Intelligent inspection robot for greenhouse planting, control system and method

Abstract

The invention belongs to the technical field of intelligent robots, and particularly relates to an intelligent inspection robot for greenhouse planting and a control system and method.High-definition images are captured through a camera and transmitted to a control chip in real time to be processed, and phenotype recognition is conducted on crops in a greenhouse; the laser radar and the camera work cooperatively, real-time three-dimensional modeling is carried out on the environment in the greenhouse, and an accurate map is generated; the control chip receives original data from the camera and the laser radar, performs advanced processing such as image recognition and three-dimensional modeling, and analyzes the growth condition of crops, environmental parameters and spatial layout in the greenhouse; the driving motor drives the crawler belt to move according to an instruction of the control chip to execute a monitoring task; according to the intelligent greenhouse planting decision-making control device, precise sensing and precise operation of the environment in a greenhouse are achieved through automatic and intelligent means; through real-time monitoring and regulation of environmental parameters in the greenhouse, a more suitable growth environment is provided for crops.

Description

Technical field:

[0001] The present invention belongs to the technical field of intelligent robots, and specifically relates to an inspection robot, control system and method for intelligent greenhouse planting. The automatic inspection crawler robot is configured with navigation, camera and deep learning sub-modules to perform automatic inspection and crop phenotype recognition on crops in the greenhouse. Background technology:

[0002] Greenhouse agricultural robots are a type of equipment that integrates artificial intelligence, machinery, and automation technologies. They are widely used in modern agricultural production and conduct all-weather inspections and management in smart greenhouses to improve agricultural production efficiency and crop quality. At present, decision-making models usually simply rely on manually recorded crop phenotypic data, which is prone to errors and insufficient to provide decision support, resulting in insufficient generalization of the model. The same model has huge differences in decision accuracy in different planting environments. If the farmer has a low level of education, insufficient ability to interpret crop phenotypic data, and insufficient professional knowledge to understand the complex information behind crop phenotypic data, it is difficult to make practical decisions from phenotypic data. In order to obtain accurate crop phenotypic information, farmers need to frequently inspect greenhouses, which not only consumes a lot of manpower, but also increases management costs. Although fixed cameras can be used for monitoring, it is difficult to obtain comprehensive, close-range, and real-time phenotypic information of all crops due to the fixed position of the camera.

[0003] In the prior art, Chinese patent 202110423414.5 discloses a greenhouse monitoring robot and a greenhouse detection system. The greenhouse detection robot includes a frame, a rotating structure, a sliding structure, a data acquisition module and a self-locking component; the rotating structure is installed on the frame, the sliding structure is installed on the rotating structure, the sliding structure drives the data acquisition module to extend and retract, and the self-locking component is installed between the rotating structure and the sliding structure, so that the sliding structure can only start after the rotating structure stops working. It can move freely in the greenhouse, and the data acquisition module can be sent to any position in the greenhouse by using the cooperation of the rotating structure and the sliding structure, so as to obtain more comprehensive and detailed data in the greenhouse, and then use the one-way control of the data acquisition module to protect the data acquisition module to avoid accidental touch or collision with the greenhouse. Chinese patent 202311828901.5 discloses a greenhouse inspection robot system and a multi-source information fusion data processing method, which consists of a magnetic strip, a robot system, an information receiver, and a display terminal. The magnetic strip is placed on the ground of the greenhouse. The wireless transmission unit of the robot system transmits the data from the temperature sensor pair, the humidity sensor pair, and the CO2 sensor pair to the information receiver. The display terminal includes a multi-source information fusion data processing method, including a parameter prediction model and a greenhouse parameter measurement model based on an improved Kalman filter, to achieve high-precision measurement of greenhouse temperature, humidity, CO2, and concentration data. It can realize real-time monitoring of greenhouse parameters, reduce technical costs, and improve the intelligence of greenhouse operations. Chinese patent 202311837703.5 discloses an intelligent control system for agricultural greenhouses based on a remote inspection robot interface, including a network module, a database management module, a server control module and a mobile control module; the sensor group collects the parameters of crop growth factors in the agricultural greenhouse; the server control module processes the crop growth factor parameters and network data according to the deep learning algorithm, generates agricultural tasks, and the inspection robot performs the agricultural tasks; the mobile control module displays the crop growth factor parameters to the user through the cloud server, and enables the user to obtain the task execution status of the inspection robot in real time and process the tasks of the inspection robot. It has the advantages of being able to conduct a comprehensive analysis of the crop growth environment and growth conditions, improve the coordination and intelligence of the inspection robot, and reduce the risk of agricultural investment. The above patents and technical methods in the prior art do not provide an inspection robot with a high level of intelligence. Moreover, there is a problem that the crop phenotypic data is recorded manually, which is prone to errors. The decision accuracy of the same model in different planting environments varies greatly, the generalization ability of the model is insufficient, and it is difficult for fixed monitoring equipment to obtain the phenotypic information of all crops in a comprehensive, close-range and real-time manner. Therefore, developing and designing an intelligent greenhouse planting inspection robot, control system and method to solve the above-mentioned problems has positive economic and social benefits. Summary of the invention:

