Weeding robot system based on cloud platform remote control and control method thereof

Abstract

The invention provides a weeding robot system based on cloud platform remote control and a control method thereof, and belongs to the technical field of robot weeding, and a weeding robot comprises a detection device, a laser emission device, a multi-module processing control device, a maneuvering device and a power device. The multi-module processing device comprises a GPS positioning module, an information receiving and processing module, a three-dimensional topographic data information capturing module, a control optimization operation module, a motion real-time simulation module and a command execution module. The control method comprises the steps that the weeding robot system comprises a real-time positioning subsystem, an area planning subsystem, an intelligent operation subsystem and a path planning subsystem, the multi-module processing control device controls the weeding robots to conduct path planning through the subsystems, and the multiple weeding robots are trained through a DQN and a new self-defined algorithm P; the accuracy of identifying the captured target data by the weeding robot is improved, and identification of more different types of data is facilitated, so that the adaptive capacity to environment change is enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of robot weeding, and in particular relates to a weeding robot system remotely controlled based on a cloud platform and a control method thereof. Background Art

[0002] When weeding crops, the efficiency of weeding and the damage to crops during the weeding process will affect the quality of the weeding work. The weeding device is an important factor in improving the quality of the weeding work.

[0003] The present invention discloses a multifunctional weeding robot system and a control method based on remote and precise control of a cloud platform. The technical key points are: including a high-precision new infrared sensing detection device, an improved laser emitting device, a new multi-module processing and control device, a new energy power device and a maneuvering device. The high-precision new infrared sensing and detection device can use the reflectivity difference between crops and weeds in the NIR band to identify and distinguish weeds, thereby improving the accuracy of identification and reducing the probability of damage to crops due to misjudgment during laser weeding. The external protection device in the improved laser emitting device can rotate in all directions in a plane, driving the laser emitter to rotate in all directions in a plane, thereby improving weeding efficiency. The new multi-module processing and control device can remotely locate the robot in real time and perform remote and precise control on the cloud platform. At the same time, the installed training set can help the robot plan its route.

[0004] With respect to the relevant content mentioned above, the following technical defects were found: the multifunctional weeding robot in the existing technology uses visual communication equipment to distinguish and identify weeds, which has low accuracy and is prone to misjudgment. It also has a single movement direction and a small control range, which may cause the following shortcomings:

[0005] 1. The central processing unit has a high probability of misjudging weeds and crops, which can easily damage crops and greatly reduce the quality of weed control.

[0006] 2. The controlled distance is small, resulting in a small working range, increasing workload, work costs and energy consumption, while reducing work efficiency. Summary of the Invention

[0007] The present invention aims to provide a remotely controlled weeding robot system based on a cloud platform, comprising a detection device for identifying terrain and weeds, a laser emitting device for weeding, a multi-module processing control device for controlling the operation of the weeding robot, a motorized device for carrying each device and providing mobility for the weeding robot, and a power device for providing driving force for the motorized device.

[0008] The multi-module processing and control device includes: a GPS positioning module, an information receiving and processing module, a three-dimensional terrain data information capture module, a control optimization operation module, a motion real-time simulation module and a command execution module that are mutually signal-connected;

[0009] The GPS positioning module locates in real time and sends the location information to the information receiving and processing module;

[0010] The information receiving and processing module is connected to the cloud platform signal, receives and processes the data sent by the detection device and each module, and controls the weeding robot to execute instructions through the command execution module;

[0011] The three-dimensional terrain data capture module is connected to the detection device to capture the three-dimensional spatial terrain data in real time and transmit it to the control optimization operation module for planning the driving path;

[0012] The real-time motion simulation module is used to coordinate the simultaneous operation of each robot.

[0013] Furthermore, the maneuverable device includes a carrying device and a tire motion system installed at the bottom of the carrying device. The carrying device is used to carry various devices, and the tire motion system is used to realize the movement of the weeding robot.

