Method and equipment for removing poisonous grass based on computer vision

Abstract

The invention discloses a method and equipment for removing poisonous grass based on computer vision, and relates to the field of meadow management, and the method comprises the steps: collecting and preprocessing different types of poisonous grass image data, and building a poisonous grass data set; constructing a YOLOv9 model and performing training and testing by using the poisonous weed data set to obtain a poisonous weed recognition model; image information, positioning information and environment information of all positions of the meadow are obtained, a poisonous weed recognition model is used for recognizing poisonous weed information in the image information, the three-dimensional position of each poisonous weed is determined according to the positioning information, and then a spatial distribution diagram is generated according to the three-dimensional position of each poisonous weed and the environment information of the corresponding position; based on the distribution characteristics and the biological characteristics of the toxicant grass at each position in the spatial distribution diagram, performing toxicant grass treatment according to a differential treatment strategy; and determining a high-risk area in the spatial distribution diagram by using a toxic weed diffusion model, and taking prevention measures. According to the invention, accurate identification, effective processing and diffusion management of the poisonous grass are realized, so that the production benefit of a pasture is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of grassland management, and more particularly to a method and device for removing poisonous weeds based on computer vision. Background Art

[0002] Pasture is an important resource for animal husbandry, providing cattle and sheep with rich forage and nutrients such as protein, fat, calcium, phosphorus, etc., which directly affect the growth rate, health level and production efficiency of livestock. High-quality pasture can not only promote the healthy growth of livestock, but also reduce breeding costs, increase production and product quality, thereby improving the overall economic benefits of the pasture. However, the presence of poisonous weeds poses a serious threat to pasture management, not only causing hidden dangers to the health of cattle and sheep, but also having a negative impact on economic benefits.

[0003] At present, conventional methods of removing poisonous weeds mainly rely on manual weeding and mechanical weeding. Although manual weeding can effectively eradicate poisonous weeds, it is labor-intensive, inefficient and costly, and is only suitable for small areas. Mechanical weeding, such as lawn mowers, is suitable for large-scale operations, but can only remove the above-ground part of poisonous weeds and cannot completely eradicate them, which can easily lead to the regrowth of poisonous weeds. The maintenance cost of mechanical equipment is also high, and the technical requirements for operators are relatively strict. Traditional weeding methods are difficult to achieve thorough and efficient removal of poisonous weeds in large areas of grassland, and more advanced technologies are urgently needed to make up for these shortcomings. Summary of the invention

[0004] In view of this, the present invention provides a method and device for removing poisonous weeds based on computer vision, thereby solving the problems of low efficiency and easy recurrence of traditional methods of removing poisonous weeds and improving the efficiency of removing poisonous weeds.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] The present invention discloses a method for removing poisonous weeds based on computer vision, and the specific steps are as follows:

[0007] Collect and preprocess image data of different types of poisonous weeds to establish a poisonous weed dataset;

[0008] Constructing a YOLOv9 model and using the poisonous weed dataset for training and testing to obtain a poisonous weed recognition model;

[0009] Acquire image information, positioning information and environmental information of each position of the grassland, use the poisonous weed recognition model to identify the poisonous weed information in the image information, and determine the three-dimensional position of each poisonous weed according to the positioning information, and then generate a spatial distribution map according to the three-dimensional position of the poisonous weed and the environmental information of the corresponding position;

[0010] Based on the distribution characteristics and biological characteristics of poisonous weeds at each location in the spatial distribution map, poisonous weeds are treated according to differentiated treatment strategies; high-risk areas in the spatial distribution map are determined using a poisonous weed diffusion model, and preventive measures are taken.

[0011] Furthermore, the collecting of different types of poisonous weed image data specifically includes: selecting the types of poisonous weeds that need to be identified and controlled; selecting different time and lighting shooting conditions according to the types of poisonous weeds, and shooting visible light images and infrared images of the poisonous weeds from multiple different angles;

[0012] The preprocessing includes: performing geometric correction on the visible light image and the infrared image to ensure that the two are spatially aligned, applying a filtering algorithm to remove noise in the image, and performing edge detection and contrast enhancement on the image; using a method based on a self-supervised transformer to fuse the visible light image and the infrared image; and performing data annotation on the fused image to obtain the poisonous weed data set.

[0013] Furthermore, the method based on the self-supervised transformer combines the convolutional neural network and the transformer model to efficiently fuse the visible light image and the infrared image of the poisonous weed, including: independent encoders extract the features of the two images respectively, the fusion layer fuses the features by enhancing the multimodal synergy effect through high-order spatial and channel interactions, and then the decoder restores the fused features into a fused image.