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and seek to design an automatic inspection robot including multiple functional devices to automatically inspect crops in greenhouses and identify crop phenotypes, so as to achieve scientific planting of crops.

[0005] In order to achieve the above-mentioned purpose, the main structure of an intelligent greenhouse planting decision control device involved in the present invention includes: a camera, a laser radar, a track, a power battery, a control chip, a drive motor, a chassis, and a bracket; the camera is used to capture image information in the greenhouse, including phenotypic recognition of crops, and the camera can transmit high-definition images to the control chip for processing in real time. The laser radar and the camera work together to perform real-time modeling of the environment in the greenhouse, generate accurate maps, and guide the robot's path planning; the control chip receives raw data from the camera and the laser radar, performs image recognition, three-dimensional modeling and other processing, and analyzes the growth status of crops, environmental parameters and the spatial layout in the greenhouse; according to the instructions of the control chip, the drive motor drives the track to move, so that the device can move freely in the greenhouse.

[0006] In an intelligent greenhouse planting decision-making control system disclosed in the present invention, a driving module provides power to drive the robot to move in the greenhouse, ensuring that the robot can flexibly cope with different terrains and obstacles; the driving module is composed of a motor, a crawler or a tire, etc. The motor receives instructions from the control system and transmits power to the crawler or the tire through a mechanical device; the movement of the robot can be controlled by adjusting the speed and direction, so that it can flexibly turn in a complex environment and overcome obstacles, thereby achieving comprehensive coverage and monitoring; the camera and the laser radar navigation module provide environmental perception and navigation functions, realize autonomous navigation and obstacle avoidance, collect spatial structure and obstacle distribution information inside the greenhouse, and ensure the safe movement of the robot; by emitting and receiving reflected signals from lasers, the distance to surrounding objects is measured, and a two-dimensional or three-dimensional map of the environment is generated, which can help the robot update the map in real time, identify obstacles and calculate the best path; the camera captures image information of the surrounding environment and compares it with the laser radar. The combination of optical radar data provides a more comprehensive environmental perception capability. The camera can help identify dynamic objects, crop status, and potential obstacles. The navigation module can develop an efficient and safe path plan to reduce human intervention. The camera module accurately monitors and evaluates the growth status of crops and provides real-time image data. It helps identify crop growth conditions, pests and diseases, nutritional deficiencies and other problems. The camera module uses image processing algorithms to analyze the growth status of crops by taking images of crops. For example, it can detect the color, shape and other characteristics of crops to determine their health status. Deep learning algorithms are used for image recognition to achieve automated crop health assessment. These analysis results will be transmitted back to the main control system to provide growers with scientific basis and decision support. The circuit motherboard module serves as the core control unit of the system, coordinating the work between various modules. It processes data from various sensors and cameras to execute decisions and control instructions.

[0007] Compared with the prior art, the present invention captures high-definition images through a camera, transmits them to a control chip for processing in real time, and performs phenotypic identification of crops in the greenhouse; the laser radar works in conjunction with the camera to perform real-time three-dimensional modeling of the environment in the greenhouse, generates an accurate map, and provides basic data for the robot's path planning and precise operation; the control chip receives raw data from the camera and the laser radar, performs advanced processing such as image recognition and three-dimensional modeling, and analyzes the growth status of crops, environmental parameters, and the spatial layout in the greenhouse; the drive motor drives the crawler to move and perform monitoring tasks according to the instructions of the control chip. The crawler design enhances the ground adaptability and stability of the device, so that it can move freely and work efficiently under different terrain and soil conditions; the intelligent greenhouse planting decision control device realizes accurate perception and precise operation of the environment in the greenhouse through automation and intelligent means, reduces errors caused by manual intervention, reduces labor costs, and improves the economic benefits of agricultural production; by real-time monitoring and regulation of environmental parameters in the greenhouse, a more suitable growth environment is provided for crops, which is conducive to the healthy growth, high yield and high quality of crops. Description of the drawings:

[0008] Figure 1 This is a schematic diagram of the main structural principle of the intelligent greenhouse planting inspection robot involved in the present invention.