[0014] Furthermore, the laser emitting device includes a laser emitter and an external protective shell. The external protective shell is rotatably arranged on the upper surface of the supporting device through a rotating bearing. The laser emitter is fixed to the front end of the external protective shell and rotates with the rotation of the external protective shell.

[0015] Furthermore, the power unit includes a solar photovoltaic panel located on the upper surface of the external protective shell, a battery located inside the external protective shell, a first engine that drives the tire movement system to operate, a second engine that drives the external protective shell to rotate, and a third engine that drives the multi-freedom robotic arm to rotate. The solar photovoltaic panel, the battery and the three engines are electrically connected to provide power to the three engines.

[0016] Furthermore, the detection device includes a sub-band infrared transceiver, an infrared carrying platform, a multi-stage free manipulator and a fixed seat, the fixed seat is fixed to the carrying device, the bottom end of the multi-stage free manipulator is movably fixed to the top end of the fixed seat, the top end is connected to the infrared carrying platform, and the sub-band infrared transceiver is movably fixed to the infrared carrying platform via a bearing;

[0017] The multi-level free robotic arms are movably connected to the infrared carrying platform, the fixed seat, and the robotic arms at each level through smooth bolts to achieve linkage.

[0018] A control method for a weeding robot system remotely controlled based on a cloud platform, using a weeding robot system remotely controlled based on a cloud platform, wherein: the weeding robot system includes a real-time positioning subsystem, a regional planning subsystem, an intelligent computing subsystem, and a path planning subsystem;

[0019] The multi-module processing and control device captures and filters targets through the real-time positioning subsystem, analyzes the filtered targets and divides the crop areas through the regional planning subsystem and intelligent operation subsystem, and plans the working paths of each weeding robot through the path planning subsystem, optimizes its working trajectory, and finds the optimal working route.

[0020] The information receiving and processing module is equipped with a training set. The weeding robot system trains multiple weeding robots through the DQN algorithm and algorithm P to improve the accuracy of data recognition of captured targets, enabling the weeding robots to make optimal behavior plans in response to different working environments.

[0021] Furthermore, the multi-module processing control device captures targets and filters them through the real-time positioning subsystem. Specifically, the multi-module processing control device captures the static and dynamic objects in the crop area through the three-dimensional terrain data information capture module, and filters the captured targets through the rc-low-pass filtering algorithm.

[0022] Furthermore, based on the target information transmitted by the real-time positioning subsystem, the multi-module processing and control device works in coordination with the intelligent computing subsystem and the regional planning subsystem to divide the crop area into multiple n-gonal work areas. Each n-gonal work area corresponds to a weeding robot, and ensures that each weeding robot performs weeding work in each work area at the same time.

[0023] Furthermore, the multi-module processing and control device generates multiple convex deformation simulation obstacles based on multiple n-deformation workspaces according to the captured dynamic and static target information through the path planning subsystem, draws a real-time three-dimensional spatial terrain map according to the patch function, makes judgments and predictions on spatial changes, and plans working paths for multiple robots;

[0024] When planning the work path, the path planned by each weeding robot is tested through the training set.

[0025] Furthermore, the algorithm P is specifically:

[0026] The static object data and dynamic object data recognized by the weeding robot are classified into the training set U, the known environmental information is defined as strong mutual data A, and the environmental information containing variable factors is defined as weak mutual data B;

[0027] Training the training set using algorithm P includes the following steps:

[0028] S1: Randomly extract strong mutual data A and weak mutual data B from the training set U to obtain data M;

[0029] S2: Divide the extracted data M into n sub-databases Sn;

[0030] S3: Compare and train the data in each sub-database Sn respectively;

[0031] S4: Move the new data Dn obtained from the preliminary training in each sub-database Sn into the data reinforcement training library L for reinforcement training;

[0032] S5: After the reinforcement training, the reinforcement mutual data A and the multi-mutual data B are coordinated to generate new data H;

[0033] S6: Move the generated new data H into the temporary database G until the training set U becomes an empty set

[0034] S7: Transfer the data in the temporary database G to the empty set, and repeat S1-S6;

[0035] In each cycle, the new data H generated in S6 is used to assist the weeding robot in selecting a better path planning and working behavior plan.

