor of the movement mechanism to rotate (the servo motor uses a brushless motor) through the FOC algorithm (field orientation control algorithm) to determine the forward direction of the harvester; Step S13, Harvesting Trolley Movement Control: The control system inputs a speed signal to control the brushless motor of the moving mechanism to rotate, thereby driving the platform of the harvesting trolley to move to the designated position.

[0011] Based on further optimization of the above scheme, step S2 specifically includes: Step S21, Flower Position Matching: This includes flower image segmentation and coordinate transformation. Flower image segmentation: The image output from the stereo camera is first subjected to Gaussian filtering to remove noise, then the OTSU algorithm is used for image segmentation, and color, texture, and edge features are weighted and fused to obtain the segmentation threshold. T : ; In the formula: These represent the statistics for overall color (i.e., HSV space), overall texture (i.e., LBP features), and overall edge (i.e., Canny detection), respectively. , These represent the average color characteristics of the background and the flower area, respectively. , These represent the average texture features of the background and the flower area, respectively. , These represent the average edge features of the background and the flower area, respectively.v 0 indicates the proportion of background pixels to total pixels. v 1 (x) Indicates in x The percentage of pixels in the flower area to the total number of pixels ( v 0+ v 1 (x) =1); Pixels with grayscale values ​​greater than T are identified as flower areas; Coordinate transformation: Transform the camera pixel coordinates ( x p , y p ) represents the physical coordinates of the retractable harvesting robotic arm mechanism. x cu , y cu ): ; In the formula: j 1. j 2 represents the radial distortion coefficient. p 1. p 2 represents the tangential distortion coefficient (optimized by acquiring calibration board images multiple times); Convert the baseline conversion coefficients to spatial variation coefficients: ; In the formula: They represent x direction, y Spatial variation coefficient of direction; They represent x direction, y Spatial variation increment of direction (based on the three-dimensional error model of the camera-robotic arm calibration field, by analyzing the deviation between the theoretical coordinates and the actual coordinates of feature points in the calibration field, the error is fitted in ( x , y The distribution pattern at point () was obtained. f x0 , f y0 Indicates the base conversion factor; ; In the formula: x b0 , y b0 These represent the pixel coordinate reference offset (obtained through camera intrinsic parameter calibration); x arm (t) , y arm (t)Let represent the physical coordinates of the robotic arm at time t; This represents the time difference between image acquisition and coordinate transformation; v x , v y These represent the robotic arm in x , y velocity in direction, a x , a y These represent the robotic arm in x , y Acceleration in the direction of; Step S22, Stepper Motor Control: By controlling the stepper motors of the first and second lead screws, the retractable picking robotic arm mechanism is moved above the flowers to facilitate subsequent harvesting.

[0012] Based on further optimization of the above scheme, step S3 specifically includes: Step S31: Inserting and positioning the flower stem for cutting: First, the end-point camera uses "template matching" to locate the flower head: multi-pose basic templates are pre-stored, and during actual detection, high-matching samples are automatically selected and dynamically added to the template library, solving the problem of poor adaptability of a single template to variations in flower head morphology (such as petal opening and closing, minor damage). Then, based on the ambient light and flower head variety, the matching weights of different templates are automatically adjusted (e.g., prioritizing anti-high-light templates in strong light environments), improving robustness in complex scenarios. ; In the formula: Indicates the first i The weight coefficient of each template ( ); NCC i Indicates the first i The average normalized cross-correlation coefficient between each template and historical successfully matched samples; C i Indicates the color feature matching degree; T i Indicates the time decay factor; These represent the corresponding weight coefficients ( ); NCC i By extracting template images F Sample images that successfully matched history S j Obtaining the grayscale matrix: ; In the formula: , These represent the mean gray values ​​of the template and the sample, respectively. C i By converting the images of the template and the region to be matched to HSV space, color histograms are extracted using the H, S, and V channels respectively. The Bach distance between the template and the sample histograms is then calculated. B(F,S) Get: ; In the formula: These represent the template and sample at the 1st and 2nd respectively. k The probability distribution of each interval. m The total number of intervals; T i Used to simulate the decay of flower head freshness over time: ; In the formula: This indicates the duration the template has been stored in the template library; This represents the attenuation coefficient (obtained by fitting actual varieties and a large amount of experimental data). Then, the Canny edge detection algorithm is used to extract the outline edges of the petals and screen out potential candidate regions of the flower head. Within the candidate regions, the "normalized cross-correlation coefficient (NCC)" is used to achieve accurate morphological matching of the flower head, and the geometric constraints between the flower head and the stem are used to eliminate mismatched regions. Then, the conversion relationship between pixels and physical length is dynamically corrected by the depth information of binocular vision to improve the accuracy of long-distance positioning and achieve accurate positioning of the flower head, providing accurate spatial and dimensional basis for determining the subsequent flower stem cutting position. Subsequently, a combination of primary soft limiting and secondary hard limiting is used to achieve adaptive cutting. Level 1 soft limit is a variety-based early warning system: establishing dynamic early warning thresholds. L yj When the actual length reaches the warning threshold L yj When the threshold reaches 90%, an early warning state is activated. ; In the formula: L jz This represents the standard length for the corresponding flower variety (obtained through fitting a large amount of experimental data). M cs The flower stem maturity coefficient represents the image recognition result. D zj The stem diameter is indirectly measured by pressure sensors (the stem diameter is indirectly calculated by the timing difference and pressure distribution of two sets of pressure sensors arranged on the inner walls of the two telescopic mechanisms and the inner walls on both sides of the lower end of the end cutting mechanism). z 1. z2 represents the correction factor; The method for obtaining the flower stem maturity coefficient is as follows: First, obtain the color similarity with mature flower stems by using the average RGB color value of the flower stem region. Sim c : ; In the formula: ( R djc ,G dic ,B djc () represents the average RGB color value of the area of ​​the flower stem to be detected; R cs ,G cs ,B cs () represents the average RGB color value of a mature flower stem; Then, LBP (Local Binary Pattern) was used to extract the surface texture features of the flower stem, and the matching degree between the features and the standard texture features was calculated. Sim t : Based on LBP texture features, the LBP histogram vector of the flower stem to be detected is defined as follows: H djc =[ h t1 , h t2 , ..., h t256 (256 dimensions, corresponding to the frequency of LBP values ​​from 0 to 255), the LBP histogram vector of a standard mature flower stem is H cs =[ h c1 , h c2 , ..., h c256 ]; The grayscale value of any pixel (x, y) in the grayscale image of the flower stem is g c The pixel grayscale values ​​of the 8 domains are as follows: g 0~ g 7 (arranged clockwise): ; ; The frequency of the k-th dimension (corresponding to the LBP value k) of the LBP histogram of the flower stem to be detected is: ; ; ; In the formula: These represent the corresponding weight coefficients ( (determined through training with a large number of samples). M cs ∈[0,1], where 0 represents completely immature and 1 represents completely mature; The stem diameter indirectly tested by the pressure sensor is as follows: First, when the flower stem passes through the end-cutting mechanism and the hose, the pressure sensor on the inner wall of the telescopic mechanism and the pressure sensors on both sides of the cut opening of the end-cutting structure will generate pressure trigger signals due to the compression of the flower stem (the pressure sensor on the inner wall of the telescopic mechanism generates a signal through contact with the hose and when the flower stem passes through the hose, squeezing the inner wall pressure sensor). The time difference between the triggering of the signals from the two sensors is recorded. This is combined with the conveying speed of the flower stem (calculated from the transmission speed of the telescopic mechanism). v hjz To obtain the preliminary diameter of the flower stem D start : ; In the formula: d cg0 Indicates the initial spacing of the sensors (obtained through calibration); Then, based on the pressure value distribution of the pressure sensors on both sides, that is... P left , P right The initial diameter is corrected: ; Finally, the weighted average of the detection results from the two sets of pressure sensors was calculated to obtain the stem diameter. D zj : ; In the formula: The two sets of pressure sensors, located on the inner walls of the two telescopic mechanisms and on the inner walls of the lower ends of the end cutting mechanism, respectively represent the stem diameters obtained after correction. These represent the corresponding weighting coefficients (generally) ); The duty cycle signal is dynamically adjusted based on real-time detection of contact pressure by a pressure sensor (it automatically increases when the pressure change rate exceeds a set threshold). A value that accelerates the deceleration process; conversely, a value that decreases the deceleration process. (to achieve smooth adjustment) ; In the formula: Duty(n) Indicates the first n Duty cycle during the second adjustment; Pdq This indicates the pressure value detected in real time. P yz This indicates the standard contact pressure for the corresponding variety (obtained through multiple tests). This represents the adjustment coefficient, which is dynamically adjusted according to the pressure change rate (generally 0.05 to 0.2). The condition of the stem is determined by integrating multiple parameters: Contact determination: When the real-time detected contact pressure is not less than P yz And pressure change rate When the force is not less than 0.5 N / s, it is determined that the flower stem has successfully made contact with the end cutting mechanism; ; Missing criterion: When the real-time detected contact pressure is less than P yz Furthermore, if the displacement increment of the end effector is less than 0.5 mm / 100 ms, it is determined to be jammed; Short stem warning: When the duty cycle drops to 10% and no effective pressure value is detected, first drive the telescopic mechanism to rise 2mm, then try extending and lowering 3mm to confirm whether the contact pressure is still less than [previous value]. P yz And check if the displacement increment of the end effector is still less than 0.5mm / 100ms; if so, issue an alarm. The secondary hard limit switch is a forced protection based on physical switches. The telescopic mechanism has two built-in mechanical limit switches. When the motor operating current exceeds the rated current, electronic soft protection is triggered (speed is reduced first and then stopped). At the same time, abnormal fluctuations in the acceleration of the telescopic mechanism are detected to predict mechanical failure. When the distance to the limit switch is 5mm, the speed is automatically reduced to 30% of the rated speed. After each harvest, the benchmark is dynamically corrected based on the deviation between the actual cutting position and the standard position. ; In the formula: JZ new , JZ old These represent the new reference and the original reference (reference: the mechanical zero-point reference of the telescopic mechanism, i.e., the theoretical position coordinates when the telescopic mechanism is fully retracted). WZ sj , WZ bz These represent the actual cutting position and the standard position, respectively. Represents the learning coefficient; A comprehensive calibration is performed after every 10 harvests to correct the temperature of the telescopic mechanism. ; In the formula: This indicates the temperature change after 10 harvests; Step S32, Blade Cutting: By changing the direction of the current in the electromagnets at both ends of the cutting opening, the magnetic poles of the electromagnets are made to be the same as the magnetic poles of the strong magnetic block. Using the principle of like poles repelling each other, the blade is driven by magnetic force to cut the flower branches. The blade also drives the barrier cloth to seal the cutting opening and prevent the flowers from falling.