[0009] Figure 2 This is a front view of the intelligent greenhouse planting inspection robot involved in the present invention.

[0010] Figure 3 This is a bottom structure diagram of the intelligent greenhouse planting inspection robot involved in the present invention.

[0011] Figure 4 This is a control system diagram of the intelligent greenhouse planting inspection robot involved in the present invention.

[0012] Figure 5 This is a flow chart of the control method of the intelligent greenhouse planting inspection robot involved in the present invention. Specific implementation method:

[0013] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0014] Embodiment 1:

[0015] The main structure of an intelligent greenhouse planting decision control device involved in this embodiment includes: a camera 1, a laser radar 2, a crawler 3, a power battery 4, a control chip 5, a drive motor 6, a chassis 7 and a bracket 8; wherein the crawler 3 is installed below the chassis 7, the drive motor 6 is installed inside the crawler 3 and fixed to the tire, the control chip 5 is installed below the drive motor 6, and the power battery 4 is installed on the side opposite to the control chip; a retractable bracket 8 is installed above the chassis 7, the top of the bracket 8 and the shorter side of the support bracket 8 are installed with a camera 1, and a laser radar 2 is installed on the chassis 7 on the shorter side of the support bracket 8;

[0016] The working principle and beneficial effects of this embodiment are as follows: when it is necessary to inspect the greenhouse, the camera 1 is used to capture image information in the greenhouse, including phenotypic identification of crops. The camera 1 can transmit high-definition images to the control chip 5 for processing in real time. The laser radar 2 works with the camera 1 to model the environment in the greenhouse in real time, generate accurate maps, and guide the robot's path planning; the control chip 5 receives the original data from the camera 1 and the laser radar 2, performs image recognition, three-dimensional modeling and other processing, and analyzes the growth status of crops, environmental parameters and the spatial layout in the greenhouse; according to the instructions of the control chip 5, the drive motor 6 drives the crawler 3 to move, so that the device can move freely in the greenhouse.

[0017] This embodiment uses a camera to capture image information of crops, identify and evaluate the growth status of crops, such as the health of plants, the occurrence of pests and diseases, etc., which helps farm managers take timely measures; the map and spatial information provided by the lidar can help monitor the environmental conditions inside the greenhouse, such as light, air circulation and temperature distribution, and can perform real-time image recognition and spatial modeling, analyze the current status of crops and environmental conditions, and provide support for decision-making; through the application of automation and intelligent technologies, it promotes the transformation of agriculture to smart agriculture and improves production efficiency and management level.

[0018] Embodiment 2:

[0019] The main structure of an intelligent greenhouse planting decision-making control system involved in this embodiment includes: a driving module 9, a camera and laser radar navigation module 10, a camera module 11, and a circuit main board module 12; wherein the driving module 9 provides power to drive the robot to move in the greenhouse, ensuring that the robot can flexibly cope with different terrains and obstacles; the driving module 9 is usually composed of a motor, tracks or tires, etc. The motor receives instructions from the control system and transmits power to the tracks or tires through a mechanical device; the movement of the robot can be controlled by adjusting the speed and direction, so that it can flexibly turn in a complex environment and overcome obstacles, thereby achieving comprehensive coverage and monitoring; the camera and laser radar navigation module 10 provides environmental perception and navigation functions, realizes autonomous navigation and obstacle avoidance, collects spatial structure and obstacle distribution information inside the greenhouse, and ensures the safe movement of the robot; by emitting lasers and receiving reflected signals, the distance to surrounding objects is measured to generate an environmental image. The two-dimensional or three-dimensional map of the environment can help the robot update the map in real time, identify obstacles and calculate the best path; the camera captures the image information of the surrounding environment, combines it with the data of the laser radar, and provides a more comprehensive environmental perception capability. The camera can help identify dynamic objects, the state of crops, and potential obstacles; the navigation module can develop an efficient and safe path plan to reduce human intervention; the camera module 11 accurately monitors and evaluates the growth status of crops, provides real-time image data, and helps identify the growth status, pests and diseases, and nutritional deficiencies of crops; the camera module 11 analyzes the growth status of crops by taking images of crops and using image processing algorithms; for example, it can detect the color, shape and other characteristics of crops to determine their health status; use deep learning algorithms for image recognition to achieve automated crop health assessment, and these analysis results will be transmitted back to the main control system to provide scientific basis and decision support for growers. The circuit motherboard module 12 serves as the core control unit of the system, coordinating the work between various modules; processing data from various sensors and cameras, and executing decision-making and control instructions.

