CN111837593A - Novel peucedanum praeruptorum weeding machine based on machine vision and convolutional neural network algorithm - Google Patents
Novel peucedanum praeruptorum weeding machine based on machine vision and convolutional neural network algorithm Download PDFInfo
- Publication number
- CN111837593A CN111837593A CN202010728415.6A CN202010728415A CN111837593A CN 111837593 A CN111837593 A CN 111837593A CN 202010728415 A CN202010728415 A CN 202010728415A CN 111837593 A CN111837593 A CN 111837593A
- Authority
- CN
- China
- Prior art keywords
- weeding
- peucedanum
- machine
- neural network
- convolutional neural
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A01—AGRICULTURE; FORESTRY; ANIMAL HUSBANDRY; HUNTING; TRAPPING; FISHING
- A01D—HARVESTING; MOWING
- A01D34/00—Mowers; Mowing apparatus of harvesters
- A01D34/01—Mowers; Mowing apparatus of harvesters characterised by features relating to the type of cutting apparatus
- A01D34/412—Mowers; Mowing apparatus of harvesters characterised by features relating to the type of cutting apparatus having rotating cutters
- A01D34/63—Mowers; Mowing apparatus of harvesters characterised by features relating to the type of cutting apparatus having rotating cutters having cutters rotating about a vertical axis
- A01D34/64—Mowers; Mowing apparatus of harvesters characterised by features relating to the type of cutting apparatus having rotating cutters having cutters rotating about a vertical axis mounted on a vehicle, e.g. a tractor, or drawn by an animal or a vehicle
-
- A—HUMAN NECESSITIES
- A01—AGRICULTURE; FORESTRY; ANIMAL HUSBANDRY; HUNTING; TRAPPING; FISHING
- A01D—HARVESTING; MOWING
- A01D34/00—Mowers; Mowing apparatus of harvesters
- A01D34/01—Mowers; Mowing apparatus of harvesters characterised by features relating to the type of cutting apparatus
- A01D34/412—Mowers; Mowing apparatus of harvesters characterised by features relating to the type of cutting apparatus having rotating cutters
- A01D34/63—Mowers; Mowing apparatus of harvesters characterised by features relating to the type of cutting apparatus having rotating cutters having cutters rotating about a vertical axis
- A01D34/73—Cutting apparatus
-
- A—HUMAN NECESSITIES
- A01—AGRICULTURE; FORESTRY; ANIMAL HUSBANDRY; HUNTING; TRAPPING; FISHING
- A01D—HARVESTING; MOWING
- A01D34/00—Mowers; Mowing apparatus of harvesters
- A01D34/01—Mowers; Mowing apparatus of harvesters characterised by features relating to the type of cutting apparatus
- A01D34/412—Mowers; Mowing apparatus of harvesters characterised by features relating to the type of cutting apparatus having rotating cutters
- A01D34/63—Mowers; Mowing apparatus of harvesters characterised by features relating to the type of cutting apparatus having rotating cutters having cutters rotating about a vertical axis
- A01D34/82—Other details
-
- A—HUMAN NECESSITIES
- A01—AGRICULTURE; FORESTRY; ANIMAL HUSBANDRY; HUNTING; TRAPPING; FISHING
- A01D—HARVESTING; MOWING
- A01D67/00—Undercarriages or frames specially adapted for harvesters or mowers; Mechanisms for adjusting the frame; Platforms
-
- A—HUMAN NECESSITIES
- A01—AGRICULTURE; FORESTRY; ANIMAL HUSBANDRY; HUNTING; TRAPPING; FISHING
- A01D—HARVESTING; MOWING
- A01D69/00—Driving mechanisms or parts thereof for harvesters or mowers
- A01D69/02—Driving mechanisms or parts thereof for harvesters or mowers electric
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- Environmental Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Soil Working Implements (AREA)
- Guiding Agricultural Machines (AREA)
Abstract
The invention relates to a novel peucedanum praeruptorum dunn weeder based on machine vision and a convolutional neural network algorithm, which comprises a motion control system, an image processing target detection system, a DELTA mechanical arm and a weeding mechanism. The motion control system is used for controlling the robot to move, turn and regulate the speed; in the image processing system, a camera acquires real-time image information and transmits the real-time image information to a processor for image recognition, a convolutional neural network for detecting the TENSOFLOW characteristic extraction target is adopted for detecting, classifying and positioning the peucedanum and the weeds, a path is fitted by a least square method for planning and navigating the field path according to the center position information of the peucedanum for drilling, a TENSOFLOW characteristic extraction target detection algorithm calibrates the coordinate information of the weeds, and the weeds are removed by matching with a DELTA triaxial mechanical arm and a weeding mechanism. The peucedanum praeruptorum weeding machine is compact in structure, reasonable in layout, firm, durable, high in working efficiency, labor-saving, strong in practicability and suitable for weeding among crop plants.
Description
Technical Field
The invention relates to a field weeding machine for peucedanum praeruptorum, in particular to a novel weeding machine for peucedanum praeruptorum based on machine vision and a convolutional neural network algorithm.