[0013] Based on further optimization of the above scheme, step S4 specifically includes: Flower storage cylinder coordinate calibration: During initialization, manually move the retractable harvesting robotic arm mechanism above each flower storage cylinder and record the corresponding coordinates. X , Y Shaft storage tube number: N x1 , N y1 , N x2 , N y2 … N xn , N yn (Also indicates the number of flower storage tubes); During operation, the coordinates of the flower storage tubes are called in the order of "left to right, top to bottom" (to avoid duplicate storage); Infrared reflective points are set at the edge of each flower storage tube, and a miniature camera and infrared ranging sensor are set at the bottom of the retractable picking robotic arm mechanism. Based on the initial manual calibration, the coordinate drift problem caused by mechanical error and vibration is solved through a dynamic correction mechanism of "vision + sensor fusion". During the operation, after the retractable harvesting robotic arm moves to the vicinity of the calibration coordinates, the miniature camera identifies the markers and calculates the deviation between the actual coordinates and the calibration values. Combined with the height deviation detected by the infrared range sensor The motion trajectory is corrected in real time; then, a "coordinate deviation database" is established to record the deviation patterns under different working durations and temperatures, and the least squares method is used to predict the deviation trend and compensate and correct it in advance.

[0014] The following are the technical effects of the present invention: This invention employs an "adaptive window filtering + Kalman filtering + weighted fusion" approach, combined with attitude data from gyroscopes and accelerometers, to filter high-frequency vibration noise and correct sensor drift errors. Based on the fused attitude data, four sets of electric telescopic rods adjust the platform's height and level in real time, effectively offsetting non-periodic vibrations caused by chassis movement and eliminating the shaking effect of vibrations on the binocular camera and end-effector camera, ensuring positioning and recognition accuracy. The end-effector camera uses a template library mechanism of "pre-stored basic templates + dynamically supplemented high-matching samples," automatically adjusting template matching weights based on ambient light and flower variety to solve the adaptability problem of single templates to petal opening and closing and minor damage. Simultaneously, the OTSU algorithm is used to segment flower images, combined with Canny edge detection to extract flower head contours, and binocular visual depth information is used to correct the conversion relationship between pixels and physical length, providing spatial basis for precise cutting. Furthermore, the speed, acceleration, and time difference correction of the robotic arm are introduced during coordinate transformation to avoid positioning deviations caused by the "time lag" in image acquisition of the robotic arm's movement, further improving the accuracy of the robotic arm moving to the harvesting position.

[0015] During the cutting process, the end-cutting mechanism of this invention adjusts the repulsive force between the electromagnet and the strong magnetic block by changing the direction of the electromagnet current. This flexibly adapts to the stem characteristics of different flower varieties such as roses (thick and thorny), lilies (thick and hollow), and carnations (thin and brittle). This avoids cutting residue caused by insufficient magnetic force on thick stems and prevents stem breakage caused by excessive magnetic force on thin stems, thus improving the cutting success rate. At the same time, the barrier cloth at the cutting edge moves synchronously with the blade, sealing the cutting edge during cutting. This prevents thorny stems from getting stuck in the blade groove and prevents cut flowers from falling out of the cutting edge, reducing flower loss during harvesting. Furthermore, by combining a first-level soft limit (dynamic warning threshold based on variety) and a second-level hard limit (mechanical switch + electronic soft protection), the cutting parameters can be dynamically adjusted according to the maturity and diameter of the flower stem. At the same time, by monitoring the current and predicting acceleration anomalies, damage to the cutting mechanism caused by mechanical overload is avoided, further ensuring the integrity of the flower stem. During storage, the use of partitioned flower storage tubes prevents leaf compression damage. In addition, the addition of preservative solutions to the flower storage tubes extends the shelf life of cut flowers, improves the leaf integrity rate, and further enhances the efficiency of flower harvesting. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall structure of the flower harvesting cart in an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of the moving mechanism of the flower harvesting cart in an embodiment of the present invention.

[0018] Figure 3This is a schematic diagram of the self-balancing harvesting mechanism of the flower harvesting cart in an embodiment of the present invention.

[0019] Figure 4 This is a schematic diagram of the guide rail slide mechanism of the flower harvesting cart in an embodiment of the present invention.