[0020] The circuit mainboard involved in this embodiment integrates a variety of electronic components, including processors, memories, power management modules and communication interfaces; receives data from the camera module and the laser radar navigation module, processes and analyzes the data; based on the algorithm and the designed logic, the mainboard performs calculation, decision-making and control functions; for example, the received image information and environmental data will be processed to generate the robot's action instructions and command the drive module to execute the movement; in addition, the circuit mainboard may also be responsible for connecting to the external network to facilitate remote monitoring and control, thereby realizing more complex instructions and information sharing; the drive module 9 improves the flexibility and adaptability of the robot in the greenhouse, ensuring comprehensive coverage and monitoring of the crop growth environment. Reduces the interruption of operations caused by terrain restrictions, improves work efficiency and stability; the camera and laser radar navigation modules improve the intelligence level of the robot's operation, realizes autonomous navigation and obstacle avoidance, reduces manual intervention, improves the accuracy and efficiency of inspections, and ensures comprehensive monitoring of the crop growth environment; the camera module 11 realizes accurate detection and evaluation of the crop growth status, provides a scientific basis for planting decisions, improves the refinement level of crop management, and helps to achieve precision agriculture and improve crop yield and quality.

[0021] The functions of each module in the intelligent greenhouse planting decision-making control system involved in this embodiment complement each other. By integrating multiple sensing and control technologies, it comprehensively monitors the crop growth environment and improves the intelligence level of agricultural management. It can not only improve work efficiency, but also provide data support for precision agriculture, thereby achieving higher crop yields and quality.

[0022] Embodiment 3:

[0023] This embodiment is improved on the basis of embodiment 2. The driving module 9 includes: a crawler, a driving motor and a power battery. The crawler-driven chassis is designed with a width of 70 cm and a height of 40 cm. The crawler design provides stable mobility, adapts to the complex terrain environment in the greenhouse, and meets the inspection passability requirements of the ridges in the greenhouse. The chassis includes a power battery module to ensure that a 1000-meter inspection is completed without additional charging; the camera and laser radar navigation module 10 is responsible for navigation and map modeling. The camera and laser radar navigation module 10 is used to model the environment in the greenhouse in real time, generate accurate maps, guide the path planning and navigation of the robot, and realize real-time obstacle avoidance through the combination of laser radar and camera to ensure the safe movement of the robot during the inspection process; the camera module 11 includes a TOF camera and its retractable bracket. The 8 TOF cameras provide high-precision depth of field detection capabilities, can capture three-dimensional information of crops, and provide image data containing distance for subsequent analysis of crop phenotypes using embedded image recognition. The camera bracket is designed to be retractable and can adjust the angle and height of the camera to adapt to different periods and inspection requirements during the growth of crops.