Background
Radix peucedani is a Chinese medicinal material with the functions of depressing qi, resolving phlegm, dispersing wind and clearing heat, and its rhizome is used as medicine. The rhizome traditional Chinese medicinal materials cannot use herbicides, fertilizers and pesticides in the planting process, otherwise the quality and the pesticide effect of the rhizome traditional Chinese medicinal materials are affected, and even the rhizome traditional Chinese medicinal materials have opposite effects. In addition, the labor cost is increased, the labor is difficult, the labor is expensive, the large-scale production is difficult, and the increase of the income of medical farmers is seriously restricted. The manual weeding consumes time and labor, the working efficiency is very low, and the cost required by the manual weeding is high every year. Aiming at solving the problem of 'inorganic availability' in the key link of the current planting of the peucedanum praeruptorum.
The traditional mechanical weeding mainly aims at weeding among crop rows, has larger weeding error, can not effectively remove weeds, and can not effectively remove weed among plants, so that the weeding of the peucedanum praeruptorum is still a big problem. The peucedanum praeruptorum weeding also has the characteristic that weeding operation is mainly concentrated in the seedling stage of the peucedanum praeruptorum, and when seedlings grow, weeds cannot influence the growth of the seedlings, so that weeding in the seedling stage is completed.
The convolutional neural network has obvious advantages in target detection. With the development of artificial intelligence technology, robots are increasingly designed and applied in the field of agricultural science. The invention designs the peucedanum root weeder based on the convolutional neural network deep learning algorithm, the deep learning algorithm based on target detection is adopted, the weeding structure adopts a four-wheel type frame structure, and the weeding mode adopts mechanical weeding.
Disclosure of Invention
The invention provides a weeding machine based on a machine vision technology and a neural network algorithm target positioning algorithm, wherein a convolutional neural network target detection algorithm is adopted to identify and distinguish peucedanum praeruptorum and weeds, so that the identification precision is effectively improved. The weed position information is transmitted to the actuating mechanism for control through serial port communication, the requirement of quick reaction under working conditions is met by adopting a Delta parallel mechanical arm with quick reaction, and the execution precision of the whole machine is guaranteed to be +/-1 mm, and the integral accuracy is more than or equal to 90%.
In order to solve the problems, the invention adopts the following technical scheme to realize:
a novel whiteflower hogfennel root weeder based on machine vision and a convolutional neural network algorithm comprises a motion control system, an image processing target detection system and a weeding mechanism, and also comprises a battery for charging and discharging, wherein the battery adopts a lithium battery, so that the electric quantity is large, and the endurance time is long; the motion control system comprises a frame, a damping device, wheels, a brushless direct current motor, an STM32 motion controller, a ZM6615 direct current brushless driver and a Hall sensor; the support frame of each wheel is provided with a damping device, and the brushless direct current motor is used as mobile power output and is electrically connected with the STM32 motion controller through a ZM6615 direct current brushless driver; the image processing target detection system comprises an electrical cabinet, an industrial camera and an industrial computer; the electrical cabinet is internally integrated with an electrical equipment motor driving circuit, image processing hardware and an MCU (microprogrammed control unit) singlechip, and the image processing target detection system realizes image target detection and positioning algorithm based on a TENSORFLOW deep learning convolutional neural network algorithm; the weeding mechanism comprises a speed reducer, a DELTA mechanical arm, a mechanical arm motor, a mechanical driving arm, a ball hinge, a mechanical driven arm, a weeding motor, a movable platform and a weeding cutter; DELTA arm is installed under the frame, and the arm motor adopts three 57 step motor, and each other becomes 120, motor output torque 2.2N/M, and mechanical driving arm links to each other with the reduction gear axle, and mechanical driving arm and mechanical driven arm all link to each other through the ball hinge with moving the platform, and the weeding motor is installed on moving the platform, and the weeding cutter is installed epaxially at the weeding motor.
Further, the STM32 motion controller adopts an STM32F407VET6 microcontroller, an STM32F407VET6 single chip microcomputer is arranged in the STM32F407VET6 microcontroller, receives weed position information transmitted by an industrial personal computer, and is electrically connected with the STM32 motion controller through an HBS57 closed-loop driver.
Furthermore, four wheels are respectively provided with a set of independent vibration damper, the driving motors of the wheels adopt brushless direct current motors, a closed loop feedback system is formed by using encoders, the rotating speed of the motors is monitored in real time to control the speed of the vehicle so as to cooperate with a weeding mechanism to complete weeding, the four wheels are respectively provided with one brushless direct current motor, the four wheels can be steered by applying differential speed and differential reverse steering, and the differential reverse steering can realize in-situ steering.
Furthermore, the mechanical arm adopts a three-axis parallel connection DELTA mechanical arm, and has the advantages of high movement speed, high response speed, flexibility, high precision and good stability, and can realize various complex movements; 57 the stepping motor is connected with the mechanical driving arm through a speed reducer with the speed reduction ratio of 10:1 to improve output torque, the mechanical driving arm is made of aluminum alloy, the mechanical driven arm is made of a high-strength carbon fiber tube, and the mechanical driving arm is connected with the mechanical driven arm through a ball hinge.