[0020] Figure 5 This is a schematic diagram of the retractable picking robotic arm mechanism of the flower harvesting cart in an embodiment of the present invention.

[0021] Figure 6 This is a schematic diagram of the end-cutting mechanism of the flower harvesting cart in an embodiment of the present invention.

[0022] Figure 7 This is a schematic diagram of the end-cutting mechanism of the flower harvesting cart from another perspective in an embodiment of the present invention.

[0023] Among them, 10. Frame; 20. Moving mechanism; 21. Rubber wheel; 22. Servo motor; 23. Servo motor connector; 24. Motor fixing part; 25. Brushless motor; 31. Electric telescopic rod; 32. Single rod universal ball joint bearing; 40. Platform; 51. First lead screw; 52. First stepper motor; 53. Sliding crossbar; 54. Second lead screw; 55. Second stepper motor; 56. Slider; 61. Connector; 62. Telescopic mechanism; 63. Hose; 64. Barrier cover; 65. Cutting support; 66. Inertial sensor; 71. Tool holder; 72. Cutting kerf; 73. Return spring disc; 74. Barrier cloth; 75. Blade; 76. Electromagnet; 77. Strong magnet; 81. Housing; 82. Flower storage tube; 90. Binocular camera. Detailed Implementation

[0024] The technical solutions in the embodiments of the present invention will be clearly and completely described below. In the following description, specific details such as specific system structures and technologies are presented for illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention.

[0025] Example 1: A flower harvesting cart based on stem-adaptive cutting and vibration compensation, such as Figure 1 As shown: Includes a frame 10, a moving mechanism 20, a harvesting self-balancing mechanism, a platform 40, a guide rail slide mechanism, a retractable harvesting robotic arm mechanism, an end-effector cutting mechanism, a storage mechanism, and a binocular camera 90. The frame 10 has an overall portal frame structure (e.g., Figure 1As shown, the frame 10 includes four supporting vertical rods that are parallel to each other, with the bottom ends of two supporting vertical rods on the same side connected by a connecting crossbar. A set of moving mechanisms 20 is installed at each of its four corners (i.e., the moving mechanisms 20 are set at the bottom ends of the four supporting vertical rods) to control the overall movement of the harvesting trolley. The moving mechanism 20 includes rubber wheels 21, a servo motor 22, a servo motor connector 23, a motor mounting bracket 24, and a brushless motor 25 (e.g., ...). Figure 2 As shown), the top of the servo motor 22 is connected to the bottom of the frame 10, and the bottom of the servo motor 22 is equipped with a motor fixing component 24 through the servo motor connector 23. A brushless motor 25 is installed inside the motor fixing component 24, and a rubber wheel 21 is fixedly sleeved on the outer wall of the output shaft of the brushless motor 25. There are four sets of harvesting self-balancing mechanisms, which are arranged sequentially at the four corners of the top of the frame 10 (corresponding to the top of the four supporting vertical rods). The top of the harvesting self-balancing mechanism (i.e., the end away from the frame) is connected to the bottom surface of the platform 40. The transverse cross-section of the platform 40 is a rectangular frame structure (as shown). Figure 1 (As shown); the harvesting self-balancing mechanism includes an electric telescopic rod 31 and a single-rod universal ball joint bearing 32 (as shown). Figure 3 As shown), one end of the electric telescopic rod 31 is connected to the top of the frame 10, and the other end is connected to the bottom surface of the platform 40 through the single-rod universal ball joint bearing 32.

[0026] A guide rail slide mechanism is located on the end face of platform 40, and a retractable harvesting robotic arm mechanism is located in the middle of the frame of platform 40. An end-cutting mechanism is located at the bottom of the retractable harvesting robotic arm mechanism. For example... Figure 4 As shown, the guide rail slide mechanism includes a first lead screw 51, a first stepper motor 52, a sliding crossbar 53, a second lead screw 54, a second stepper motor 55, and a slider 56. There are two first lead screws 51, each corresponding to a long rod on the platform 40 and mounted on the end face of the platform 40. The first lead screws 51 are driven by the first stepper motor 52. The two ends of the sliding crossbar 53 are slidably connected to the end faces of the long rods on the platform 40. The two ends of the sliding crossbar 53 are threaded through and connected by the corresponding first lead screws 51. A second lead screw 54 is mounted on the end face of the sliding crossbar 53 and is driven by the second stepper motor 55. A slider 56 is slidably mounted on the end face of the sliding crossbar 53, with the second lead screw 54 threaded through and connected to the middle of the slider 56. Figure 5 As shown: The retractable harvesting robotic arm mechanism includes a connector 61, a telescopic mechanism 62, a flexible hose 63, a barrier cover 64, a cutting support 65, and an inertial sensor 66. The bottom surface of the connector 61 is fixedly connected to the end face of the slider 56, and telescopic mechanisms 62 are respectively set at both ends of the connector 61. The two telescopic mechanisms 62 are located at the two ends of a diagonal line of the connector 61, and the bottom of the two telescopic mechanisms 62 is provided with a ring-shaped cutting support 65. The top surface of the cutting support 65 is provided with a barrier cover 64 (e.g., ...) through multiple evenly distributed connecting rods. Figure 5As shown, in this embodiment there are four connecting rods, and a flexible hose 63 is provided at the top of the barrier cover 64. An end camera is provided inside the barrier cover 64, and an inertial sensor 66 is provided on the end face of the connector 61 (e.g., Figure 1 (As shown); the bottom of the barrier cover 64 is set with a sloping structure (the angle of the sloping structure is generally 45°), and the end cutting mechanism is set on the upper side of the cutting support 65 corresponding to the bottom of the barrier cover 64, including a blade holder 71, a cutting nozzle 72, a return spring disc 73, a barrier cloth 74, a blade 75, an electromagnet 76, and a strong magnet 77 (as shown). Figure 6 or Figure 7 As shown), the outer sidewall of the tool holder 71 is penetrated and fixedly connected by multiple connecting rods. A cutting opening 72 is opened in the middle of the tool holder 71 (the cutting opening 72 corresponds to the middle through hole of the cutting support 65 and the middle through hole of the barrier cover 64), and electromagnets 76 are respectively installed on the two sidewalls of the cutting opening 72 (e.g., Figure 7 As shown), the blade 75 is slidably disposed in the cutting opening 72 and a strong magnetic block 77 is fixedly disposed on the end face of the blade 75. A return spring disc 73 is disposed at the top of the inclined surface of the blade holder 71. One end of the barrier cloth 74 is fixedly connected to the outer wall of the return spring disc 73 and wrapped around the outer wall of the return spring disc 73, and the other end is fixedly connected to the end face of the strong magnetic block 77.

[0027] The storage mechanism consists of two sets, which are symmetrically arranged on the inner walls of both sides of the rack 10 (e.g., Figure 1 As shown), the storage mechanism includes a housing 81 and flower storage tubes 82. The housing 81 is fixedly installed on the inner wall of the frame 10, and the flower storage tubes 82 are arranged in an array inside the housing 81. The binocular camera 90 is installed on the bottom surface of the platform 40 corresponding to the retractable harvesting robotic arm mechanism, and its viewing direction is relative to the forward direction of the harvesting cart (e.g., ...). Figure 1 (As shown).