[0024] The crawler design of the drive module 9 involved in this embodiment with a width of 70 cm and a height of 40 cm provides good stability and adaptability, allowing the robot to move freely on complex terrain and ridges in the greenhouse; the layout of the crawler helps to disperse the pressure of the robot on the soil and reduce the risk of crop damage; the drive motor provides power for the crawler, enabling it to move flexibly at different speeds and directions, and the motor adjusts the speed and steering through the instructions of the control system to adapt to the operation requirements; the built-in power battery module ensures that the robot can conduct inspections up to 1,000 meters without additional charging, effectively extending the robot's working time and improving the operation efficiency; the crawler provides stable support, allowing the robot to easily cope with various complex terrains in the greenhouse; the long-lasting power battery supports long-term inspections, reduces the charging frequency, and improves the overall efficiency; the camera and laser radar navigation module 10 scans the surrounding environment through the laser radar, collects the spatial information of the environment in real time, and generates an accurate 2D or 3D map; at the same time, the camera captures image information and recognizes environmental features; the distance data provided by the laser radar is combined with the image information obtained by the camera, allowing the robot to identify obstacles in real time during the inspection process and intelligently plan to avoid paths, thereby ensuring mobile safety. The accurate environmental model enables the robot to make autonomous decisions, optimize inspection routes, and improve work efficiency; the real-time obstacle avoidance capability ensures that the robot can move safely in complex greenhouse environments to avoid damaging crops or its own equipment. Camera module 11 uses time-of-flight (TOF) technology. The eight TOF cameras can provide high-precision depth information and capture the three-dimensional form of crops; by emitting light signals and measuring the signal return time, accurate depth images are constructed; the retractable bracket design enables the height and angle of the camera to be adjusted to adapt to crops at different growth stages and specific inspection needs, so as to obtain the best viewing angle for imaging. The efficient depth of field detection capability helps analyze the growth status and health status of crops, provides support for subsequent image recognition analysis, and directly improves the level of refinement of crop management; the adjustable bracket allows the system to flexibly respond to changes in the crop growth process and ensure accurate data collection during the inspection process.

[0025] The intelligent greenhouse planting decision control system involved in this embodiment can achieve: the robot conducts efficient inspections in the greenhouse, monitors the growth status and environmental indicators of crops around the clock, and ensures the best growth conditions; provides professional decision support through real-time data analysis to help farm managers conduct scientific agricultural management; the application of precision agricultural technology helps to improve crop yields and quality, reduce pests and other risks; automatic inspections and intelligent perception greatly reduce dependence on manual operations, release human resources, and make farm management more efficient. It can not only improve the level of intelligent agricultural management, but also provide farmers with more scientific crop management tools to promote agricultural modernization and sustainable development.

[0026] Embodiment 4:

[0027] This embodiment is improved on the basis of Embodiment 3, and the circuit main board module 12 includes an embedded computing submodule, a deep learning submodule and a communication submodule; wherein the embedded computing submodule processes the data from the camera in real time, and obtains crop phenotypic data analysis through an embedded image recognition algorithm; the communication submodule ensures that the robot can transmit and communicate data with the remote server; the deep learning submodule in the robot judges whether the crop phenotypic data is abnormal through the environmental data (greenhouse air temperature and humidity, soil temperature and humidity) calibrated by experts and the correlation model of the crop growth stage, and can also issue an alarm for abnormal conditions of the crops; the inspection robot can be connected to a large model in the cloud. When the user does not understand the data, he can ask questions to the large model. The large model adds the crop phenotypic data of the inspection robot to the question and searches the existing knowledge base to give the user a text or voice explanation.

[0028] The embedded computing submodule involved in this embodiment improves the speed of data processing and image recognition, ensures timely monitoring and evaluation of crop growth status, reduces data transmission delays and bandwidth occupancy, and reduces the system's dependence on external computing resources; the deep learning submodule realizes intelligent evaluation and judgment of crop growth status, improves the accuracy and scientificity of decision-making, and enhances the system's autonomous learning and adaptability, enabling it to continuously optimize and improve crop growth management strategies; the communication submodule realizes remote management and control of the system, improves the convenience and efficiency of management, promotes data sharing and communication, and provides strong support for the intelligence and precision of agricultural production.

[0029] Embodiment 5:

[0030] This embodiment is improved on the basis of embodiment 4, and the deep learning submodule includes:

[0031] The data processing unit receives the embedded computing submodule to process the data from the camera in real time, and obtains crop phenotypic data analysis through the embedded image recognition algorithm, including the following table:

[0032]

[0033]

[0034] This embodiment uses the average calculation of the inspection area to reduce the error accumulation of single plant identification and obtain relatively more accurate information with lower calculation accuracy requirements. All phenotypic information is automatically filled in the table or database without manual recording, judgment and entry into the database.