[0036] Compared with the prior art, the beneficial effects of the present invention are mainly reflected in:

[0037] 1. This invention utilizes infrared recognition, laser technology, and cloud platform remote control to accurately and efficiently distinguish and identify weeds. This reduces the probability of misjudgment during the identification process by multi-module processing and control devices, minimizes damage to crops caused by lasers during weeding, and achieves efficient weeding. Cloud platform remote control technology increases the robot's controlled distance and working range, thereby reducing workload and costs.

[0038] 2. The weeding robot trained with the DQN algorithm in this invention can select the optimal behavior based on its current working state. The use of the newly defined independent algorithm P greatly improves the accuracy of the weeding robot in identifying the captured target data and can help the robot recognize more different types of data, thereby enhancing its ability to adapt to environmental changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a schematic diagram of the structure of the weeding robot remotely controlled based on the cloud platform of the present invention.

[0040] Figure 2 for Figure 1 Schematic diagram of the detection device structure.

[0041] Figure 3 for Figure 1 Schematic diagram of the laser emitting device structure.

[0042] Figure 4 for Figure 1 Schematic diagram of the structure of the medium maneuvering device.

[0043] Figure 5 Schematic diagram of the connection relationship between the subsystems in the weeding robot system of the present invention.

[0044] Figure 6 Schematic diagram of the training set training process in the control method of the present invention.

[0045] Figure 7 This is a three-dimensional spatial topographic map drawn using training set data in the present invention.

[0046] Figure 8 This is a two-dimensional topographic map drawn using the training set data in the present invention.

[0047] Figure 9 The path planning diagram is drawn using MATLAB based on the path planning subsystem and training set data in the present invention.

[0048] Figure 10 This is a pseudo code diagram for path planning in the present invention.

[0049] Among them, 1. Detection device; 101. Band-splitting infrared transceiver; 102. Bearing; 103. Free robotic arm; 104. Infrared carrying platform; 2. Laser emitting device; 201. Laser emitter; 202. External protective shell; 3. Multi-module processing and control device; 4. Power unit; 5. Maneuvering device; 501. Carrying device; 502. Tire movement system. DETAILED DESCRIPTION

[0050] The following will be described in more detail with reference to schematic diagrams of a cloud platform-based remotely controlled weeding robot system and its control method of the present invention, which shows a preferred embodiment of the present invention. It should be understood that those skilled in the art can modify the present invention described herein while still achieving the beneficial effects of the present invention. Therefore, the following description should be understood as a general knowledge for those skilled in the art and not as a limitation of the present invention.

[0051] like Figure 1 As shown, a weeding robot system based on remote control of a cloud platform includes a detection device, a laser emitting device, a multi-module processing control device, a maneuvering device and a power device.

[0052] Detection device 1

[0053] refer to Figure 2, including: a sub-band infrared transceiver 101, an infrared carrying platform 104, a free robotic arm 103 and a fixed seat, the fixed seat is fixed to the carrying device 501, the bottom end of the free robotic arm 103 is movably fixed to the top of the fixed seat, and the top is connected to the infrared carrying platform 104, the outer surface of the sub-band infrared transceiver 101 is made of ASTMA36 steel, and is movably fixed to the infrared carrying platform 104 through a bearing 102 to achieve 360° plane rotation; the free robotic arm 103 and the infrared carrying platform 104 form a multi-level free robotic arm, and the free robotic arm 103 and the infrared carrying platform 104 are movably connected by a smooth bolt to achieve the linkage of the multi-level free robotic arm.