[0028] Example 2: As another preferred embodiment of the technical solution of the present invention, a method for harvesting flowers, using a flower harvesting cart as shown in Example 1, includes the following specific steps: Step S1: First, activate the self-balancing harvesting mechanism to level the platform and avoid the effects of displacement and vibration caused by ground elevation differences; then, use the moving mechanism to control the harvesting trolley to move above the flower rows; including: Step S11, Platform Self-Balancing Control: The sampling frequency of the gyroscope of the inertial sensor is... f , per 1 / f Output angular velocity data w High-frequency noise is removed through adaptive window filtering; window size... N Dynamically adjusted by the rate of change of the signal: ; In the formula: w t Indicates the current angular velocity.w t-1 This represents the angular velocity at the previous moment; w max Indicates the maximum value of angle SU; k This represents the adjustment coefficient (obtained from a large amount of experimental data, typically 10). w fil Indicates the angular velocity after filtering; Then, a Kalman filter model is constructed to fuse attitude data from the gyroscope and accelerometer (within the inertial sensor). Through Kalman filter prediction and update steps, the optimal attitude estimate is obtained. : Equations of state: ; Observation equation: ; In the formula: X k This represents the matrix composed of attitude angles and attitude angular velocities, i.e. A represents the state transition matrix, i.e. B represents the control input matrix, i.e. ; U k This indicates a control input (e.g., the output of an accelerometer). W k Indicates process noise; Z k This represents the attitude observation matrix of the gyroscope and accelerometer, i.e. ; H Represents the observation matrix, i.e. ; V k Indicates observation noise; Because gyroscope sensors have good instantaneous dynamic response but can produce angular drift, while accelerometers have good static accuracy but can fluctuate during motion or when subjected to disturbances; therefore, the simply filtered angular velocity and acceleration data are weighted and fused. ; In the formula: C gyro , C acc These represent the confidence levels of the gyroscope and accelerometer, respectively, calculated based on the gyroscope drift and the accelerometer noise level. These represent the fusion coefficients of the gyroscope and accelerometer, respectively; ; In the formula: This indicates the gyroscope angle at the previous moment. Indicates the gyroscope is in The change in angle during the integral over time. Indicates the angle of the accelerometer; According to the fusion and Construct an attitude gradient differential formula that considers both the current attitude deviation and the attitude change trend to achieve more agile balance adjustment: ; In the formula: Indicates the first i The elongation of the root telescopic mechanism; k 1. k 2 represents the attitude angle adjustment coefficient and the attitude angle change rate adjustment coefficient, respectively (obtained from a large amount of experimental data); Step S12, Controlling the forward direction of the harvester: The control system transmits movement signals to the harvester via the SBUS communication protocol, and then controls the servo motor of the movement mechanism to rotate using the FOC algorithm (Field Oriented Control algorithm) to determine the forward direction of the harvester; specifically: The desired rotation angle of the servo motor is input, and the motor angle measured by the servo motor encoder is subtracted from the desired angle to obtain the current servo motor angle deviation 'e'. Then, the appropriate current output from the servo motor is calculated via the p control. I q Obtain the appropriate voltage required by the servo motor U q : ; In the formula: K p express p Control coefficient; R This indicates the phase resistance of the servo motor; The inverse Parker transform and the inverse Clarke transform are used to determine the vector voltage required by the motor. U q Converted to three-phase voltage for controlling the motor: ; Parker inverse transform: U d and U q Transform back to the α-β coordinate system: ; In the formula: Represents the electrical angle (i.e., the product of the mechanical angle and the number of pole pairs); Inverse Clarke Transform: Converts voltage in the α-β coordinate system into three-phase voltage to drive the inverter. ; pass u a +u b + u c =0, inverse solution yields: ; In the formula: u a , u b , u c This represents the three-phase voltage driving the inverter; Step S13, Harvesting Trolley Movement Control: The control system inputs a speed signal to control the brushless motor of the moving mechanism to rotate, thereby moving the platform of the harvesting trolley to the designated position, specifically: Desired speed input via remote control v hope Subtract the motor rotation speed measured by the encoder to obtain the current speed error. v e The motor's rotation speed v z for: ; In the formula: n , n p These represent the current number of revolutions of the motor and the number of revolutions at the time of the last sampling, respectively. These represent the current motor angle and the motor angle at the time of the last calculation, respectively. T s Indicates the sampling time; The speed value obtained at this time will fluctuate greatly. If used directly, it will cause vibration and noise in the motor. Therefore, it needs to be filtered. ; In the formula: T f Represents the time constant; dt Indicates a time interval; Then, the required vector current output by the motor is calculated using PI control. I q Then, it is converted into the vector voltage required by the motor using Ohm's law. U q : ; In the formula: K p , K i These represent the PI control coefficients; Finally, the required vector voltage for the motor is obtained through Parker inverse transform and Clark inverse transform. Uq This is converted into the three-phase voltage for controlling the motor (see the above description for the specific methods of Parker inverse transformation and Clark inverse transformation). Step S2: Use a binocular camera to transmit images of the flowers below the platform, match the flower image positions with the actual positions to obtain the position that the retractable harvesting robotic arm needs to reach (i.e., the harvesting position), and control the retractable harvesting robotic arm to move to the harvesting position via the guide rail slide mechanism; including: Step S21, Flower Position Matching: This includes flower image segmentation and coordinate transformation. Flower image segmentation: The image output from the stereo camera (640x480 resolution) is first subjected to Gaussian filtering (kernel size=3x3, standard deviation 1.5) to remove noise, and then the OTSU algorithm is used for image segmentation. A segmentation threshold is obtained by weighted fusion of color, texture, and edge features. T : ; In the formula: These represent the statistics for overall color (i.e., HSV space), overall texture (i.e., LBP features), and overall edge (i.e., Canny detection), respectively. , These represent the average color characteristics of the background and the flower area, respectively. , These represent the average texture features of the background and the flower area, respectively. , These represent the average edge features of the background and the flower area, respectively. v 0 indicates the proportion of background pixels to total pixels. v 1 (x) Indicates in x The percentage of pixels in the flower area to the total number of pixels ( v 0+ v 1 (x) =1); Pixels with grayscale values ​​greater than T are identified as flower areas; Coordinate transformation: Transform the camera pixel coordinates ( x p , y p ) represents the physical coordinates of the retractable harvesting robotic arm mechanism. x cu , y cu ): ; In the formula: j 1. j 2 represents the radial distortion coefficient. p 1.p 2 represents the tangential distortion coefficient (optimized by acquiring calibration board images multiple times); Convert the baseline conversion coefficients to spatial variation coefficients: ; In the formula: They represent x direction, y Spatial variation coefficient of direction; They represent x direction, y Spatial variation increment of direction (based on the three-dimensional error model of the camera-robotic arm calibration field, by analyzing the deviation between the theoretical coordinates and the actual coordinates of feature points in the calibration field, the error is fitted in ( x , y The distribution pattern at point () was obtained. f x0 , f y0 Indicates the base conversion factor; ; In the formula: x b0 , y b0 These represent the pixel coordinate reference offset (obtained through camera intrinsic parameter calibration); x arm (t) , y arm (t) Let represent the physical coordinates of the robotic arm at time t; This represents the time difference between image acquisition and coordinate transformation; v x , v y These represent the robotic arm in x , y velocity in direction, a x , a y These represent the robotic arm in x , y Acceleration in the direction of; Step S22, Stepper Motor Control: By controlling the stepper motors of the first and second lead screws, the retractable harvesting robotic arm mechanism is moved above the flowers to facilitate subsequent harvesting; specifically: A sinusoidal current with a 90° phase difference is applied to the A and B phase collars of the two stepper motors of the first and second lead screws: ; In the formula: Imax Indicates the maximum input current; Indicates electrical angle; Thus, the two coils will generate a magnetic field that changes in a regular manner. After the two phase magnetic fields are vectored together, a continuously changing rotating magnetic field can be formed. Therefore, the vector of the combined magnetic field can be rotated by an angle smaller than the motor step angle, so as to accurately position the motor rotor at any angle between the two inherent parts of the motor.