Furthermore, the weeding cutter adopts a self-sharpening blade, is arranged on a movable platform of the DELTA mechanical arm, changes the position along with the movement moment of the DELTA mechanical arm, and rotates at high speed to cut off weeds.
Furthermore, the machine vision of the image processing target detection system adopts a high-definition industrial camera to acquire real-time image information, a TENSOFLOW frame convolutional neural network YoloV4 target detection algorithm is used for identifying weeds, and the results are transmitted to an executing mechanism, so that the weeding function is realized; the convolutional neural network adopts a target detection convolutional network algorithm of feature extraction, the trunk feature extraction network adopts CSPDACKNET53, the feature extraction capability is enhanced by using SPP and PANET structures, the MOSAIC data is enhanced, and CIOU is used as LOSS regression and MISH activation functions; the input image is a 416 × 416 × 3RGB three-channel color picture, and pixel coordinate values, scores, and classifications of the detection target are output.
Further, the image processing target detection system detects the center coordinates of the detected peucedanum of the drill by adopting a target detection depth learning algorithm and fits a lane line by adopting a least square method; in the moving process of the robot, if the number of missing seedlings on one side is small, path fitting is not influenced; when the number of the missing seedlings on one side is large, the path can be planned according to the plants on the other side, so that the autonomous navigation of the weeding machine is realized, and the weeding machine walks among the planted peucedanum praeruptorum plants.
Further, after the image processing target detection system transmits the image acquired by the industrial camera to an industrial computer in real time, the trained model industrial computer is used for fitting an ideal path, and then the industrial computer sends the planned path to the STM32 motion controller in a USB-to-serial port mode; the STM32 motion controller sends a pulse modulation (PWM) pulse signal to a signal control end of the ZM6615 direct-current brushless driver, and the signal control end is used for driving two front wheels of the whiteflower hogfennel weeding machine, and the whiteflower hogfennel weeding machine realizes autonomous motion; when the current weed killer deviates, the rotating speed of the brushless direct current motors on the two sides of the peucedanum weed killer is adjusted to adjust the movement direction, and meanwhile, the Hall sensor converts the rotating speed of the brushless direct current motors into pulse signals and transmits the pulse signals to the STM32 movement controller for accurate steering control; the brushless direct current motor shown in the control process is connected with the STM32 motion controller through a ZM6615 direct current brushless driver, and the STM32 motion controller outputs PWM pulse signals to control the speed and walk in the field through a motor driving circuit, so that the advancing, turning and speed regulation are realized.
Further, the image processing target detection system detects the position of weeds by adopting a deep learning algorithm, and performs sampling stacking through multiple convolution, pooling and stacking of residual edges; after deep network feature extraction, selecting a prediction box with the maximum score by adopting non-maximum inhibition, and finally outputting the position coordinate of the weeds; and calibrating the specific relation between the actual position coordinates and the pixel coordinates through the calibration plate, and determining the actual position of the coordinate point in the graph.
Further, after the image processing target detection system transmits the image acquired by the industrial camera to an industrial computer in real time, the trained model is used for target detection by the industrial computer, and weeds and score conditions of a prediction frame are marked; calculating the actual spatial coordinates of the weeds through calibration, and then sending the weed position information to a controller-STM 32 motion controller of a DELTA mechanical arm by an industrial computer in a mode of converting USB into TTL serial ports; the STM32 motion controller calculates the weed position information to control the joint node of DELTA mechanical arm to realize the displacement motion of weeding mechanism by using the rotating angle of 57 stepping motors. The MOTION of the DELTA mechanical arm adopts a space interpolation algorithm to realize a space arbitrary moving track, and adopts an ADEPT MOTION portal MOTION path for avoiding mistakenly removing the peucedanum and solving the problem of mechanism vibration during cross-region weeding.
Compared with the prior art, the invention has the following beneficial effects: compared with the prior art, the invention adopts the four-wheel vehicle chassis to additionally arrange the damping device to be suitable for weeding on the terrain; the invention effectively identifies and distinguishes weeds and crops by using machine vision and digital image technology, thereby avoiding injuring the crops by mistake; the power is output by the direct current motor and is connected with the track driving wheel through the conveying device, and the weeding mechanical structure adopts a three-shaft parallel shaft mechanical arm to install a mechanical cutter for cutting weeds.
Drawings
Fig. 1 is a schematic diagram of the overall structure of a novel peucedanum weeder based on machine vision and a convolutional neural network algorithm.
Fig. 2 is a schematic diagram of a frame structure of a novel peucedanum weeding machine based on machine vision and a convolutional neural network algorithm.
FIG. 3 is a schematic structural diagram of a Delta mechanical arm in a weeding mechanism of a novel peucedanum weeding machine based on machine vision and a convolutional neural network algorithm.
Fig. 4 is a flow chart of a target detection system of the novel peucedanum weeder based on machine vision and a convolutional neural network algorithm.
Fig. 5 is a flow chart of the novel machine vision and convolutional neural network algorithm-based whiteflower hogfennel weeding machine.