[0054] Laser launch device 2

[0055] refer to Figure 3 , including a laser emitter 201 and an external protective shell 202. The external protective shell 202 is made of AISI316 stainless steel and is rotatably arranged on the upper surface of the carrying device 501 through a rotary bearing. The laser emitter 201 uses organic glass and is fixed to the front end of the external protective shell 202. It rotates with the rotation of the external protective shell 202 to achieve laser weeding.

[0056] Multi-module processing control device 3

[0057] It includes a real-time positioning module, an information receiving and processing module, a three-dimensional terrain data information capture module, a control optimization operation module, a motion real-time simulation module and a command execution module.

[0058] The real-time positioning module includes a GPS positioning device, which is connected to the information receiving and processing module signal and is used to send the position of the weeding robot in real time.

[0059] The information receiving and processing module is the central processing unit, responsible for receiving and processing data transmitted by each module. It then controls the weeding robot to execute instructions through the command execution module. An internally installed training set is used for subsequent training using the DQN algorithm and the newly defined algorithm P to help the weeding robot plan its path.

[0060] The three-dimensional terrain data information capture module signal is connected to the band-splitting infrared transceiver 101 to capture the collected three-dimensional spatial terrain data and transmit it to the control optimization operation module for planning the driving path, reducing the probability of overlapping driving paths of multiple weeding robots and preventing collisions.

[0061] The real-time motion simulation module is used to simulate the driving path to obtain the optimal control parameters, further optimize the control of multiple weeding robots, and coordinate the simultaneous operation of each weeding robot.

[0062] Maneuvering device 5

[0063] refer to Figure 4 , including a carrying device 501 and a tire motion system 502 installed at the bottom of the carrying device 501. The carrying device 501 is made of AISI 316 stainless steel and is used to carry various devices. The tire motion system 502 uses lightweight motion bearings and lightweight tires. The outer tires are made of natural rubber and have complex patterns, which reduce the energy consumption used for additional work of the wheels during travel and improve energy utilization. It is driven by a first engine to achieve the movement of the weeding robot.

[0064] Power Unit 4

[0065] It includes a solar photovoltaic panel located on the upper surface of an external protective shell, a high-storage battery located inside the external protective shell, a first engine that drives the tire movement system to operate, a second engine that drives the external protective shell to rotate, and a third engine that drives the multi-freedom robotic arm to rotate. The solar photovoltaic panel, the battery and the three engines are electrically connected to provide power to the three engines.

[0066] A control method for a weeding robot system based on remote control of a cloud platform, such as Figure 5 As shown, the weeding robot system includes the following four subsystems.

[0067] Real-time positioning subsystem:

[0068] It can capture and filter multiple random capture targets in each crop area, including static and dynamic targets, and upload the data to the intelligent computing subsystem to analyze each captured target;

[0069] Regional planning subsystem:

[0070] It can analyze the captured targets and divide the entire crop area through the intelligent computing subsystem in the central control system to ensure that each weeding robot works in its own crop area;

[0071] Intelligent computing subsystem:

[0072] It can analyze the target capture information transmitted by the real-time positioning subsystem and the regional planning subsystem, and divide the crop areas;

[0073] Path planning subsystem:

[0074] It can plan the robot's working path, optimize its working trajectory, and reduce the probability of collision between robots when working. At the same time, it can find the optimal working route and improve weeding efficiency.

[0075] The three-dimensional terrain data information capture module can capture static and dynamic objects in the crop area through the real-time positioning subsystem and transmit the captured targets to the light wave filtering module;

[0076] The light wave filtering module uses the rc-low-pass filtering algorithm to filter the captured information. The following variables are defined below:

[0077] Yn: the output value of the captured target;

[0078] Ym: the output value of the last captured target;

[0079] Xn: the sampling value of the captured target;

[0080] T: Sampling period for capturing targets;

[0081] f: cutoff frequency;

[0082] a: filter coefficient, and a∈[0,1]

[0083] Filter the captured target information.