[0029] In this embodiment, the stepper motor used by the harvesting trolley has a step angle of 1.8°, so the motor needs 200 steps to rotate one revolution. Dividing each step of the motor into 16 equal parts and setting the lead screw to 4mm, then: microstep angle = / 16=0.1125°; Number of pulses per motor revolution = 200 * 16 = 3200 pulses / revolution; Pulse equivalent = 4mm / number of pulses per motor revolution = 0.00125mm; Pulse frequency = (speed · number of pulses per motor revolution) / 4mm = 4kHz (set speed is 10mm / s); X , Y The pulse signals of the two-axis stepper motors (i.e., the first stepper motor and the second stepper motor) are output by the timer of the STM32H750 microcontroller, and the timer frequency is set to 10kHz. To ensure synchronous movement of the two axes, a "pulse synchronous transmission" strategy is adopted: for every X-axis pulse output, one Y-axis pulse is output synchronously to avoid positioning deviation caused by "X-axis arriving first and Y-axis arriving later". Once the end-cutting mechanism reaches above the flower, proceed to step S3.

[0030] Step S3: The flower is inserted using a telescopic mechanism, and then the flower image is recognized by the end-effector camera to determine the cutting position of the flower stem. The flower stem is then cut using the end-effector cutting mechanism; including: Step S31: Inserting and positioning the flower stem for cutting: First, the end-point camera uses "template matching" to locate the flower head: multi-pose basic templates are pre-stored, and during actual detection, high-matching samples are automatically selected and dynamically added to the template library, solving the problem of poor adaptability of a single template to variations in flower head morphology (such as petal opening and closing, minor damage). Then, based on the ambient light and flower head variety, the matching weights of different templates are automatically adjusted (e.g., prioritizing anti-high-light templates in strong light environments), improving robustness in complex scenarios. ; In the formula: Indicates the first iThe weight coefficient of each template ( ); NCC i Indicates the first i The average normalized cross-correlation coefficient between each template and historical successfully matched samples; C i Indicates the color feature matching degree; T i Indicates the time decay factor; These represent the corresponding weight coefficients ( ); NCC i By extracting template images F Sample images that successfully matched history S j Obtaining the grayscale matrix: ; In the formula: , These represent the mean gray values ​​of the template and the sample, respectively. C i By converting the images of the template and the region to be matched to HSV space, color histograms are extracted using the H, S, and V channels respectively. The Bach distance between the template and the sample histograms is then calculated. B(F,S) Get: ; In the formula: These represent the template and sample at the 1st and 2nd respectively. k The probability distribution of each interval. m The total number of intervals; T i Used to simulate the decay of flower head freshness over time: ; In the formula: This indicates the duration the template has been stored in the template library; This represents the attenuation coefficient (obtained by fitting actual varieties and a large amount of experimental data). Then, the Canny edge detection algorithm is used to extract the outline edges of the petals and filter out potential candidate regions of the flower head. Within the candidate regions, the "normalized cross-correlation coefficient (NCC)" is used to achieve accurate morphological matching of the flower head, and the geometric constraints between the flower head and the stem are used to eliminate mismatched regions (i.e., based on the geometric relationship that "the center of the flower head is usually located on the extension line of the stem", the distance from the center of the flower head to the extension line of the stem is calculated, and if it exceeds the threshold, it is judged as a mismatch). Then, by dynamically correcting the conversion relationship between pixels and physical length using binocular visual depth information, the accuracy of long-distance positioning is improved, enabling precise positioning of the flower head and providing accurate spatial and dimensional basis for determining the subsequent flower stem cutting position; including: Depth calculation: In binocular vision, depth z deep The relationship with disparity (the difference in the horizontal coordinate of the same point between the left and right cameras) is as follows: ; Always: f Indicates the camera's focal length; B Represented as the baseline distance between the two cameras; Pixel equivalent dynamic adjustment: Set reference distance z The pixel equivalent at 0 is k 0 (unit: mm / pixel) indicates real-time distance z deep Pixel equivalent at k(z) for: ; Physical length and pixel distance d The relationship is: physical length = d · k(z) This enables dynamic correction of pixel physical transformation.