Description of reference numerals: the device comprises a frame 1, a damping device 11, wheels 12, a brushless direct current motor 13, an electrical cabinet 2, an industrial camera 3, a DELTA mechanical arm 4, a stepping motor 41 to 57, a mechanical driving arm 42, a spherical hinge 43, a mechanical driven arm 44, a weeding motor 45, a movable platform 46 and a weeding cutter 47.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1, the novel machine vision and convolutional neural network algorithm-based whiteflower hogfennel weeding machine comprises three parts, namely a motion control system, an image processing target detection system and a weeding mechanism, and also comprises a battery for charging and discharging; the motion control system comprises a frame 1, a damping device 11, wheels 12, a brushless direct current motor 13, an STM32 motion controller, a ZM6615 direct current brushless driver and a Hall sensor; the support frame of each wheel 12 is provided with a damping device 11, and a brushless direct current motor 13 is used as mobile power output and is electrically connected with an STM32 motion controller through a ZM6615 direct current brushless driver; the image processing target detection system comprises an electrical cabinet 2, an industrial camera 3 and an industrial computer; an electrical equipment motor driving circuit, image processing hardware and an MCU singlechip are integrated in the electrical cabinet 2, and an image processing target detection system realizes an image target detection and positioning algorithm based on a TENSORFLOW deep learning convolutional neural network algorithm; the weeding mechanism comprises a speed reducer, a DELTA mechanical arm 4, a mechanical arm motor, a mechanical driving arm 42, a spherical hinge 43, a mechanical driven arm 44, a weeding motor 45, a movable platform 46 and a weeding cutter 47; DELTA mechanical arm 4 is installed under frame 1, and the mechanical arm motor adopts three 57 step motor 41, and each other becomes 120, motor output torque 2.2N/M, and mechanical master arm 42 links to each other with the reduction gear axle, and mechanical master arm 42 and mechanical slave arm 44 all link to each other through ball hinge 43 with moving platform 46, and weeding motor 45 installs on moving platform 46, and weeding cutter 47 installs on weeding motor 45's axle.
As shown in fig. 2, the STM32 motion controller adopts an STM32F407VET6 microcontroller, an STM32F407VET6 single chip microcomputer is arranged in the STM32F407VET6 microcontroller, receives weed position information transmitted by an industrial personal computer, and is electrically connected with the STM32 motion controller through an HBS57 closed-loop driver; four wheels 12 are respectively provided with a set of independent vibration damper 11, the driving motors of the wheels 12 adopt brushless direct current motors 13, a closed loop feedback system is formed by using encoders, the motor rotating speed is monitored in real time to control the vehicle speed so as to cooperate with a weeding mechanism to complete weeding, the four wheels 12 are respectively provided with one brushless direct current motor 13, the differential speed and the differential reverse direction can be applied during steering, and the differential reverse direction can realize in-situ steering.
As shown in fig. 3, 57 the stepping motor 41 is connected with the mechanical driving arm 42 through a speed reducer with the speed reduction ratio of 10:1 to improve output torque, the mechanical driving arm 42 is made of aluminum alloy, the mechanical driven arm 44 is made of high-strength carbon fiber tubes, the mechanical driving arm 42 is connected with the mechanical driven arm 44 through a ball hinge 43, the weeding cutter 47 is a self-sharpening blade and is arranged on a movable platform 46 of the DELTA mechanical arm 4, the position of the weeding cutter 47 of the self-sharpening blade is changed along with the movement time of the DELTA mechanical arm 4, and the weeding cutter 47 of the self-sharpening blade rotates at high speed to cut off weeds; the machine vision of the image processing target detection system adopts a high-definition industrial camera 3 to obtain real-time image information, and a TENSORFLOW frame convolutional neural network YoloV4 target detection algorithm is used for identifying weeds and transmitting results to an actuating mechanism to realize a weeding function;
the convolutional neural network adopts a target detection convolutional network algorithm of feature extraction, the trunk feature extraction network adopts CSPDACKNET53, the feature extraction capability is enhanced by using SPP and PANET structures, the MOSAIC data is enhanced, and CIOU is used as LOSS regression and MISH activation functions; the input image is a 416 x 3RGB three-channel color picture, and the pixel coordinate value, the score and the classification of the detection target are output;
the image processing target detection system detects the center coordinates of the detected peucedanum of the drill by adopting a target detection deep learning algorithm and fits a lane line by adopting a least square method; in the moving process of the robot, if the number of missing seedlings on one side is small, path fitting is not influenced; when the number of the missing seedlings on one side is large, the path can be planned according to the plants on the other side, so that the autonomous navigation of the weeding machine is realized, and the weeding machine walks among the planted peucedanum praeruptorum plants.