[0014] Furthermore, the image information includes: visible light image, infrared image, three-axis shooting angle and focal length of the poisonous weed;

[0015] The positioning information includes: GPS positioning data of the shooting location and laser radar three-dimensional point cloud data;

[0016] The environmental information includes: soil moisture, light intensity, temperature, humidity, and precipitation;

[0017] The poisonous weed information includes: poisonous weed type, size, and state.

[0018] Furthermore, the distribution characteristics include concentrated continuous distribution and sporadic distribution, which are determined according to the density threshold of poisonous weeds in the area; the biological characteristics include root depth and growth years; the differentiated treatment strategy specifically includes:

[0019] Concentrated and continuous poisonous weeds: Cover large areas of poisonous weeds with pesticides and use laser technology to establish isolation zones to prevent spread; determine the amount of pesticides and spraying paths based on the density and distribution of poisonous weeds;

[0020] Scattered poisonous weeds: precise cutting with laser transmitters, or removal with mechanical claws and local spraying;

[0021] Toxic weeds with deep roots: comprehensive treatment through mechanical claw removal, deep spraying and heat treatment;

[0022] Poisonous weeds with shallow roots: Use mechanical claws to remove them directly.

[0023] Furthermore, the poisonous weed diffusion model is obtained by:

[0024] Obtain historical data on the growth of various types of poisonous weeds, the historical data including the type, density, location, removal time and environmental information of the poisonous weeds; establish a machine learning model and a data regression model, use the historical data for data regression and training, and obtain a poisonous weed diffusion model.

[0025] Furthermore, the preventive measures include: establishing isolation zones in the high-risk areas by physical or chemical means, regularly testing the isolation effects of the isolation zones, and performing secondary weeding operations inside and outside the isolation zones.

[0026] The present invention also discloses a poisonous weed removal device based on computer vision, comprising: a quadruped robot chassis, a mechanical arm module, a sensor module, and a layered combined central processing unit; the mechanical arm module, the sensor module, and the layered combined central processing unit are all fixedly mounted on the quadruped robot chassis;

[0027] The robotic arm module includes: a laser emitting robotic arm, a pesticide spraying robotic arm, a robotic claw, a liftable robotic arm platform, and a liftable working platform;

[0028] The sensor module includes: a visible light camera, a thermal infrared camera, a laser radar, a temperature sensor, a humidity sensor, a rainfall sensor, and a GPS module;

[0029] The layered combined central processor includes a real-time control layer, an AI processing layer, and a communication and recording layer; the real-time control layer obtains positioning information based on the data uploaded by the laser radar and the GPS module, and performs operation control; the AI ​​processing layer performs image processing and AI recognition on the data uploaded by the visible light camera and the thermal infrared camera, determines the weeding strategy executed by the robot and controls the execution of the robotic arm module; the communication and recording layer is used to communicate and record data with the cloud platform, as well as process environmental information collection and generate spatial distribution maps.

[0030] Furthermore, the liftable work platform is fixedly mounted on the upper part of the chassis of the quadruped robot, and the mechanical claw is fixedly mounted on the liftable work platform, so that the mechanical claw moves up and down when the liftable work platform moves;

[0031] The lifting robot arm platform is fixedly mounted on the lower part of the quadruped robot chassis, and the pesticide spraying robot arm and the laser emitting robot arm are both fixedly mounted on the bottom of the lifting robot arm platform, so that the pesticide spraying robot arm and the laser emitting robot arm move up and down when the lifting robot arm platform moves;

[0032] The laser radar is fixedly mounted on the front end of the chassis of the quadruped robot through a laser radar bracket; the visible light camera and the thermal infrared camera are both mounted on the chassis of the quadruped robot through slide rails.

[0033] Furthermore, the quadruped robot chassis includes a fuselage, a drive motor and four legs that can move independently. The fuselage is composed of an upper cover, a lower cover and a baffle to form an overall frame to provide a supporting structure. The drive motor is fixedly installed on the fuselage through a motor base and a flange. The drive motor drives the legs to move through a connecting rod.