[0031] Subsequently, a combination of primary soft limiting and secondary hard limiting is used to achieve adaptive cutting. Level 1 soft limit is a variety-based early warning system: establishing dynamic early warning thresholds. L yj When the actual length reaches the warning threshold L yj When the threshold reaches 90%, an early warning state is activated. ; In the formula: L jz This represents the standard length for the corresponding flower variety (obtained through fitting a large amount of experimental data). M cs The flower stem maturity coefficient represents the image recognition result. D zj The stem diameter is indirectly measured by pressure sensors (the stem diameter is indirectly calculated by the timing difference and pressure distribution of two sets of pressure sensors arranged inside the telescopic mechanism and at the lower end of the end cutting mechanism). z 1. z 2 represents the correction factor; The method for obtaining the flower stem maturity coefficient is as follows: First, obtain the color similarity with mature flower stems by using the average RGB color value of the flower stem region. Simc : ; In the formula: ( R djc ,G dic ,B djc () represents the average RGB color value of the area of ​​the flower stem to be detected; R cs ,G cs ,B cs () represents the average RGB color value of a mature flower stem; Then, LBP (Local Binary Pattern) was used to extract the surface texture features of the flower stem, and the matching degree between the features and the standard texture features was calculated. Sim t : Based on LBP texture features, the LBP histogram vector of the flower stem to be detected is defined as follows: H djc =[ h t1 , h t2 , ..., h t256 (256 dimensions, corresponding to the frequency of LBP values ​​from 0 to 255), the LBP histogram vector of a standard mature flower stem is H cs =[ h c1 , h c2 , ..., h c256 ]; The grayscale value of any pixel (x, y) in the grayscale image of the flower stem is g c The pixel grayscale values ​​of the 8 domains are as follows: g 0~ g 7 (arranged clockwise): ; ; The frequency of the k-th dimension (corresponding to the LBP value k) of the LBP histogram of the flower stem to be detected is: ; ; ; In the formula: These represent the corresponding weight coefficients ( (determined through training with a large number of samples). M cs∈[0,1], where 0 represents completely immature and 1 represents completely mature; The stem diameter indirectly tested by the pressure sensor is as follows: First, when the flower stem passes through the end-cutting mechanism and the hose, the pressure sensor on the inner wall of the telescopic mechanism and the pressure sensors on both sides of the cut opening of the end-cutting structure will generate pressure trigger signals due to the compression of the flower stem (the pressure sensor on the inner wall of the telescopic mechanism generates a signal through contact with the hose and when the flower stem passes through the hose, squeezing the inner wall pressure sensor). The time difference between the triggering of the signals from the two sensors is recorded. This is combined with the conveying speed of the flower stem (calculated from the transmission speed of the telescopic mechanism). v hjz To obtain the preliminary diameter of the flower stem D start : ; In the formula: d cg0 Indicates the initial spacing of the sensors (obtained through calibration); Then, based on the pressure value distribution of the pressure sensors on both sides, that is... P left , P right The initial diameter is corrected: ; Finally, the weighted average of the detection results from the two sets of pressure sensors was calculated to obtain the stem diameter. D zj : ; In the formula: The two sets of pressure sensors, located on the inner walls of the two telescopic mechanisms and on the inner walls of the lower ends of the end cutting mechanism, respectively represent the stem diameters obtained after correction. These represent the corresponding weighting coefficients (generally) ); The contact pressure is monitored in real time by a pressure sensor, thereby dynamically adjusting the duty cycle signal (automatically increasing when the pressure change rate exceeds a set threshold). A value that accelerates the deceleration process; conversely, a value that decreases the deceleration process. (to achieve smooth adjustment) ; In the formula: Duty(n) Indicates the first n Duty cycle during the second adjustment; P dq This indicates the pressure value detected in real time. P yz This indicates the standard contact pressure for the corresponding variety (obtained through multiple tests). This represents the adjustment coefficient, which is dynamically adjusted according to the pressure change rate (generally 0.05 to 0.2). The condition of the stem is determined by integrating multiple parameters: Contact determination: When the real-time detected contact pressure is not less than P yz And pressure change rate When the force is not less than 0.5 N / s, it is determined that the flower stem has successfully made contact with the end cutting mechanism; ; Missing criterion: When the real-time detected contact pressure is less than P yz Furthermore, if the displacement increment of the end effector is less than 0.5 mm / 100 ms, it is determined to be jammed; Short stem warning: When the duty cycle drops to 10% and no effective pressure value is detected, first drive the telescopic mechanism to rise 2mm, then try extending and lowering 3mm to confirm whether the contact pressure is still less than [previous value]. P yz And check if the displacement increment of the end effector is still less than 0.5mm / 100ms; if so, issue an alarm. The secondary hard limit switch is a forced protection based on physical switches. The telescopic mechanism has two built-in mechanical limit switches. When the motor operating current exceeds the rated current, electronic soft protection is triggered (speed is reduced first and then stopped). At the same time, abnormal fluctuations in the acceleration of the telescopic mechanism are detected to predict mechanical failure. When the distance to the limit switch is 5mm, the speed is automatically reduced to 30% of the rated speed. After each harvest, the benchmark is dynamically corrected based on the deviation between the actual cutting position and the standard position. ; In the formula: JZ new , JZ old These represent the new reference and the original reference (reference: the mechanical zero-point reference of the telescopic mechanism, i.e., the theoretical position coordinates when the telescopic mechanism is fully retracted). WZ sj , WZ bz These represent the actual cutting position and the standard position, respectively. Represents the learning coefficient; A comprehensive calibration is performed after every 10 harvests to correct the temperature of the telescopic mechanism. ; In the formula: This indicates the temperature change after 10 harvests; Step S32, Blade Cutting: By changing the direction of the current in the electromagnets at both ends of the cutting opening, the magnetic poles of the electromagnets are made to be the same as the magnetic poles of the strong magnet. Utilizing the principle of like poles repelling each other, the blade is driven by magnetic force to cut the flower stem. The blade then moves a barrier cloth to seal the cutting opening and prevent the flowers from falling out. Specifically: When installing the strong magnet and the electromagnet, ensure that the N pole of the strong magnet is aligned with the N pole of the electromagnet, thus obtaining an air gap of 0.1–0.5 mm between them. The magnitude of the repulsive force between them is obtained as follows: When the two are in close contact (under ideal conditions) and the air gap is 0mm, the rated attraction force of the electromagnet is: F dc The repulsive force between the two is approximately 294 N, at which point the repulsive force is (0.8~1.0). F dc ; However, under actual conditions, an air gap of 0.1–0.5 mm will appear between the two. The repulsive force decreases sharply with the increase of the air gap in an inverse square relationship, as shown by fitting a large amount of empirical data: ; In the formula: F Represents repulsive force. qx Indicates the air gap; Therefore, under actual conditions, the repulsive force between the two is 20–25. kgf When the electromagnet is energized, it transmits a force of 20-25 degrees to the blade. kgf The thrust causes the blade to quickly cut along a 45° inclined linear guide rail, and then enters the cutting completion judgment.

[0032] Cutting completion determination: A miniature pressure sensor is attached to the rear end of the blade, and the miniature pressure sensor collects the reaction force every 0.001 seconds. F reac When reaction force F reac The cutting is considered complete when the peak value drops sharply from 5-15 kgf to below 0.1 kgf for more than 0.01 seconds; if F reac >1.5· F tar ( F tar If the target cutting force is applied, the anti-jamming logic is triggered. The anti-jamming logic is as follows: a reverse PWM signal (i.e., duty cycle = -30%, I = -0.6A) is output for 0.5s, thereby driving the blade to retract 0.5mm, and then resumes forward cutting until the cutting work is completed.

[0033] Step S4: Place the cut flowers into the corresponding flower storage tube to collect the flowers; then, the retractable harvesting robotic arm resets and awaits the next harvesting operation, including: Flower storage cylinder coordinate calibration: During initialization, manually move the retractable harvesting robotic arm mechanism above each flower storage cylinder and record the corresponding coordinates. X , Y Shaft storage tube number: N x1 , N y1 , N x2 , N y2 … N xn , N yn (Also indicates the number of flower storage tubes); During operation, the coordinates of the flower storage tubes are called in the order of "left to right, top to bottom" (to avoid duplicate storage); Infrared reflective points are set at the edge of each flower storage tube, and a miniature camera and infrared ranging sensor are set at the bottom of the retractable picking robotic arm mechanism. Based on the initial manual calibration, the coordinate drift problem caused by mechanical error and vibration is solved through a dynamic correction mechanism of "vision + sensor fusion". During the operation, after the retractable harvesting robotic arm moves to the vicinity of the calibration coordinates, the miniature camera identifies the markers and calculates the deviation between the actual coordinates and the calibration values. Combined with the height deviation detected by the infrared range sensor The motion trajectory is corrected in real time; then, a "coordinate deviation database" is established to record the deviation patterns under different working durations and temperatures, and the least squares method is used to predict the deviation trend and compensate and correct it in advance.

[0034] In addition, a high-precision ultrasonic liquid level sensor is installed inside the flower storage cylinder to collect the liquid level height H in real time and record the change curve (dH / dt). During operation, based on the number of times the harvesting arm stores flowers (each stored flower stem corresponds to a fixed amount of liquid consumption ΔV), the theoretical liquid level drop value is calculated and compared with the actual detection value to determine whether there is a leak or sensor malfunction. A "dynamic early warning threshold" is also set. When the number of harvests that the remaining liquid can support is less than or equal to a preset value (e.g., 50 times), an alarm is triggered in advance. At the same time, the optimal replenishment amount is predicted based on historical replenishment data, and the "recommended replenishment amount" is displayed via remote control. If an abnormality in dH / dt is detected (e.g., a sudden drop in a short period of time), a "leakage alarm" is immediately triggered, and the equipment is suspended.

[0035] Example 3: As another preferred embodiment of the technical solution of the present invention, based on the solution of embodiment 2, a flower harvesting method further includes structural reset, specifically: a photoelectric limit switch is provided at the cutting end of the end cutting mechanism. When the blade retracts to the initial position, the light path of the limit switch is blocked, the switch outputs a low level, and the controller determines that the blade reset is complete; if no low level is detected within 1 second, the pull-in current is output again until the reset is successful.

[0036] During cutting, the blade pulls the barrier cloth, causing the return spring to rotate, and the return spring to store torque (maximum torque set at 0.5 N·m); a magnetorheological damper is added between the return spring and its spool, and its damping force can be adjusted in real time via an electrical signal; a torque sensor is installed to detect the real-time torque during the release process of the return spring. NJ Combined with the rotational speed signal from the Hall sensor n xh Establish a "torque-speed-damping" mapping model: when n xh ≤ n aq0 When the safe speed threshold is reached, the damper does not operate to ensure recovery efficiency; when n xh > n aq0 At that time, according to NJ and n xh The ratio of the graded increase of damping force (e.g.) NJ / n xh The larger the value, the greater the increase in damping force, thus rapidly reducing the rotational speed; When the mechanism is nearly fully retracted, the damping force increases linearly to its maximum value regardless of the rotational speed, achieving a "soft landing" at the end and extending the life of the mechanism.