The image processing target detection system transmits the image acquired by the industrial camera 3 to an industrial computer in real time, then the trained model industrial computer is used for fitting an ideal path, and then the industrial computer sends the planned path to an STM32 motion controller in a USB-to-serial port mode; the STM32 motion controller sends a pulse modulation (PWM) pulse signal to a signal control end of the ZM6615 direct-current brushless driver, and the signal control end is used for driving two front wheels of the whiteflower hogfennel weeding machine, and the whiteflower hogfennel weeding machine realizes autonomous motion; when the current weed killer deviates, the rotating speed of the brushless direct current motor 13 on the two sides of the peucedanum weed killer is adjusted to adjust the moving direction, and meanwhile, the Hall sensor converts the rotating speed of the brushless direct current motor 13 into a pulse signal and transmits the pulse signal to the STM32 motion controller for accurate steering control; the brushless direct current motor 13 and the STM32 motion controller are connected through a ZM6615 direct current brushless driver in the control process, and the STM32 motion controller outputs PWM pulse signals to control the speed and walk in the field through a motor driving circuit, so that the advancing, turning and speed regulation are realized.
As shown in fig. 4 and 5, the image processing target detection system detects the position of the weeds by using a deep learning algorithm, and performs upsampling stacking through multiple convolution, pooling and stacking of residual edges; after deep network feature extraction, selecting a prediction box with the maximum score by adopting non-maximum inhibition, and finally outputting the position coordinate of the weeds; and calibrating the specific relation between the actual position coordinates and the pixel coordinates through the calibration plate, and determining the actual position of the coordinate point in the graph.
As shown in fig. 4 and 5, the image processing target detection system transmits the image acquired by the industrial camera 3 to the industrial computer in real time, and then the trained model is used for target detection by the industrial computer to mark weeds and score conditions of the prediction frame; calculating the actual spatial coordinates of the weeds through calibration, and then sending the weed position information to a controller-STM 32 motion controller of the DELTA mechanical arm 4 by the industrial computer in a USB-to-TTL serial port mode; the STM32 motion controller calculates the joint node of the control DELTA mechanical arm 4 by the received weed position information, and realizes the displacement motion of the weeding mechanism by using the rotating angle of the 57 stepping motor 41. The MOTION of the DELTA mechanical arm 4 adopts a space interpolation algorithm to realize a space arbitrary moving track, and adopts an ADEPT MOTION portal MOTION path for avoiding mistakenly removing the peucedanum and solving the problem of mechanism vibration during cross-region weeding.
As shown in fig. 4 and 5, in the workflow diagram of the whiteflower hogfennel weeding machine, when the weeding machine is powered on the planting field of the planted whiteflower hogfennel, the system starts to work, the industrial camera collects the video information of the farmland in real time and transmits the video information to the industrial computer, each frame of image is obtained through the computer processing, the system processes the image into a pixel image with the size of 416 × 416, then the image is input into a target detection depth learning frame based on TENSOFLOW to perform prediction classification and positioning, and the actual center coordinates of the whiteflower hogfennel and the weeds on the ground are obtained through the calibration of pixel coordinates and the actual physical coordinates. The industrial computer fits a travel route graph through a least square method according to the central coordinates of the peucedanum praeruptorum and sends the travel route graph to the STM32F407VET6 motion controller through a USB-to-serial port mode, and then the STM32F407VET6 motion controller outputs PWM pulse information to control the motion of the machine. Meanwhile, the industrial computer sends the position coordinates of the weeds to the STM32FF407VET6 controller of the DELTA mechanical arm 4, and then the STM32FF407VET6 controller of the DELTA mechanical arm 4 controls the three 57 stepping motors 41 to enable the DELTA mechanical arm 4 to reach the position of the weeds, thereby realizing the weeding function.
While only the preferred embodiments of the present invention have been described, it should be understood by those skilled in the art that the present invention is not limited to the above-described embodiments, and that various changes and modifications may be made without departing from the spirit and scope of the invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (10)
1. A novel peucedanum weeder based on machine vision and convolution neural network algorithm is characterized in that: the whiteflower hogfennel root weeder comprises three parts, namely a motion control system, an image processing target detection system and a weeding mechanism, and also comprises a battery for charging and discharging;
the motion control system comprises a frame (1), a damping device (11), wheels (12), a brushless direct current motor (13), an STM32 motion controller, a ZM6615 direct current brushless driver and a Hall sensor; the support frame of each wheel (12) is provided with a damping device (11), and a brushless direct current motor (13) is used as a mobile power output and is electrically connected with an STM32 motion controller through a ZM6615 direct current brushless driver;
the image processing target detection system comprises an electrical cabinet (2), an industrial camera (3) and an industrial computer; an electrical equipment motor driving circuit, image processing hardware and an MCU singlechip are integrated in the electrical cabinet (2), and the image processing target detection system realizes image target detection and positioning algorithm based on a convolutional neural network algorithm of TENSORFLOW deep learning;
the weeding mechanism comprises a speed reducer, a DELTA mechanical arm (4), a mechanical arm motor, a mechanical driving arm (42), a ball hinge (43), a mechanical driven arm (44), a weeding motor (45), a movable platform (46) and a weeding cutter (47); DELTA arm (4) are installed under frame (1), and the arm motor adopts three 57 step motor (41), and each other becomes 120, motor output torque 2.2N/M, and mechanical master arm (42) link to each other with the reduction gear axle, and mechanical master arm (42) and mechanical slave arm (44) all link to each other through ball hinge (43) with moving platform (46), and install on moving platform (46) weeding motor (45), and weeding cutter (47) are installed on weeding motor's (45) epaxial.