[0084] Yn=Ym+a(Xn-Ym)

[0085]

[0086] The rc-low-pass filter can attenuate high-frequency bands except the cutoff frequency, thereby making motion state predictions based on the motion history data of dynamic and static captured targets, that is, making judgments and predictions on the robot's next working path, and correcting the newly captured target data to obtain a more realistic three-dimensional space coordinate sequence.

[0087] The real-time positioning subsystem can also be used to track and monitor the real-time working status and working position of multiple weeding robots. At the same time, it can transmit information and identify each other among multiple robots, ensuring the information connection of the weeding robot group and effectively improving work efficiency.

[0088] The regional planning subsystem can analyze the target capture information transmitted by the real-time positioning subsystem, divide the crop area into sub-unit work areas, and evenly distribute the robots in each sub-unit work area to improve work efficiency and maximize the utilization of work resources;

[0089] The intelligent computing subsystem works in conjunction with the regional planning subsystem, using information transmitted by the real-time positioning subsystem to cooperate with the regional planning subsystem to divide the crop area into multiple n-gonal work areas, ensuring that each robot can perform weeding work in its own work area at the same time.

[0090] It should be understood that in each work zone, in order to prevent the robots in each zone from exceeding their own work zone, the intelligent computing subsystem will plan the boundaries of each zone.

[0091]

[0092] Among them, i is the number of the convex polygon simulation obstacle in the workspace, and temp is the dynamic environment parameter captured in real time.

[0093] The path planning subsystem can simulate and generate multiple convex polygonal obstacles based on the captured dynamic and static target information and the polygonal workspace planned by the intelligent computing subsystem, draw a real-time three-dimensional spatial terrain map based on the patch function, make judgments and predictions on spatial changes, and plan the robot's work path.

[0094] patch('XData',X,'YData',Y,'ZData',Z)

[0095] Among them, 'XData' is the longitude coordinate of the real-time capture input, X is the longitude coordinate of the convex polygon simulation obstacle, 'YData' is the latitude coordinate of the real-time capture input, Y is the latitude coordinate of the convex polygon simulation obstacle, 'ZData' is the height coordinate of the real-time capture input, Z is the height coordinate of the convex polygon simulation obstacle.

[0096] In the path planning subsystem, it is necessary to add a training set to each robot to test the path planning. For a certain motion state u of the robot, it is converted into data and input into the external environment to obtain the new motion state data S and r.

[0097] y=r+γ*maxQ(S,u;w)

[0098] L=0.5[yQ(S,u;w)]^2

[0099] Among them, y is the target value output by the robot model learning, r is the immediate action benefit value, γ is the discount factor, which takes the value [0, 1], Q is the cumulative expected value of learning, (S, a; w) is the combined parameter of state S and action u under parameter w, and L is the optimization value of learning loss.

[0100] Now define a new algorithm P

[0101] The data of static and dynamic objects that the robot can recognize are classified into the training set U. The known environmental information (including the known robot's own motion state and properties, robot working status, etc.) is defined as strong mutual data A, and the environmental information containing variable factors (including weather changes, soil properties, obstacles, and crop growth conditions, etc.) is defined as weak mutual data B, where the change parameter is defined as d.

[0102] d = rand('T, 'W, 'G) mod M

[0103] Among them, 'T is a dynamic environmental variable related to meteorological changes and crop growth status, 'W is a weight variable of weak interaction data B related to the importance of soil properties, and 'G is a global variable of strong interaction A related to robot properties.

[0104] The principle of algorithm P is as follows: Figure 6 As shown:

[0105] Strong mutual data A and weak mutual data B are randomly extracted from the entire data set U, and the obtained data M is divided into n sub-databases Sn. Each database compares and trains the data in a single database, and each group of new data Dn obtained from the preliminary training is moved into the data reinforcement training database L. After the reinforcement training, the new data H generated by the coordination of strong mutual data A and weak mutual data B is moved into the temporary database G until U becomes an empty set, and the data in G is transferred to U, and the cycle is repeated X times.