Claims

1. A flower harvesting cart based on stem-adaptive cutting and vibration compensation, characterized in that: The system includes a frame, a moving mechanism, a self-balancing harvesting mechanism, a platform, a guide rail slide mechanism, a retractable harvesting robotic arm mechanism, an end-cutting mechanism, a storage mechanism, and a binocular camera. The frame has a portal frame structure with a set of moving mechanisms installed at each of its four corners. There are four sets of self-balancing harvesting mechanisms, which are sequentially located at the four corners of the top of the frame. The top of the self-balancing harvesting mechanism is connected to the bottom of the platform, and the platform has a rectangular frame structure in its transverse cross-section. The guide rail slide mechanism is located on the end face of the platform, and the retractable harvesting robotic arm mechanism is located in the middle of the platform frame. The end-cutting mechanism is located at the bottom of the retractable harvesting robotic arm mechanism. There are two sets of storage mechanisms, which are symmetrically located on the inner walls of both sides of the frame. The binocular camera is located on the bottom of the platform, corresponding to the retractable harvesting robotic arm mechanism, and its viewing direction is towards the direction of the harvesting trolley's movement.

2. The flower harvesting cart based on stem adaptation cutting and vibration compensation according to claim 1, characterized in that: The moving mechanism includes a rubber wheel, a servo motor, a servo motor connector, a motor mounting bracket, and a brushless motor. The top of the servo motor is connected to the bottom of the frame, and the bottom of the servo motor is fitted with a motor mounting bracket via the servo motor connector. The brushless motor is installed inside the motor mounting bracket, and the rubber wheel is fixedly sleeved on the outer wall of the output shaft of the brushless motor. The harvesting self-balancing mechanism includes an electric telescopic rod and a single-rod universal ball joint bearing. One end of the electric telescopic rod is connected to the top of the frame, and the other end is connected to the bottom of the platform via the single-rod universal ball joint bearing.

3. A flower harvesting cart based on stem-adaptive cutting and vibration compensation according to claim 1 or 2, characterized in that: The guide rail slide mechanism includes a first lead screw, a first stepper motor, a sliding crossbar, a second lead screw, a second stepper motor, and a slider. There are two first lead screws, each corresponding to a long rod on the platform end face. The first lead screws are driven by the first stepper motor. The two ends of the sliding crossbar are slidably connected to the end faces of the long rods on the platform. The two ends of the sliding crossbar are threaded through and connected by the corresponding first lead screws. The end face of the sliding crossbar is provided with a second lead screw, which is driven by the second stepper motor. The slider is slidably provided on the end face of the sliding crossbar, and the middle of the slider is threaded through and connected by the second lead screw.

4. A flower harvesting cart based on stem-adaptive cutting and vibration compensation according to claim 2 or 3, characterized in that: The retractable harvesting robotic arm mechanism includes a connector, a telescopic mechanism, a flexible hose, a barrier cover, a cutting support, and an inertial sensor. The bottom surface of the connector is fixedly connected to the end face of the slider, and telescopic mechanisms are respectively set at both ends of the connector. The two telescopic mechanisms are located at the two ends of a diagonal line of the connector, and a ring-shaped cutting support is set at the bottom of the two telescopic mechanisms. The top surface of the cutting support is equipped with a barrier cover through multiple evenly distributed connecting rods, and a flexible hose is set at the top of the barrier cover. An end-effector camera is installed inside the barrier cover, and the inertial sensor is set on the end face of the connector. The bottom of the barrier cover is set with a sloping structure. The structure includes a cutting mechanism located at the bottom of the barrier cover on the upper side of the cutting support, comprising a blade holder, a cutting opening, a return spring disc, a barrier cloth, a blade, an electromagnet, and a strong magnet. The outer sidewall of the blade holder is penetrated and fixedly connected by multiple connecting rods. A cutting opening is opened in the middle of the blade holder, and electromagnets are respectively installed on the two sidewalls of the cutting opening. The blade is slidably installed in the cutting opening, and a strong magnet is fixedly installed on the end face of the blade. A return spring disc is installed at the top of the inclined surface of the blade holder. One end of the barrier cloth is fixedly connected to the outer wall of the return spring disc and wrapped around the outer wall of the return spring disc, and the other end is fixedly connected to the end face of the strong magnet.

5. A flower harvesting cart based on stem-adaptive cutting and vibration compensation according to claim 4, characterized in that: The storage mechanism includes an outer shell and flower storage tubes. The outer shell is fixedly installed on the inner wall of the frame, and the flower storage tubes are arranged in an array inside the outer shell.

6. A flower harvesting cart based on stem-adaptive cutting and vibration compensation according to claim 5, characterized in that: The specific steps for the harvesting cart to harvest flowers are as follows: Step S1: First, activate the self-balancing harvesting mechanism to keep the platform level and avoid the effects of displacement and vibration caused by ground elevation differences; then, use the moving mechanism to control the harvesting trolley to move above the flower ridge. Step S2: Use a binocular camera to transmit images of flowers below the platform, match the flower image positions with the actual positions to obtain the position that the retractable picking robotic arm needs to reach, and control the retractable picking robotic arm to move to the harvesting position through the guide rail slide mechanism. Step S3: The flower is inserted through the telescopic mechanism, and the flower image is identified by the end camera to determine the cutting position of the flower stem. The flower stem is then cut by the end cutting mechanism. Step S4: Place the cut flowers into the corresponding flower storage tube to collect the flowers; then, the retractable picking robotic arm mechanism resets and waits for the next harvesting operation.

7. A flower harvesting cart based on stem-adaptive cutting and vibration compensation according to claim 6, characterized in that: Step S1 specifically involves: Step S11, Platform Self-Balancing Control: The sampling frequency of the gyroscope of the inertial sensor is... f , per 1 / f Output angular velocity data w High-frequency noise is removed through adaptive window filtering; window size... N Dynamically adjusted by the rate of change of the signal: ； In the formula: w t Indicates the current angular velocity. w t-1 This represents the angular velocity at the previous moment; w max Indicates the maximum value of angle SU; k Indicates the adjustment coefficient; w fil Indicates the angular velocity after filtering; Then, a Kalman filter model is constructed to fuse attitude data from the gyroscope and accelerometer. Through Kalman filter prediction and update steps, the optimal attitude estimate is obtained. : Equations of state: ; Observation equation: ; In the formula: X k This represents the matrix composed of attitude angles and attitude angular velocities, i.e. ; A represents the state transition matrix, i.e. B represents the control input matrix, i.e. ; U k Indicates control input; W k Indicates process noise; Z k This represents the attitude observation matrix of the gyroscope and accelerometer, i.e. ; H Represents the observation matrix, i.e. ; V k Indicates observation noise; The angular velocity and acceleration data after simple filtering are then weighted and fused: ； In the formula: C gyro , C acc These represent the confidence levels of the gyroscope and accelerometer, respectively, calculated based on the gyroscope drift and the accelerometer noise level. These represent the fusion coefficients of the gyroscope and accelerometer, respectively; ； In the formula: This indicates the gyroscope angle at the previous moment. Indicates the gyroscope is in The change in angle during the integral over time. Indicates the angle of the accelerometer; According to the fusion and Construct an attitude gradient differential formula that considers both the current attitude deviation and the attitude change trend to achieve more agile balance adjustment: ； In the formula: Indicates the first i The elongation of the root telescopic mechanism; k 1. k 2 represents the attitude angle adjustment coefficient and the attitude angle change rate adjustment coefficient, respectively; Step S12: Controlling the forward direction of the harvester: The control system transmits the movement signal to the harvester through the SBUS communication protocol, and then controls the servo motor of the movement mechanism to rotate through the FOC algorithm to determine the forward direction of the harvester; Step S13, Harvesting Trolley Movement Control: The control system inputs a speed signal to control the brushless motor of the moving mechanism to rotate, thereby driving the platform of the harvesting trolley to move to the designated position.