[0034] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method and equipment for removing poisonous weeds based on computer vision, which not only overcomes the shortcomings of low efficiency and easy recurrence of poisonous weeds in traditional methods, but also significantly saves labor costs and improves operational efficiency. Through the combination of intelligent operating system, flexible operation platform and multifunctional mechanical arm, the robot can realize accurate identification, effective treatment and diffusion management of poisonous weeds, providing a modern and efficient solution for grassland management, thereby improving the production efficiency and environmental protection level of pastures. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0036] Figure 1 It is a schematic diagram of the overall process of an embodiment of the present invention.

[0037] Figure 2 It is a front view of the poisonous weed removal equipment according to an embodiment of the present invention.

[0038] Figure 3 This is a front view of the chassis of the quadruped robot according to an embodiment of the present invention.

[0039] Figure 4 Schematic diagram of a liftable work platform according to an embodiment of the present invention.

[0040] Figure 5 Schematic diagram of a lifting robot arm platform according to an embodiment of the present invention.

[0041] Figure 6 Schematic diagram of a robotic arm according to an embodiment of the present invention.

[0042] Figure 7 Schematic diagram of a mechanical claw according to an embodiment of the present invention.

[0043] Figure 8 Schematic diagram of the chassis and legs of a quadruped robot according to an embodiment of the present invention.

[0044] In the figure, 1. chassis of the quadruped robot; 2. power structure of the legs; 3. platform of the lifting mechanical arm; 4. lifting working platform; 5. storage box for poisonous weeds; 101. upper cover of the fuselage; 102. lower cover of the fuselage; 103. outer baffle of the fuselage; 104. first inner baffle of the fuselage; 105. second inner baffle of the fuselage; 201. outer motor; 202. motor flange; 203. connecting rod; 204. inner motor; 205. motor protective shell; 206. motor base; 207. first thigh connecting rod ; 208, second thigh link; 209, rear thigh link; 210, calf link; 211, mechanical foot; 301, control box; 302, central lifting column; 303, horizontal bracket; 304, laser emitting mechanical arm; 305, pesticide spraying mechanical arm; 306, laser emitting head; 307, pesticide spraying head; 401, bottom control panel; 402, upper lifting panel; 403, circular track; 404, movable round rod; 405, connecting block; 406, three-ring connecting rod. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] The embodiment of the present invention discloses a method for removing poisonous weeds based on computer vision, such as Figure 1 As shown, the specific steps are as follows:

[0047] Collect and preprocess image data of different types of poisonous weeds to establish a poisonous weed dataset;

[0048] Build a YOLOv9 model and use the poisonous weed dataset for training and testing to obtain a poisonous weed recognition model;

[0049] Obtain image information, positioning information and environmental information of each position of the grassland, use a poisonous weed recognition model to identify the poisonous weed information in the image information, and determine the three-dimensional position of each poisonous weed according to the positioning information, and then generate a spatial distribution map according to the three-dimensional position of the poisonous weed and the environmental information of the corresponding position;

[0050] Based on the distribution characteristics and biological characteristics of poisonous weeds at each location in the spatial distribution map, poisonous weeds are treated according to differentiated treatment strategies; the poisonous weed diffusion model is used to determine the high-risk areas in the spatial distribution map, and preventive measures are taken.

[0051] In a specific embodiment, collecting image data of different types of poisonous weeds specifically includes: selecting the types of poisonous weeds that need to be identified and controlled; selecting different time and lighting shooting conditions according to the types of poisonous weeds, and shooting visible light images and infrared images of the poisonous weeds from multiple different angles;

[0052] Specifically, first, we selected the types of poisonous weeds that needed to be identified and controlled. Based on the recommendations of agricultural experts, we identified several poisonous weeds that were most harmful to crops as the main targets for data collection.

[0053] Then, data collection standards were formulated, and images were collected using high-resolution digital cameras and thermal infrared cameras. The best time for collection was spring and summer, when poisonous weeds grew vigorously. The collection location was a real farmland scene, ensuring that there were enough samples under different lighting conditions and weather conditions. A batch of data was collected in the farmland scene every two hours to ensure the diversity and representativeness of the samples.

[0054] Finally, the image should be taken from multiple angles to ensure that the poisonous weeds are fully captured and their clarity and center position in the image are guaranteed, and to ensure the synchronization of visible light images and thermal infrared images, so as to facilitate subsequent image registration and fusion. At the same time, in order to increase the diversity of data, different lighting conditions and background environments should be covered during the collection.

[0055] Preprocessing includes: geometric correction of visible light images and infrared images to ensure that the two are spatially aligned, application of filtering algorithms to remove noise in images, and edge detection and contrast enhancement of images; using a method based on self-supervised transformers to fuse visible light images and infrared images; and data annotation of the fused images to obtain a poisonous weed dataset.