[0035] In this embodiment, the embedded computing submodule processes the camera data in real time, and uses the embedded image recognition algorithm. The system can accurately analyze the phenotypic data of the crops. These accurate data provide growers with a comprehensive view of the crop growth status, which helps to achieve precise planting management; automated phenotypic data analysis reduces the workload of manual inspections and records, saves human resources, and improves the efficiency of data collection and analysis; by calculating the phenotypic data according to the average of the inspection area, the system can effectively reduce the error accumulation that may be caused by the identification of a single plant, while maintaining the calculation accuracy, improving the accuracy and reliability of the data, and providing a more solid basis for planting decisions; the system can store the automatically filled phenotypic information in a table or database to facilitate subsequent data analysis and mining.

[0036] Embodiment 6:

[0037] The present embodiment involves a smart greenhouse planting decision control method comprising the following steps:

[0038] Photograph the greenhouse environment and perform real-time modeling to generate accurate maps to guide the robot’s path planning and navigation;

[0039] Capture the three-dimensional information of crops and provide image data containing distance for subsequent analysis of crop phenotypes using embedded image recognition;

[0040] Through the correlation model between environmental data and crop growth stages calibrated by experts, it is possible to judge whether the crop phenotypic data is abnormal, mark out or not use the phenotypic data that is misidentified, and issue an alarm for abnormal crop conditions;

[0041] The inspection robot is connected to a multimodal big model in the cloud. When users do not understand the data, they can ask questions to the big model. The big model adds the inspection robot's crop phenotypic data and searches the existing knowledge base to the questions and provides users with text or voice explanations.

[0042] The camera involved in this embodiment captures visual information in the greenhouse and provides rich environmental details. The lidar accurately measures the distance and constructs the spatial structure of the environment by emitting laser pulses and measuring their return time. The combination of the two can generate a three-dimensional environmental model that contains both visual features and accurate spatial information. The TOF camera calculates the distance between the object and the camera by measuring the time difference from the emission of the laser pulse to its reflection and return by the object. The eight cameras are distributed at different angles and positions and can capture the three-dimensional information of the crop in all directions. Through the fusion processing of multi-camera data, a high-precision three-dimensional model of the crop can be generated. The captured three-dimensional information is converted into image data containing distance. This data not only contains the visual features of the crop, but also contains the depth information corresponding to each pixel. The deep learning module in the robot uses deep learning algorithms to analyze crop phenotypic data. These algorithms are trained with a large amount of environmental data calibrated by experts and crop growth stage data, and can learn the normal operation of crops. growth characteristics and patterns; by comparing the crop phenotypic data obtained in real time with the trained model, the deep learning module can identify abnormal data that does not match the normal growth pattern; for the identified abnormal data, the deep learning module will mark or ignore it to avoid interference with subsequent analysis. At the same time, for serious abnormal situations, the module will trigger an alarm mechanism to remind users to take timely measures to solve the problem; a large knowledge base and reasoning model are deployed on the cloud, which contains rich agricultural knowledge, crop growth laws, expert experience and other information; users can ask questions to the cloud-based large model through the interactive interface with the inspection robot. These questions may involve the interpretation of crop phenotypic data, analysis of growth status or suggestions for handling abnormal situations; after receiving the user's questions, the cloud-based large model will combine the crop phenotypic data provided by the inspection robot and the information in its own knowledge base for comprehensive analysis and reasoning, generate corresponding text or voice explanations based on the analysis results, and return them to the user through the inspection robot.

[0043] This embodiment uses the collaborative work of the camera and the laser radar to capture the environmental information in the greenhouse in real time, including terrain, obstacles, etc., and generate a high-precision three-dimensional map. Based on the generated map, the system can automatically plan the optimal inspection path to ensure that the robot moves efficiently and safely in the greenhouse, improve the accuracy and efficiency of the inspection, reduce navigation errors caused by complex environments, reduce the risk of robot collision and damage, and extend the service life; the TOF camera uses time-of-flight technology to accurately measure the distance on the surface of an object, capture the three-dimensional morphology and growth environment of crops, and provide image data containing distance information, which provides richer data support for subsequent image recognition and analysis, improves the accuracy and reliability of crop phenotypic data analysis, provides a scientific basis for planting decisions, enriches the means of monitoring crop growth status, and helps to promptly discover and solve problems in the growth process; Using deep learning algorithms, the system can automatically analyze crop phenotypic data to determine whether it is abnormal. For phenotypic data that is identified incorrectly, the system can automatically mark or ignore it to avoid interference with planting decisions. At the same time, for abnormal crop conditions, the system can promptly issue alarms, thereby improving the level of intelligence in planting management, reducing the impact of human factors on crops, and promptly discovering and dealing with problems in the crop growth process, ensuring the healthy growth and high yield and quality of crops. Users can interact with the inspection robot through the cloud-based big model and ask questions about crop phenotypic data. The big model uses the inspection robot's crop phenotypic data and its own knowledge base to provide users with detailed answers and explanations, thereby improving the user experience and enabling users to more intuitively understand the crop growth status and data analysis results, enhancing the intelligence and interactivity of the system, and providing users with more convenient and efficient data query and consulting services.