8. A flower harvesting cart based on stem-adaptive cutting and vibration compensation according to claim 6, characterized in that: Step S2 specifically involves: Step S21, Flower Position Matching: This includes flower image segmentation and coordinate transformation. Flower image segmentation: The image output from the stereo camera is first subjected to Gaussian filtering to remove noise, then the OTSU algorithm is used for image segmentation, and color, texture, and edge features are weighted and fused to obtain the segmentation threshold. T : ； In the formula: These represent the statistics for overall color, overall texture, and overall edge, respectively. , These represent the average color characteristics of the background and the flower area, respectively. , These represent the average texture features of the background and the flower area, respectively. , These represent the average edge features of the background and the flower area, respectively. v 0 indicates the proportion of background pixels to total pixels. v 1 (x) Indicates in x The proportion of pixels in the flower area to the total number of pixels; Pixels with grayscale values ​​greater than T are identified as flower areas; Coordinate transformation: Transform the camera pixel coordinates ( x p , y p ) represents the physical coordinates of the retractable harvesting robotic arm mechanism. x cu , y cu ): ； In the formula: j 1. j 2 represents the radial distortion coefficient. p 1. p 2 represents the tangential distortion coefficient; Convert the baseline conversion coefficients to spatial variation coefficients: ； In the formula: They represent x direction, y Spatial variation coefficient of direction; They represent x direction, y Incremental spatial change in direction; f x0 , f y0 Indicates the base conversion factor; ； In the formula: x b0 , y b0 These represent the pixel coordinate baseline offset; x arm (t) , y arm (t) Let represent the physical coordinates of the robotic arm at time t; This represents the time difference between image acquisition and coordinate transformation; v x , v y These represent the robotic arm in x , y velocity in direction, a x , a y These represent the robotic arm in x , y Acceleration in the direction of; Step S22, Stepper Motor Control: By controlling the stepper motors of the first and second lead screws, the retractable picking robotic arm mechanism is moved above the flowers to facilitate subsequent harvesting.

9. A flower harvesting cart based on stem-adaptive cutting and vibration compensation according to claim 6, characterized in that: Step S3 specifically involves: Step S31: Inserting and positioning the flower stem for cutting: First, the end-point camera uses "template matching" to locate the flower head position: that is, it pre-stores multiple pose basic templates, automatically selects high-matching samples during actual detection, and dynamically supplements them into the template library; then, it automatically adjusts the matching weight of different templates according to the ambient light and flower head variety. ； In the formula: Indicates the first i The weight coefficients of each template; NCC i Indicates the first i The average normalized cross-correlation coefficient between each template and historical successfully matched samples; C i Indicates the color feature matching degree; i>T i Indicates the time decay factor; These represent the corresponding weight coefficients; NCC i By extracting template images F Sample images that successfully matched history S j Obtaining the grayscale matrix: ； In the formula: , These represent the mean gray values ​​of the template and the sample, respectively. C i By converting the images of the template and the region to be matched to HSV space, color histograms are extracted using the H, S, and V channels respectively. The Bach distance between the template and the sample histograms is then calculated. B(F,S) Get: ； In the formula: These represent the template and sample at the 1st and 2nd respectively. k The probability distribution of each interval. m The total number of intervals; T i Used to simulate the decay of flower head freshness over time: ； In the formula: This indicates the duration the template has been stored in the template library; Indicates the attenuation coefficient; Then, the Canny edge detection algorithm is used to extract the outline edges of the petals and screen out potential candidate areas of the flower head. Within the candidate areas, the "normalized cross-correlation coefficient" is used to achieve accurate morphological matching of the flower head, and the geometric constraints between the flower head and the stem are used to eliminate mismatched areas. Then, the conversion relationship between pixels and physical length is dynamically corrected by the depth information of binocular vision to improve the accuracy of long-distance positioning and achieve accurate positioning of the flower head, providing accurate spatial and dimensional basis for determining the subsequent flower stem cutting position. Subsequently, a combination of primary soft limit and secondary hard limit is used to achieve adaptive cutting. Level 1 soft limit is a variety-based early warning system: establishing dynamic early warning thresholds. L yj When the actual length reaches the warning threshold L yj When the threshold reaches 90%, an early warning state is activated. ； In the formula: L jz Indicates the standard length for the corresponding flower variety; M cs The flower stem maturity coefficient represents the value recognized by the image. D zj This indicates the stem diameter indirectly measured by the pressure sensor. z 1. z 2 represents the correction factor; The duty cycle signal is dynamically adjusted based on real-time monitoring of the contact pressure by a pressure sensor. ； In the formula: Duty(n) Indicates the first n Duty cycle during the second adjustment; P dq This indicates the pressure value detected in real time. P yz Indicates the standard contact pressure for the corresponding variety; This represents the adjustment coefficient, which is dynamically adjusted based on the rate of pressure change. The condition of the stem is determined by integrating multiple parameters: Contact determination: When the real-time detected contact pressure is not less than P yz And pressure change rate When the force is not less than 0.5 N / s, it is determined that the flower stem has successfully made contact with the end cutting mechanism; ； Missing criterion: When the real-time detected contact pressure is less than P yz Furthermore, if the displacement increment of the end effector is less than 0.5 mm / 100 ms, it is determined to be jammed; Short stem warning: When the duty cycle drops to 10% and no effective pressure value is detected, first drive the telescopic mechanism to rise 2mm, then try extending and lowering 3mm to confirm whether the contact pressure is still less than [previous value]. P yz And check if the displacement increment of the end effector is still less than 0.5mm / 100ms; if so, issue an alarm. The secondary hard limit switch is a forced protection based on physical switches. The telescopic mechanism has two built-in mechanical limit switches. When the motor operating current exceeds the rated current, the electronic soft protection is triggered. At the same time, abnormal fluctuations in the acceleration of the telescopic mechanism are detected to predict mechanical failure. When the distance to the limit switch is 5mm, the speed is automatically reduced to 30% of the rated speed. After each harvest, the benchmark is dynamically corrected based on the deviation between the actual cutting position and the standard position. ； In the formula: JZ new , JZ old These represent the new benchmark and the original benchmark, respectively. WZ sj , WZ bz These represent the actual cutting position and the standard position, respectively. Represents the learning coefficient; A comprehensive calibration is performed after every 10 harvests to correct the temperature of the telescopic mechanism. ； In the formula: This indicates the temperature change after 10 harvests; Step S32, Blade Cutting: By changing the direction of the current in the electromagnets at both ends of the cutting opening, the magnetic poles of the electromagnets are made to be the same as the magnetic poles of the strong magnetic block. Using the principle of like poles repelling each other, the blade is driven by magnetic force to cut the flower branches. The blade also drives the barrier cloth to seal the cutting opening and prevent the flowers from falling.

Images