[0056] Specifically, the visible light image and the thermal infrared image are geometrically calibrated to ensure that they are aligned in space. A filtering algorithm is applied to remove noise from the image to improve image quality, and contrast enhancement and edge detection are performed on the image to improve image details and edge information.

[0057] The method based on self-supervised transformer (STFNet) can efficiently fuse visible light images and thermal infrared images of poisonous weeds by combining convolutional neural network (CNN) and transformer models. First, independent encoders extract features of the two images respectively, and the fusion layer enhances the multimodal synergy effect through high-order spatial and channel interactions. Then the decoder restores the fused features into a fused image. This method uses a fusion loss function that combines pixel-level loss and structural similarity (SSIM) loss to ensure that the fused image provides higher visual quality and accurate poisonous weed identification while retaining the details and structural information of the input image.

[0058] The LabelImg tool was used to annotate the data in VOC format. The output was an XML file that was converted into the txt format used by YOLO through Python code. The annotated categories included the detailed classification of each poisonous weed. When annotating, make sure that the selected area accurately covers the main part of the poisonous weed and exclude blurry or poor quality images.

[0059] The dataset is divided into training set, validation set and test set in a ratio of 8:1:1. The training set is used for model training, the validation set is used to adjust model parameters and prevent overfitting during training, and the test set is used to finally evaluate the performance of the model on unseen data. Through reasonable division, the reliability and effectiveness of the model in practical applications are ensured.

[0060] In a specific embodiment, a method based on a self-supervised transformer combines a convolutional neural network and a transformer model to efficiently fuse visible light images and infrared images of poisonous weeds, including: independent encoders extract features of the two images respectively, a fusion layer fuses features by enhancing multimodal synergy effects through high-order spatial and channel interactions, and then a decoder restores the fused features into a fused image.

[0061] In a specific embodiment, YOLOv9 is selected as the target detection network because of its high efficiency and accuracy, which is suitable for poisonous weed detection. The image is preprocessed using methods such as scale transformation, data enhancement, normalization, denoising and smoothing, image occlusion, multi-image combination, histogram equalization, image cutting, local area interest, etc. In model training, Windows 11 operating system, Python 3.9 and Torch 2.0 training environment are used. The training is set to 1200 epochs, breakpoint continuation training is adopted, the optimizer uses AdamW, the initial learning rate is 0.001, and the cosine annealing scheduler is used for adjustment. The batch size is set to 32, and mAP50 (average accuracy at 50% intersection-over-combination threshold) is calculated after each training. The preservation strategy is to retain the latest training and the model with the highest mAP50 value, and finally use the model with the highest mAP50 value.

[0062] In a specific embodiment, the image information includes: a visible light image, an infrared image, a three-axis shooting angle and a focal length of the poisonous weed;

[0063] Positioning information includes: GPS positioning data of the shooting location and LiDAR 3D point cloud data;

[0064] Environmental information includes: soil moisture, light intensity, temperature, humidity, and precipitation;

[0065] The poisonous weed information includes: the type, size and status of the poisonous weeds.

[0066] In a specific embodiment, the distribution characteristics include concentrated continuous distribution and sporadic distribution, which are determined according to the poisonous weed density threshold in the area; the biological characteristics include root depth and growth years; and the differentiated treatment strategies specifically include:

[0067] Concentrated and continuous poisonous weeds: Cover large areas of poisonous weeds with pesticides and use laser technology to establish isolation zones to prevent spread; determine the amount of pesticides and spraying paths based on the density and distribution of poisonous weeds;

[0068] Scattered poisonous weeds: precise cutting with laser transmitters, or removal with mechanical claws and local spraying;

[0069] Toxic weeds with deep roots: comprehensive treatment through mechanical claw removal, deep spraying and heat treatment;

[0070] Poisonous weeds with shallow roots: Use mechanical claws to remove them directly.

