Facility horticulture greenhouse-oriented semantic perception-driven fine flame soil disinfection system

Abstract

The application discloses a facility horticulture greenhouse intelligent flame soil disinfection system and method based on multi-source visual perception. The system is composed of a multi-source visual perception module, an edge computing decision module, a flame execution module and a tracked mobile chassis. The system uses an improved YOLOv8n lightweight model and binocular spatial positioning technology to realize real-time dynamic perception of weeds, crops and key facilities in the complex environment of the greenhouse. The decision module is built-in with a soil heat penetration depth control algorithm, which quantifies heat input by adjusting the linear speed of the chassis to ensure the disinfection effect of deep soil. At the same time, the system adopts a nonlinear decision mapping logic, has an automatic compensation function of the safe distance of mulch heat radiation and a heat safety self-diagnosis mechanism. The application not only solves the closed-loop collaborative control problem of weeding and disinfection of the facility greenhouse, but also guarantees the safety and smoothness of the facility in the operation process, and has the advantages of precision, high efficiency, fast real-time response and the like.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture technology, specifically to a refined flame disinfection system based on multi-source visual perception for greenhouse horticulture. This invention is particularly suitable for greenhouse cultivation environments of crops such as melons and vegetables, enabling the coordinated operation of precise flame weed control and deep soil-borne disease thermal inactivation. Background Technology

[0002] With the advancement of agricultural modernization, weeding and soil pest and disease control, as core aspects of greenhouse horticulture management, are transitioning from traditional manual methods to mechanized and intelligent processes. In greenhouse horticulture, crops are planted at high densities and have rapid growth cycles, requiring extremely stable and safe growing environments. Currently, weeding and disease control in greenhouses mainly rely on manual labor or chemical agents. Manual weeding is labor-intensive and inefficient; long-term reliance on chemical agents can easily lead to problems such as pathogen resistance, soil degradation, and agricultural product residues.

[0003] As a physical control method, flame weeding and soil disinfection technology inactivates weed cells and kills pathogens and root-knot nematodes in the soil through high-temperature heat, offering advantages such as no chemical residue and rapid response. However, existing flame disinfection equipment still faces the following technical bottlenecks in its application in protected horticulture environments:

[0004] Lack of environmental perception: Existing equipment mostly adopts fixed modes or manual control, lacking the ability to intelligently perceive the complex and unstructured environment inside greenhouses. In the confined space of greenhouses covered with plastic film, if weeds, crops, and key facilities cannot be distinguished in real time, it is easy to cause accidental spraying, resulting in serious economic losses.

[0005] Insufficient operational control precision: Weeding and soil disinfection require precise management of heat input. If the driving speed and flame output cannot be dynamically matched according to the target type and disease risk, it is difficult to ensure that heat effectively penetrates into the deep soil layers, and it is also difficult to ensure safe avoidance of heat-sensitive facilities such as mulch film.

[0006] The lack of system-level safety assurance mechanisms is a significant issue: flame operations involve high temperatures, and critical components are subjected to prolonged thermal stress. Existing research largely focuses on single identification or mechanism design, lacking a comprehensive verification system that deeply couples visual perception, thermodynamic analysis, and behavioral decision-making. This results in insufficient operational stability and safety response capabilities of the system under complex operating conditions.

[0007] Therefore, there is an urgent need to develop an intelligent flame disinfection system that integrates high-precision environmental perception, dynamic thermal dosage decision-making, and safe and reliable execution to meet the special operational needs of facility horticulture greenhouses. Summary of the Invention

[0008] This invention aims to address the aforementioned problems in the existing technology by providing a semantically perceptive-driven precision flame soil disinfection system for greenhouse horticulture. This system can achieve real-time and accurate identification of multiple targets within the greenhouse, and realizes quantitative management of heat input through the coupling of a heat penetration depth control algorithm and a motion control algorithm. Furthermore, it significantly improves the efficiency and safety of soil disinfection while ensuring the safety of the mulch film and crops.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A semantically aware, precision flame soil disinfection system for greenhouse horticulture includes:

[0011] The multi-source vision perception module is configured to acquire color images, depth maps, and point cloud data of the working environment through a binocular camera. This data is used to identify weeds, crops, mulch film, obstacles, and personnel targets within the greenhouse, and to output the target category. and relative position vector ;

[0012] The edge computing decision module, equipped with a lightweight deep learning model, is configured to receive the target category. and relative position vector Based on this, motion control commands and flamethrower execution commands are generated in real time;

[0013] The flame actuation module includes a flame nozzle array and a gas control unit, wherein the flame actuation module is in flame-emitting state. Controlled by the flamethrower execution command output by the edge computing decision module;

[0014] The tracked mobile chassis adopts a tracked walking mechanism with independent left and right drive, and is configured to perform path avoidance and inter-row cruise based on the motion control commands output by the edge computing decision module and the equivalent differential wheel kinematic model.