[0106] During the above training process:

[0107] Strong mutual data A: usually refers to subjective environmental factors related to the properties of the weeding robot itself, including relatively subjective factors such as the weeding robot's own properties and the working conditions it can withstand;

[0108] Weakly interactive data B: usually refers to objective environmental factors that are not strongly correlated with the weeding robot itself and include external variables, including climate change and soil properties (i.e., humidity, shape, fertility, etc.);

[0109] Initial reinforcement data Dn: preliminary training data obtained by randomly mixing data A and B and performing comparative training, providing resources for the generation of reinforcement training data;

[0110] Reinforced training data H: This is new data obtained by integrating and retraining the preliminary training data of each sub-unit in the reinforced training library L. This data can help the robot make better choices for subsequent path planning and work behavior, improve the robot's intelligence level, and enhance the robot's ability to self-plan its work trajectory.

[0111] H=compareΣ('D1,'D2,...,'Dn)

[0112] It should also be understood that in the training process defined above, as data is continuously extracted from the entire data set U, U will gradually become an empty set. At this time, the data temporarily stored in the temporary database G will transfer the enhanced new data to the overall database U, ensuring the cycle of the training process, while also continuously strengthening the optimization of the training data.

[0113] In summary, a weeding robot trained with the DQN algorithm can select the optimal behavior based on its current working state. The newly defined algorithm P significantly improves the accuracy of the weeding robot's recognition of captured target data and helps it identify a wider range of data types, thereby enhancing its ability to adapt to environmental changes. This training process enhances the robot's intelligent motion capabilities, enabling the weeding robot team to quickly and efficiently make decisions and plan optimal behaviors in diverse working environments.

[0114] The multifunctional weeding robot system with remote and precise control based on a cloud platform and its control method includes an infrared sensor detector and a laser emitter, which include: an infrared sensor detection device capable of accurately identifying weeds based on the differences in the NIR bands of crops and weeds, and transmitting the data to a central processing unit for data processing;

[0115] The laser emitting device can receive instructions from the central processing unit to the part, rotate to a designated position, and emit laser to remove weeds.

[0116] The infrared transceiver contained in the infrared sensing detection device can rotate autonomously and emit infrared rays when the robot is working, and then receive the NIR band reflected by the plants and transmit it to the central processor.

[0117] C:ni+1=(v*ni+g)modM

[0118] J:mj+1=(c*mj+d)modN

[0119] F: pk+=(e*pk+q)modR

[0120] Among them, C is the generated horizontal scanning sequence function, J is the function for adjusting the vertical direction of the infrared transceiver, F is the function for adjusting the transmission power and receiving sensitivity of the infrared transceiver, ni is the capture input parameter of the current scanning position, v and g are linear transformation parameters, mj is the vertical position parameter of the current state, c and d are vertical position control parameters, pk is the current infrared transceiver device state parameter, and e and q are the device dynamic adjustment coefficients

[0121] It should be understood that the above-mentioned operation method can realize the autonomous rotation of the infrared transceiver, reduce the omission rate of the robot's sensing and detection of weeds in the working area, and improve the robot's efficiency in identifying weeds.

[0122] In the present invention, the laser emitting device can take actions according to the instructions issued by the central controller, including the rotation of the external protection device and the emission of the laser.

[0123] O=(NIR-r) / (NIR+r)

[0124] A=(rcosαcosθ-rsinαsinθ)

[0125] B=∑(rsinθcosθ+rcosαsinθ)

[0126] Wherein, NIR = 780 ± 10 nm;

[0127] It should be understood that the central processor receives the infrared information transmitted by the infrared transceiver, filters the infrared band with an infrared wavelength of λ≈780nm (the infrared wavelength reflected by weeds), and issues an instruction to drive the external protection device to rotate to a specified position, and issues an instruction to make the laser emitter emit laser to remove weeds.