2. The novel machine vision and convolutional neural network algorithm-based peucedanum machine of claim 1, wherein: the STM32 motion controller adopts STM32F407VET6 microcontroller inside to be provided with STM32F407VET6 singlechip, receives the weeds position information of industrial computer transmission, through HBS57 closed loop driver and STM32 motion controller electric connection.
3. The novel machine vision and convolutional neural network algorithm-based peucedanum machine of claim 1, wherein: the four wheels (12) are respectively provided with a set of independent vibration damper (11), the driving motors of the wheels (12) adopt brushless direct current motors (13), an encoder is used for forming a closed loop feedback system, the rotating speed of the motor is monitored in real time to control the speed of the vehicle so as to cooperate with a weeding mechanism to complete weeding, the four wheels (12) are respectively provided with one brushless direct current motor (13), the differential speed and the differential reverse direction can be applied during steering, and the differential reverse direction can realize pivot steering.
4. The novel machine vision and convolutional neural network algorithm-based peucedanum machine of claim 1, wherein: the 57 stepping motor (41) is connected with the mechanical driving arm (42) through a speed reducer with the speed reduction ratio of 10:1 to improve output torque, the mechanical driving arm (42) is made of aluminum alloy, the mechanical driven arm (44) is made of a high-strength carbon fiber pipe, and the mechanical driving arm (42) is connected with the mechanical driven arm (44) through a ball hinge (43).
5. The novel machine vision and convolutional neural network algorithm-based peucedanum machine of claim 1, wherein: the weeding cutter (47) adopts a self-sharpening blade, is arranged on a movable platform (46) of the DELTA mechanical arm (4), and can rotate at high speed to cut off weeds along with the change of the position of the DELTA mechanical arm (4) at any time.
6. The novel machine vision and convolutional neural network algorithm-based peucedanum machine of claim 1, wherein: the machine vision of the image processing target detection system adopts a high-definition industrial camera (3) to acquire real-time image information, and a TENSOFLOW frame convolutional neural network YoloV4 target detection algorithm is used for identifying weeds and transmitting results to an actuating mechanism to realize a weeding function;
the convolutional neural network adopts a target detection convolutional network algorithm of feature extraction, the trunk feature extraction network adopts CSPDACKNET53, the feature extraction capability is enhanced by using SPP and PANET structures, the MOSAIC data is enhanced, and CIOU is used as LOSS regression and MISH activation functions; the input image is a 416 × 416 × 3RGB three-channel color picture, and pixel coordinate values, scores, and classifications of the detection target are output.
7. The novel machine-vision and convolutional neural network algorithm-based peucedanum machine of claim 6, wherein: the image processing target detection system detects the center coordinates of the detected peucedanum of the drill by adopting a target detection deep learning algorithm and fits a lane line by adopting a least square method; in the moving process of the robot, if the number of missing seedlings on one side is small, path fitting is not influenced; when the number of the missing seedlings on one side is large, the path can be planned according to the plants on the other side, so that the autonomous navigation of the weeding machine is realized, and the weeding machine walks among the planted peucedanum praeruptorum plants.
8. The novel machine-vision and convolutional neural network algorithm-based peucedanum machine of claim 7, wherein: the image processing target detection system transmits images acquired by the industrial camera (3) to an industrial computer in real time, then the trained model industrial computer is used for fitting an ideal path, and then the industrial computer sends the planned path to an STM32 motion controller in a USB-to-serial port mode;
the STM32 motion controller sends a pulse modulation (PWM) pulse signal to a signal control end of the ZM6615 direct-current brushless driver, and the signal control end is used for driving two front wheels of the whiteflower hogfennel weeding machine, and the whiteflower hogfennel weeding machine realizes autonomous motion; when the current weed killer deviates, the rotating speed of the brushless direct current motors (13) on the two sides of the current weed killer is adjusted to adjust the moving direction, and meanwhile, the Hall sensor converts the rotating speed of the brushless direct current motors (13) into pulse signals and transmits the pulse signals to the STM32 motion controller for accurate steering control; the brushless direct current motor (13) and the STM32 motion controller are connected through a ZM6615 direct current brushless driver in the control process, and the STM32 motion controller outputs PWM pulse signals to control the speed and walk in the field through a motor driving circuit, so that the forward, turning and speed regulation are realized.
9. The novel machine vision and convolutional neural network algorithm-based peucedanum machine of claim 1, wherein: the image processing target detection system detects the position of weeds by adopting a deep learning algorithm, and performs sampling stacking through multiple convolution, pooling and stacking of residual edges; after deep network feature extraction, selecting a prediction box with the maximum score by adopting non-maximum inhibition, and finally outputting the position coordinate of the weeds; and calibrating the specific relation between the actual position coordinates and the pixel coordinates through the calibration plate, and determining the actual position of the coordinate point in the graph.