[0044] Embodiment 7:

[0045] This embodiment is improved on the basis of embodiment 6, and the specific steps of constructing a multi-modal large model include:

[0046] Collect multiple modal information of the target, including voice information, image information and text information, pre-process the collected information, and numerically quantify the performance of each feature to obtain pre-processed information;

[0047] A model is constructed based on the collected information and trained. Through training, the initial model learns the voice and language features of the person to obtain a pre-trained model.

[0048] Optimize the pre-trained model. During the model training process, evaluate the pre-trained module and feed the evaluation results back to the model training module to reduce the difference between the pre-trained model and the target.

[0049] This embodiment collects different types of data from the target object and performs preprocessing to convert non-numerical data into numerical form through vectorization, feature extraction or label encoding. After building the model, the preprocessed data is used to train the model to learn how to map the input data to the output target by minimizing a loss function. The generalization ability of the model is verified by evaluating the model performance on a set of independent test data, and the model structure and parameters are adjusted according to the evaluation results. During the training process, the feedback mechanism helps the model to gradually reduce the difference between the model and the target output and improve the performance.

[0050] This embodiment can comprehensively capture the characteristics and information of the target by simultaneously collecting multiple types of data such as voice, image and text, providing a rich data source for subsequent model construction. The preprocessing effectively removes noise, redundancy and inconsistency in the original data, improves the quality and availability of the data, improves the reliability and consistency of the data, and provides high-quality data input for subsequent steps, so that various types of data can be processed and analyzed under a unified framework, thereby enhancing the generalization ability of the model; a model capable of processing multimodal data is constructed, and by learning voice and language features, the model can more accurately understand and respond to human voice and language input, thereby improving the efficiency and accuracy of human-computer interaction; through continuous optimization and improvement, the accuracy of the model is improved, so that it can better adapt to actual application scenarios, and the difference between the model and the target is reduced, so that the model can more accurately reflect the characteristics and laws of the target object, thereby improving the practical value and reliability of the model.

[0051] Embodiment 8:

[0052] The implementation of this embodiment is to verify the use effect of the intelligent greenhouse planting inspection robot involved in Example 1, and the following comparative experiments are carried out;

[0053] (1) Experimental equipment:

[0054] Experimental group: the intelligent greenhouse planting inspection robot involved in Example 1;

[0055] Control group: manual inspection;

[0056] (2) Experimental methods:

[0057] Conduct inspections on greenhouses of the same area and compare the inspection time and results;

[0058] (3) Experimental results:

[0059] Inspection area Inspection time Effect Experimental Group <![CDATA[50m 2 ]]> 30min excellent Control group <![CDATA[50m 2 ]]> 67min qualified

[0060] It can be seen that the intelligent greenhouse planting inspection robot involved in Example 1 is superior to the control group in terms of inspection time and effect due to the configuration of navigation, camera and deep learning sub-module structures, thereby improving the efficiency and accuracy of automatic inspection of greenhouse planting.

Claims

1. An intelligent greenhouse planting decision control device, characterized in that: The main structure includes: camera, lidar, tracks, power battery, control chip, drive motor, chassis, and bracket; the camera is used to capture image information in the greenhouse and transmit high-definition images to the control chip for processing in real time. The lidar works with the camera to model the environment in the greenhouse in real time, generate accurate maps, and guide the robot's path planning; the control chip receives raw data from the camera and lidar, performs image recognition and three-dimensional modeling, and analyzes the growth conditions of crops, environmental parameters, and the spatial layout in the greenhouse. The drive motor drives the tracks to move according to the instructions of the control chip.

2. The intelligent greenhouse planting decision control device according to claim 1 is characterized in that: The crawler is installed under the chassis, the drive motor is installed inside the crawler and fixed to the tire, the control chip is installed under the drive motor, and the power battery is installed on the side opposite to the control chip; A retractable bracket is installed above the chassis, and cameras are installed on the top of the bracket and on the shorter side of the supporting bracket. A laser radar is installed on the chassis on the shorter side of the supporting bracket.