[0128] The infrared band reflected by weeds is the near-infrared spectrum, while the infrared band reflected by most crops is usually the mid-infrared spectrum; due to the difference in the infrared wavelengths reflected by weeds and crops, the infrared sensing detection device and central processor can accurately identify weeds.

[0129] refer to Figure 7-10 The robot path planning of the multifunctional weeding robot system and its control method based on remote and precise control of the cloud platform of the present invention is based on the following principles:

[0130] Based on the data obtained from the terrain data acquisition module and the regional planning subsystem, three-dimensional space simulation obstacles are drawn, the starting point and end point are locked, and the training data obtained from the training set are used to draw three-dimensional space and two-dimensional plane path planning simulation diagrams. This reduces the probability of dynamic mutual collisions and individual collisions of the weeding robot group, and can plan the optimal working motion trajectory within its own working area, greatly reducing work costs and improving work efficiency.

[0131] Using the control method of the present invention, the pseudo code of the path planning is as follows: Figure 10 shown.

[0132] Figure 5 and 6The flowcharts show the possible architectures, functions and operations of the systems, methods, etc. according to various embodiments of the present invention. Based on this, each box in the flowchart can represent a database, a portion of which contains one or more executable instructions for implementing the specified logical functions. In some alternative implementations, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, as well as the combination of boxes in the flowchart, can be implemented by a dedicated hardware- or software-based system that performs the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions. It is well known to those skilled in the art that implementation by hardware, implementation by software, and implementation by a combination of software and hardware are all equivalent.

[0133] The above description is merely a preferred embodiment of the present invention and does not limit the present invention in any way. Any person skilled in the art who, without departing from the scope of the present invention, makes any equivalent substitution, modification, or other changes to the technical solution and technical content disclosed in the present invention shall be deemed to be within the scope of the present invention and still fall within the scope of protection of the present invention.

Claims

1. A weeding robot system remotely controlled by a cloud platform, characterized in that: It includes a detection device for identifying terrain and weeds, a laser emitting device for weeding, a multi-module processing control device for controlling the operation of the weeding robot, a maneuvering device for carrying each device and providing a mobile function for the weeding robot, and a power device for providing driving force for the maneuvering device; The multi-module processing and control device includes: a GPS positioning module, an information receiving and processing module, a three-dimensional terrain data information capturing module, a control optimization operation module, a motion real-time simulation module and a command execution module that are mutually signal-connected; The GPS positioning module locates in real time and sends the position information to the information receiving and processing module; The information receiving and processing module is connected to the cloud platform signal, receives and processes the data sent by the detection device and each module, and controls the weeding robot to execute instructions through the command execution module; The three-dimensional terrain data capture module is connected to the detection device to capture three-dimensional spatial terrain data in real time and transmit it to the control optimization operation module for planning the driving path; The real-time motion simulation module is used to coordinate the robots to run simultaneously.

2. The weeding robot system based on cloud platform remote control according to claim 1, characterized in that: The maneuvering device includes a carrying device and a tire motion system installed at the bottom of the carrying device. The carrying device is used to carry various devices, and the tire motion system is used to realize the movement of the weeding robot.

3. The weeding robot system based on cloud platform remote control according to claim 2, characterized in that: The laser emitting device includes a laser emitter and an external protective shell. The external protective shell is rotatably arranged on the upper surface of the carrying device through a rotary bearing. The laser emitter is fixed to the front end of the external protective shell and rotates with the rotation of the external protective shell.

4. The cloud platform-based remote-controlled weeding robot system according to claim 3, characterized in that: The power device includes a solar photovoltaic panel located on the upper surface of the external protective shell, a battery located inside the external protective shell, a first engine that drives the tire movement system to operate, a second engine that drives the external protective shell to rotate, and a third engine that drives the multi-freedom robotic arm to rotate. The solar photovoltaic panel, the battery and the three engines are electrically connected to provide power to the three engines.