[0015] Furthermore, the edge computing decision module incorporates a heat penetration depth control algorithm for soil pathogen inactivation; the edge computing decision module is configured to adjust the linear velocity of the tracked mobile chassis based on the risk of soil-borne diseases in the area to be disinfected. To control the duration of flame exposure on a unit area of ​​soil This allows for disinfection at the predetermined depth. instantaneous soil temperature The temperature threshold required for pathogen inactivation is met; wherein, the instantaneous soil temperature According to the formula Sure, The soil surface equivalent temperature is positively correlated with the output power of the flame actuator module. For soil thermal diffusivity, This is the error compensation function.

[0016] Furthermore, the edge computing decision module is configured to determine the linear velocity of the tracked mobile chassis. The velocity should be controlled within the range of 0.2 m / s to 0.4 m / s to ensure the effective heat input required for pathogen inactivation.

[0017] Furthermore, the lightweight deep learning model is built on an improved YOLOv8n architecture, with 3.01M model parameters, and an average inference time of 2.6ms per image on the edge computing platform to which the edge computing decision module belongs.

[0018] Furthermore, the lightweight deep learning model achieves a precision of 0.955 for identifying weeds, crops, and plastic film targets within the greenhouse, and an average precision (mAP@0.5) of 0.902.

[0019] Furthermore, the edge computing decision module is configured to use a nonlinear decision mapping operator. Transforming perceived information into execution instructions satisfies the mapping relationship: ,in, For the first The perception tensor of each target is a priority operator based on the target category.

[0020] Furthermore, the nonlinear decision mapping operator It is configured to: when the multi-source visual perception module identifies a human target, force the output of an instruction to stop the tracked mobile chassis and shut down the flame execution module's flame-spraying state, and this instruction has a higher priority than the response instruction to weeds, crops, mulch film or obstacle targets.

[0021] Furthermore, the edge computing decision module is also configured to perform an automatic compensation function for the thermal radiation safety distance of the mulch film; the edge computing decision module determines the lateral offset of the mulch film edge relative to the system's centerline based on the information identified by the multi-source visual perception module. Real-time calculation of dynamic safety distance And enter the preset obstacle avoidance threshold at the edge of the mulch film. Automatically correct yaw rate .

[0022] Furthermore, the edge computing decision module is also configured to: when the edge of the mulch film enters a preset obstacle avoidance threshold. And the dynamic safety distance When the temperature drops below the minimum safety threshold, the flame execution module's flame emission state is turned off.

[0023] Furthermore, the flame execution module also includes a temperature sensor for real-time monitoring of the main body temperature of the flame nozzle array; the edge computing decision module has built-in thermal safety self-diagnosis logic, configured to automatically execute a deceleration cooling action or switch to pulse flame mode when the temperature sensor detects that the main body temperature is approaching a preset critical risk area of ​​900℃; the discrete update process of the equivalent differential wheel kinematic model satisfies:

[0024] ,

[0025] in, , , They represent the first Chassis position coordinates and heading angle at each sampling time Linear velocity, Angular velocity, The sampling period; the edge computing decision module is configured to adjust... and Control the tracked mobile chassis to move smoothly along the avoidance path.

[0026] Compared with the prior art, the beneficial effects of this application are as follows:

[0027] Precise and efficient disinfection: Through the coordinated adjustment of heat penetration depth control algorithm and driving speed, the heat input is quantitatively controlled, ensuring the disinfection effect of deep soil.

[0028] Real-time performance and lightweight design: An improved lightweight recognition model is adopted, which significantly improves the processing speed and recognition accuracy on the edge computing platform, meeting the real-time operation requirements of the complex environment of the greenhouse.

[0029] High facility safety: It has a dedicated function for compensating for thermal radiation from the mulch film and a self-diagnostic logic for thermal safety, which effectively avoids the risk of thermal damage to the mulch film and other facilities caused by flames.

[0030] Stable and controllable operation: The use of tracked differential drive and smooth path planning enhances the prototype's passability and movement stability on soft greenhouse ground. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the overall system structure and operating environment provided in an embodiment of the present invention.

[0032] Figure label:

[0033] 100 is the multi-source visual perception module, 110 is the edge computing decision module, 120 is the flame execution module, 130 is the tracked mobile chassis, 140 is the gas control unit, and 150 is the dynamic safety distance at the edge of the ground film. Detailed Implementation

[0034] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.

[0035] Example 1:

[0036] System overall architecture implementation:

[0037] The present invention provides a semantic perception-driven fine flame soil disinfection system for facility horticulture greenhouses. Its hardware architecture mainly consists of a multi-source visual perception module, an edge computing decision module, a flame execution module, and a tracked mobile chassis.