3. An intelligent greenhouse planting decision control system, characterized in that: The main structure includes: a drive module, a camera and lidar navigation module, a camera module, and a circuit mainboard module; the drive module provides power to drive the robot to move in the greenhouse; the camera and lidar navigation module provides environmental perception and navigation functions, collects information on the spatial structure and obstacle distribution inside the greenhouse, measures the distance to surrounding objects through laser emission and reception of reflected signals, and generates a two-dimensional or three-dimensional map of the environment, so that the robot can update the map in real time, identify obstacles and calculate the optimal path; the camera module monitors and evaluates the growth status of crops, analyzes the growth status of crops by taking images of crops, and uses image processing algorithms to perform image recognition; the circuit mainboard module is the core control unit, which coordinates the work between various modules, processes data from various sensors and cameras, and executes decisions and control instructions.

4. The intelligent greenhouse planting decision control system according to claim 3 is characterized in that: The driving module includes an electric motor, tracks or tires. The electric motor receives instructions and transmits power to the tracks or tires through mechanical devices; the movement of the robot is controlled by adjusting the speed and direction.

5. An intelligent greenhouse planting decision control system according to claim 3 or 4, characterized in that: The circuit mainboard module integrates electronic components, including a processor, memory, a power management module and a communication interface; it receives data from the camera module and the lidar navigation module, and processes and analyzes the data; the circuit mainboard module is connected to an external network for remote monitoring and control.

6. The intelligent greenhouse planting decision control system according to claim 3 is characterized in that: The driving module includes: tracks, drive motors and power batteries. The track-driven chassis is 70 cm wide and 40 cm high. The chassis contains a power battery module and can complete a 1,000-meter inspection at a time. The camera and lidar navigation module is responsible for navigation and map modeling. It performs real-time modeling of the greenhouse environment, generates maps, guides the robot's path planning and navigation, and avoids obstacles in real time through the combination of lidar and cameras. The camera module includes a TOF camera and its retractable bracket. The TOF camera captures three-dimensional information of crops and provides image data containing distance for subsequent embedded image recognition analysis of crop phenotypes. The camera bracket is retractable to adjust the angle and height of the camera.

7. The intelligent greenhouse planting decision control system according to claim 5, characterized in that: The circuit main board module includes an embedded computing submodule, a deep learning submodule and a communication submodule; among them, the embedded computing submodule processes the data from the camera in real time, and obtains crop phenotypic data analysis through an embedded image recognition algorithm; the deep learning submodule judges whether the crop phenotypic data is abnormal through an expert-calibrated correlation model of environmental data and crop growth stages, and issues an alarm for abnormal crop conditions; the communication submodule transmits and communicates data with a remote server, connects to a large model in the cloud, and provides text or voice explanations for users' questions based on crop phenotypic data and an existing knowledge base.

8. The intelligent greenhouse planting decision control system according to claim 7, characterized in that: The deep learning submodule includes: a data processing unit, which receives the embedded computing submodule to process the data from the camera in real time, and obtains and analyzes the crop phenotypic data through the embedded image recognition algorithm.

9. An intelligent greenhouse planting decision control method, characterized in that: The following steps are involved: Photograph the greenhouse environment, model it in real time, generate maps, and plan the robot’s path and navigation; Capture the three-dimensional information of crops and provide image data containing distance; Through the correlation model between environmental data and crop growth stages calibrated by experts, it is possible to judge whether the crop phenotypic data is abnormal, mark out or not use the phenotypic data that is misidentified, and issue an alarm for abnormal crop conditions; The inspection robot connected to the multimodal large model in the cloud can explain questions based on crop phenotypic data and knowledge base.

10. The intelligent greenhouse planting decision control method according to claim 9, characterized in that: The specific steps to build a multimodal large model include: Collect the modal information of the target, including voice information, image information and text information, perform preprocessing, and numerically quantify the performance of each feature to obtain preprocessing information; The model is constructed based on the collected information, and the model is trained. The initial model is learned through training to obtain a pre-trained model; Optimize the pre-trained model. During the model training process, evaluate the pre-trained module and feed the evaluation results back to the model training module.

Images