10. The novel machine-vision and convolutional neural network algorithm-based peucedanum machine of claim 9, wherein: the image processing target detection system transmits the image acquired by the industrial camera (3) to an industrial computer in real time, and then the trained model is used for detecting the target of the industrial computer to mark weeds and score conditions of a prediction frame;
calculating the actual spatial coordinates of the weeds through calibration, and then sending the weed position information to a controller-STM 32 motion controller of a DELTA mechanical arm (4) by an industrial computer in a USB-to-TTL serial port mode; the STM32 motion controller calculates and obtains the joint node for controlling the DELTA mechanical arm (4) by using the rotating angle of the 57 stepping motor (41) through the fed weed position information to realize the displacement motion of the weeding mechanism. The MOTION of the DELTA mechanical arm (4) adopts a space interpolation algorithm to realize a space arbitrary moving track, and adopts an ADEPT MOTION portal MOTION path for avoiding mistakenly removing the peucedanum and solving the problem of mechanism vibration during cross-region weeding.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010728415.6A CN111837593A (en) | 2020-07-24 | 2020-07-24 | Novel peucedanum praeruptorum weeding machine based on machine vision and convolutional neural network algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010728415.6A CN111837593A (en) | 2020-07-24 | 2020-07-24 | Novel peucedanum praeruptorum weeding machine based on machine vision and convolutional neural network algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111837593A true CN111837593A (en) | 2020-10-30 |
Family
ID=72946991
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010728415.6A Pending CN111837593A (en) | 2020-07-24 | 2020-07-24 | Novel peucedanum praeruptorum weeding machine based on machine vision and convolutional neural network algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111837593A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112544204A (en) * | 2020-12-03 | 2021-03-26 | 武汉理工大学 | Weeding robot based on RISC-V embedded processor and control method |
| CN112868444A (en) * | 2020-12-28 | 2021-06-01 | 山东农业大学 | Garlic bud seedling membrane breaking robot and using method thereof |
| CN112931136A (en) * | 2021-03-30 | 2021-06-11 | 青岛理工大学 | Full-automatic tree planter and tree planting method |
| CN113381667A (en) * | 2021-06-25 | 2021-09-10 | 哈尔滨工业大学 | Seedling searching walking system and method based on ROS and image processing |
| CN113498639A (en) * | 2021-09-13 | 2021-10-15 | 农业农村部南京农业机械化研究所 | Weeding device and weeding method |
| CN113954986A (en) * | 2021-11-09 | 2022-01-21 | 昆明理工大学 | Tobacco leaf auxiliary material conveying trolley based on vision |
| CN113994885A (en) * | 2021-11-17 | 2022-02-01 | 安徽荃银高科种业股份有限公司 | Method for seed production, cost saving, impurity removal, castration and purity preservation of hybrid corn |
| CN114169500A (en) * | 2021-11-30 | 2022-03-11 | 电子科技大学 | Neural network model processing method based on small-scale electromagnetic data |
| CN114617114A (en) * | 2022-04-19 | 2022-06-14 | 浙江理工大学 | Solar weeding robot and control method thereof |
| CN114700969A (en) * | 2022-04-19 | 2022-07-05 | 浙江理工大学 | Weeding robot based on XY I-shaped sliding table parallel flexible manipulator and control method |
| CN114700927A (en) * | 2022-04-19 | 2022-07-05 | 浙江理工大学 | Combined weeding robot with parallel mechanical arms and flexible mechanical arm and control method |
| CN115119546A (en) * | 2022-07-19 | 2022-09-30 | 华南农业大学 | Rice field self-adaptive flexible mechanical weeding machine based on visual-tactile fusion perception |
| WO2023050783A1 (en) * | 2021-09-29 | 2023-04-06 | 农业农村部南京农业机械化研究所 | Weeding robot and method and apparatus for planning weeding path thereof, and medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU738825B3 (en) * | 2000-09-29 | 2001-09-27 | Gerard De Thomasis | Mower |
| US20150245559A1 (en) * | 2012-01-11 | 2015-09-03 | Jian Yang | Lawn mower with an automatic clutch |
| CN105230224A (en) * | 2015-11-06 | 2016-01-13 | 嘉兴职业技术学院 | Intelligent weeding machine and weed removing method thereof |
| CN111165092A (en) * | 2020-02-28 | 2020-05-19 | 山东农业大学 | Parallel arm weeding robot and weed weeding method |
| CN210841314U (en) * | 2019-07-17 | 2020-06-26 | 安徽农业大学 | Novel automatic whiteflower hogfennel root weeder |
-
2020
- 2020-07-24 CN CN202010728415.6A patent/CN111837593A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU738825B3 (en) * | 2000-09-29 | 2001-09-27 | Gerard De Thomasis | Mower |
| US20150245559A1 (en) * | 2012-01-11 | 2015-09-03 | Jian Yang | Lawn mower with an automatic clutch |