[0038] In actual operation, the visual perception module uses a ZED X binocular camera mounted on the front gimbal of the chassis to collect real-time RGB images and depth information of weeds, crops (such as melon seedlings), mulch, obstacles, and workers inside the greenhouse. The edge computing module uses the NVIDIA Jetson AGX Orin platform to run deep learning models and solve control commands. The flame execution module includes a flame nozzle array with an adjustable mounting height to accommodate disinfection needs at different ridge heights. The chassis adopts a tracked structure to ensure good passability and steering accuracy on the soft, damp ground of the greenhouse.

[0039] Visual recognition and model deployment and implementation

[0040] The system's lightweight deep learning model is built on an improved version of YOLOv8n, with only 3.01M parameters, making it ideal for deployment on resource-constrained edge computing devices.

[0041] Detection performance: Trained on a large amount of real-world data from greenhouses, the model achieved a single-frame inference speed of 2.6ms on an edge computing platform. For key targets within the greenhouses, its recognition accuracy reached 0.955, and its mean accuracy (mAP@0.5) reached 0.902, effectively solving the recognition problem under conditions of uneven lighting and severe occlusion.

[0042] Semantic mapping: The model's recognition result is output as the target category. and the corresponding relative position vector This information will serve as input for subsequent behavioral decisions.

[0043] Implementation Logic of Soil Thermal Infiltration Disinfection Algorithm

[0044] The core innovation of this invention lies in achieving precise control over the depth of soil disinfection. The edge computing decision module incorporates a thermal penetration depth control algorithm for inactivating soil pathogens.

[0045] Control principle: The system is based on the preset disinfection depth and soil thermal diffusivity The required effective flame action time is calculated using Formula 1. .

[0046] Dynamic adjustment: Since the effective thermal effect length of the flamethrower is fixed, the system dynamically adjusts the chassis linear velocity. The total heat input is controlled within the range of 0.2~0.4 m / s. For example, if it is necessary to kill deeper soil-borne pathogens, the system will automatically reduce the heat input. This increases the duration of flame exposure per unit area, ensuring the instantaneous temperature of deep soil layers. The inactivation threshold has been reached.

[0047] Kinematic Models and Behavioral Decision Implementation

[0048] The system's motion control employs an equivalent differential wheel kinematic model. During the sampling period... Internally, the system updates the prototype's pose in the global coordinate system in real time using Formula 3. , , ).

[0049] Priority mechanism: The system uses a nonlinear decision mapping operator. (Formula 2) Handling multi-target conflicts. For example, when the visual perception module detects both weeds and people simultaneously, the system determines that people are the highest priority stop, and will forcibly override the flame command, executing the command. , Safety protection actions.

[0050] Obstacle Avoidance Path: When an obstacle or a crop that needs protection (such as melon seedlings) is detected, the system enters an Avoid state, adjusting its linear velocity accordingly. and angular velocity Plan a smooth obstacle avoidance curve.

[0051] Implementation of Mulch Film Protection and Thermal Safety Self-Diagnosis

[0052] The system has a dedicated protection logic for the plastic film commonly laid in greenhouses.

[0053] Distance compensation: The vision system calculates the lateral offset between the edge of the plastic film and the center line of the prototype in real time. .like Entering the danger threshold The system automatically calculates the dynamic safety distance. And fine-tune the yaw rate Keep the flame-spraying center away from the mulch film to prevent heat radiation from melting the mulch film.

[0054] Thermal safety monitoring: Based on ANSYS instantaneous thermal analysis results, the system continuously monitors the temperature of the nozzle body. Once it detects that prolonged continuous operation has caused the body temperature to approach the critical 900℃ range, the system will automatically switch to pulse jet mode or reduce the operating intensity, executing forced thermal protection logic to prevent damage to prototype components or fire risks.

[0055] Typical Work Process Description

[0056] In a typical melon greenhouse operation:

[0057] A. The system is cruising normally between lines (linear speed approximately 0.4 m / s).

[0058] B. The vision module identifies areas with weeds and soil-borne disease risks within 0.5m ahead;

[0059] C. The decision module calculates the required disinfection depth, Reduce to 0.2 m / s and activate the flame. ;

[0060] D. If any staff member enters the detection area at this time, the system will immediately recognize the entry and trigger a Stop command to cut off the gas supply and apply the brakes.

[0061] E. Once the risk is eliminated, the system resumes operation and automatically compensates for the travel path based on the amount of film offset, completing the refined disinfection task.

[0062] The above description is only a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. Any equivalent modifications or changes made by those skilled in the art based on the content disclosed in the present invention should be included within the scope of protection set forth in the claims.