5. The weeding robot system based on cloud platform remote control according to claim 4 is characterized in that: The detection device includes a sub-band infrared transceiver, an infrared carrying platform, a free mechanical arm and a fixed seat, wherein the fixed seat is fixed to the carrying device, the bottom end of the free mechanical arm is movably fixed to the top end of the fixed seat, and the top end is connected to the infrared carrying platform, and the sub-band infrared transceiver is movably fixed to the infrared carrying platform via a bearing; The free robotic arm is movably connected to the infrared carrying platform, the fixing seat and the robotic arms at all levels through smooth bolts to achieve linkage.

6. A control method for a cloud-based remote-controlled weeding robot system, using the cloud-based remote-controlled weeding robot system according to any one of claims 1 to 5, characterized in that: The weeding robot system includes a real-time positioning subsystem, an area planning subsystem, an intelligent operation subsystem, and a path planning subsystem; The multi-module processing and control device captures and filters targets through a real-time positioning subsystem, analyzes the filtered targets and divides crop areas through a regional planning subsystem and an intelligent computing subsystem, and plans the working paths of each weeding robot through a path planning subsystem, optimizes its working trajectory, and finds the optimal working route. The information receiving and processing module is provided with a training set. The multifunctional weeding robot system trains the multiple weeding robots through the DQN algorithm and algorithm P to improve the accuracy of data recognition of captured targets, so that the weeding robots can make optimal behavior plans in response to different working environments.

7. The control method of the weeding robot system based on cloud platform remote control according to claim 6, characterized in that: The multi-module processing control device captures targets and filters them through a real-time positioning subsystem. Specifically, the multi-module processing control device captures targets of static and dynamic objects in the crop area through a three-dimensional terrain data information capture module, and filters the captured targets through an rc-low-pass filtering algorithm.

8. The control method of the weeding robot system based on cloud platform remote control according to claim 7, characterized in that: The multi-module processing and control device divides the crop area into multiple n-gonal work areas based on the target information transmitted by the real-time positioning subsystem, through the coordinated work of the intelligent operation subsystem and the regional planning subsystem. Each n-gonal work area corresponds to a weeding robot, and ensures that each weeding robot performs weeding work in each work area at the same time.

9. The control method of the weeding robot system based on cloud platform remote control according to claim 8, characterized in that: The multi-module processing and control device generates multiple convex deformation simulation obstacles based on multiple n-deformation workspaces according to the captured dynamic and static target information through the path planning subsystem, draws a real-time three-dimensional spatial terrain map according to the patch function, makes judgments and predictions on spatial changes, and plans working paths for multiple robots; When planning the work path, the path planned by each weeding robot is tested through the training set.

10. The control method of the weeding robot system based on cloud platform remote control according to claim 9, characterized in that: The algorithm P is specifically: The static object data and dynamic object data recognized by the weeding robot are classified into the training set U, the known environmental information is defined as strong mutual data A, and the environmental information containing variable factors is defined as weak mutual data B; Training the training set using algorithm P includes the following steps: S1: Randomly extract strong mutual data A and weak mutual data B from the training set U to obtain data M; S2: Divide the extracted data M into n sub-databases Sn; S3: Compare and train the data in each sub-database Sn respectively; S4: Move the new data Dn obtained from the preliminary training in each sub-database Sn into the data reinforcement training library L for reinforcement training; S5: After the reinforcement training, the reinforcement mutual data A and the multi-mutual data B are coordinated to generate new data H; S6: Move the generated new data H into the temporary database G until the training set U becomes an empty set S7: Transfer the data in the temporary database G to the empty set, and repeat S1-S6; In each cycle, the new data H generated in S6 is used to assist the weeding robot in selecting a better path planning and working behavior plan.

Images