| CN105230224A (en) * | 2015-11-06 | 2016-01-13 | 嘉兴职业技术学院 | Intelligent weeding machine and weed removing method thereof |
| CN210841314U (en) * | 2019-07-17 | 2020-06-26 | 安徽农业大学 | Novel automatic whiteflower hogfennel root weeder |
| CN111165092A (en) * | 2020-02-28 | 2020-05-19 | 山东农业大学 | Parallel arm weeding robot and weed weeding method |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112544204A (en) * | 2020-12-03 | 2021-03-26 | 武汉理工大学 | Weeding robot based on RISC-V embedded processor and control method |
| CN112868444A (en) * | 2020-12-28 | 2021-06-01 | 山东农业大学 | Garlic bud seedling membrane breaking robot and using method thereof |
| CN112931136A (en) * | 2021-03-30 | 2021-06-11 | 青岛理工大学 | Full-automatic tree planter and tree planting method |
| CN113381667A (en) * | 2021-06-25 | 2021-09-10 | 哈尔滨工业大学 | Seedling searching walking system and method based on ROS and image processing |
| CN113498639A (en) * | 2021-09-13 | 2021-10-15 | 农业农村部南京农业机械化研究所 | Weeding device and weeding method |
| CN113498639B (en) * | 2021-09-13 | 2021-12-10 | 农业农村部南京农业机械化研究所 | Weeding device and weeding method |
| WO2023050783A1 (en) * | 2021-09-29 | 2023-04-06 | 农业农村部南京农业机械化研究所 | Weeding robot and method and apparatus for planning weeding path thereof, and medium |
| CN113954986A (en) * | 2021-11-09 | 2022-01-21 | 昆明理工大学 | Tobacco leaf auxiliary material conveying trolley based on vision |
| CN113994885A (en) * | 2021-11-17 | 2022-02-01 | 安徽荃银高科种业股份有限公司 | Method for seed production, cost saving, impurity removal, castration and purity preservation of hybrid corn |
| CN114169500A (en) * | 2021-11-30 | 2022-03-11 | 电子科技大学 | Neural network model processing method based on small-scale electromagnetic data |
| CN114169500B (en) * | 2021-11-30 | 2023-04-18 | 电子科技大学 | Neural network model processing method based on small-scale electromagnetic data |
| CN114700969A (en) * | 2022-04-19 | 2022-07-05 | 浙江理工大学 | Weeding robot based on XY I-shaped sliding table parallel flexible manipulator and control method |
| CN114700927A (en) * | 2022-04-19 | 2022-07-05 | 浙江理工大学 | Combined weeding robot with parallel mechanical arms and flexible mechanical arm and control method |
| CN114617114A (en) * | 2022-04-19 | 2022-06-14 | 浙江理工大学 | Solar weeding robot and control method thereof |
| CN115119546A (en) * | 2022-07-19 | 2022-09-30 | 华南农业大学 | Rice field self-adaptive flexible mechanical weeding machine based on visual-tactile fusion perception |
| CN115119546B (en) * | 2022-07-19 | 2024-03-12 | 华南农业大学 | Paddy field self-adaptive flexible mechanical weeding machine based on visual-tactile fusion perception |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111837593A (en) | Novel peucedanum praeruptorum weeding machine based on machine vision and convolutional neural network algorithm | |
| Torii | Research in autonomous agriculture vehicles in Japan | |
| Tillett et al. | A robotic system for plant-scale husbandry | |
| Keicher et al. | Automatic guidance for agricultural vehicles in Europe | |
| Nagasaka et al. | An autonomous rice transplanter guided by global positioning system and inertial measurement unit | |
| CN101412217B (en) | Autonomous field robot | |
| Noguchi | Agricultural vehicle robot | |
| KR20190031391A (en) | Intelligent agricultural robot system | |
| Yang et al. | Development and application of a wheel-type robot tractor | |
| KR20200095225A (en) | An weed removal robot using image processor | |
| Pota et al. | Agricultural robotics: A streamlined approach to realization of autonomous farming | |
| Emmi et al. | Mobile robotics in arable lands: Current state and future trends | |
| Berenstein et al. | Robustly adjusting indoor drip irrigation emitters with the toyota hsr robot | |
| Möller | Computer vision–a versatile technology in automation of agricultural machinery | |
| CN108508889B (en) | Small intelligent weeding and soil loosening device and control method | |
| Lai et al. | A human-robot cooperative vehicle for tea plucking | |
| Wang et al. | Smart agricultural in-field service robot: from toy to tool | |
| CN112868623B (en) | Plant protection machinery navigation control method based on multiple sensors | |
| CN113467453A (en) | Inspection robot and method for controlling inspection robot to run based on fuzzy PID | |
| Meti et al. | IoT and Solar Energy Based Multipurpose Agricultural Robot for Smart Farming | |
| US20220124959A1 (en) | Autonomous folding farm implement and method | |
| Nagasaka et al. | An autonomous field watching-dog robot for information collection in agricultural fields | |
| Hameed et al. | Task and motion planning for selective weed conrol using a team of autonomous vehicles | |
| Mengoli | A robotic platform for precision agriculture and applications | |
| Zhang et al. | Development of a Human-driven tractor following a Robot System |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20201030 |