Claims

1. A semantically perceptive-driven precision flame soil disinfection system for greenhouse horticulture, characterized in that, include: The multi-source vision perception module is configured to acquire color images, depth maps, and point cloud data of the working environment through a binocular camera. This data is used to identify weeds, crops, mulch film, obstacles, and personnel targets within the greenhouse, and to output the target category. and relative position vector ; The edge computing decision module, equipped with a lightweight deep learning model, is configured to receive the target category. and relative position vector Based on this, motion control commands and flamethrower execution commands are generated in real time; The flame actuation module includes a flame nozzle array and a gas control unit, wherein the flame actuation module is in flame-emitting state. Controlled by the flamethrower execution command output by the edge computing decision module; The tracked mobile chassis adopts a tracked walking mechanism with independent left and right drive, and is configured to perform path avoidance and inter-row cruise based on the motion control commands output by the edge computing decision module and the equivalent differential wheel kinematic model.

2. The semantic-aware driven precision flame soil disinfection system for facility horticulture greenhouses according to claim 1, characterized in that, The edge computing decision module incorporates a heat penetration depth control algorithm for soil pathogen inactivation; the edge computing decision module is configured to adjust the linear speed of the tracked mobile chassis based on the risk of soil-borne diseases in the area to be disinfected. To control the duration of flame exposure on a unit area of ​​soil This allows for disinfection at the predetermined depth. instantaneous soil temperature The temperature threshold required for pathogen inactivation is met; wherein, the instantaneous soil temperature According to the formula Sure, The soil surface equivalent temperature is positively correlated with the output power of the flame actuator module. For soil thermal diffusivity, This is the error compensation function.

3. A semantically perceptive-driven precision flame soil disinfection system for facility horticulture greenhouses according to claim 2, characterized in that, The edge computing decision module is configured to determine the linear velocity of the tracked mobile chassis. The velocity should be controlled within the range of 0.2 m / s to 0.4 m / s to ensure the effective heat input required for pathogen inactivation.

4. A semantically perceptive driven precision flame soil disinfection system for facility horticulture greenhouses according to claim 1, characterized in that, The lightweight deep learning model is built on an improved YOLOv8n architecture, with 3.01M model parameters. The average inference time for a single image on the edge computing platform to which the edge computing decision module belongs is 2.6ms.

5. A semantically perceptive driven precision flame soil disinfection system for facility horticulture greenhouses according to claim 4, characterized in that, The lightweight deep learning model achieved a precision of 0.955 for identifying weeds, crops, and plastic film targets in the greenhouse, and a mean precision (mAP@0.5) of 0.

902.

6. A semantically perceptive driven precision flame soil disinfection system for facility horticulture greenhouses according to claim 1, characterized in that, The edge computing decision module is configured to use a nonlinear decision mapping operator. Transforming perceived information into execution instructions satisfies the mapping relationship: ,in, For the first The perception tensor of each target is a priority operator based on the target category.

7. A semantically perceptive driven precision flame soil disinfection system for facility horticulture greenhouses according to claim 6, characterized in that, The nonlinear decision mapping operator It is configured to: when the multi-source visual perception module identifies a human target, force the output of an instruction to stop the tracked mobile chassis and shut down the flame execution module's flame-spraying state, and this instruction has a higher priority than the response instruction to weeds, crops, mulch film or obstacle targets.

8. A semantically perceptive driven precision flame soil disinfection system for facility horticulture greenhouses according to claim 1, characterized in that, The edge computing decision module is also configured to perform an automatic compensation function for the thermal radiation safety distance of the mulch film; the edge computing decision module determines the lateral offset of the mulch film edge relative to the system's centerline based on the information identified by the multi-source visual perception module. Real-time calculation of dynamic safety distance And enter the preset obstacle avoidance threshold at the edge of the mulch film. Automatically correct yaw rate .

9. A semantically perceptive driven precision flame soil disinfection system for facility horticulture greenhouses according to claim 8, characterized in that, The edge computing decision module is also configured to: enter a preset obstacle avoidance threshold at the edge of the plastic film. And the dynamic safety distance When the temperature drops below the minimum safety threshold, the flame execution module's flame emission state is turned off.

10. A semantically perceptive driven precision flame soil disinfection system for facility horticulture greenhouses according to claim 1, characterized in that, The flame execution module also includes a temperature sensor for real-time monitoring of the main body temperature of the flame nozzle array; the edge computing decision module has built-in thermal safety self-diagnosis logic and is configured to automatically execute a deceleration cooling action or switch to pulse flame mode when the temperature sensor detects that the main body temperature is approaching a preset critical risk area of ​​900℃; the discrete update process of the equivalent differential wheel kinematic model satisfies: ， in, , , They represent the first Chassis position coordinates and heading angle at each sampling time Linear velocity, Angular velocity, The sampling period; the edge computing decision module is configured to adjust... and Control the tracked mobile chassis to move smoothly along the avoidance path.

Images