[0071] Specifically, for the concentrated and continuous poisonous weed strategy, when the robot identifies that the poisonous weeds are concentrated and continuous, the focus of the treatment method is to efficiently cover a large area and ensure the uniform distribution of the agent to prevent the poisonous weeds from continuing to spread. Once the images collected by the robot camera are processed and the areas where the poisonous weeds are concentrated are identified, the system first generates a spatial distribution map of the poisonous weeds in the area. Using the three-dimensional point cloud data generated by a lidar (such as the Velodyne Puck VLP-16), the system can accurately mark the boundaries, area, and density of each piece of poisonous weeds. Through image fusion technology, combined with visible light and thermal infrared images, the health status, size, and location of the poisonous weeds can be further confirmed. The poisonous weed distribution map can provide precise guidance for subsequent spraying operations. After the poisonous weeds are identified, when the poisonous weed spatial distribution map is generated, the poisonous weed density D can be calculated as an indicator for evaluating the regional concentration. The formula is: Where D is the density of poisonous weeds, expressed in the number of plants per square meter, n is the number of poisonous weeds, and A is the area of ​​the area (square meters). The formula helps the robot assess the concentration of poisonous weeds and decide whether to choose to spray pesticides as the main treatment method.

[0072] For concentrated and continuous poisonous weeds, the robot prefers to spray pesticides. In order to ensure that the pesticide can evenly cover the entire poisonous weed area, the system will analyze the density and distribution characteristics of the poisonous weeds, calculate the required amount of pesticides, and formulate a spraying path. The spraying path generates the optimal route through the Dijkstra algorithm to ensure that the robot can cover the entire poisonous weed area efficiently and without omission. In order to ensure the effective distribution of pesticides, the dosage V can be calculated based on the density D of the poisonous weeds and the area A of the area, the formula is: V = k × D × A; where V represents the total amount of pesticides required to be sprayed (liters), k is the applicable coefficient of the pesticide, which is adjusted according to the type, size and recommended amount of pesticides of the poisonous weeds, and D represents the density of the poisonous weeds, in units of the number of plants / square meter. A represents the area of ​​the spraying area (square meters). The robot will calculate the dosage of the pesticide in real time according to the distribution of the poisonous weeds to ensure that the pesticide is not wasted and the poisonous weeds can be accurately covered. The specific operation of pesticide spraying is as follows: the pesticide spraying device on the robot arm will move slowly according to the predetermined path to ensure that every piece of poisonous weed can be sprayed evenly. The system will monitor the location and amount of each spraying in real time to ensure that there will be no over- or under-application of pesticide. A high-precision pressure nozzle is used to ensure that the pesticide can be evenly distributed on the surface and roots of the poisonous weeds in the form of fine mist. This spraying method helps the pesticide to be absorbed quickly and increases the weed control effect.

[0073] In order to prevent the further spread of poisonous weeds, the system will also establish isolation belts along the edges of the poisonous weed areas. The isolation belts are generated by laser emission devices. The laser will cut or burn the surrounding weeds along the edges of the poisonous weed dense areas to form a "poisonous weed-free" area to prevent the poisonous weeds from spreading to other areas. When the laser equipment is used to establish isolation belts, the laser power P should be adjusted according to the burning temperature of the poisonous weeds and the soil moisture. The formula is: Where: P represents the power of the current laser device, P 0 is the base power, the default power for dry soil, H s Indicates the current soil moisture, H 0 The formula ensures that the laser equipment can work under different soil moisture conditions and adjusts the power to ensure the cutting effect. The specific steps of the laser isolation operation are: the laser transmitter accurately cuts or burns along the edge according to the poisonous weed boundary data. By real-time monitoring of soil temperature and poisonous weed burning conditions, the system can accurately control the intensity and duration of the laser to avoid damage to normal crops around it. The width of the isolation belt is adjusted according to the diffusion speed and density of the poisonous weeds, and is usually set to a width of 5-10 cm to ensure that the roots of the poisonous weeds will not easily cross the isolation belt.

[0074] After spraying and establishing isolation belts, the robot will continue to monitor changes in the area. Sensors are used to monitor the growth status of poisonous weeds, the effect of pesticides, and the mortality rate of poisonous weeds. For poisonous weeds that still persist, the system can set a secondary spraying plan to further consolidate the weed control effect. After spraying, the robot needs to continuously monitor the effect of the pesticide. The spraying effect can be evaluated by the pesticide coverage rate C and the poisonous weed mortality rate M. The specific formula is: M = f(C, T); where M represents the mortality rate of poisonous weeds in percentage, C is the coverage rate of pesticides, ranging from 0 to 1, and T is the effective time of pesticides in hours. By monitoring the values ​​of C and T, the robot can determine whether secondary spraying is needed.

[0075] For the strategy of sporadic distribution of poisonous weeds, when the poisonous weeds are distributed in a sporadic manner, the focus of the treatment is precise and localized weeding operations to avoid unnecessary waste of pesticides and reduce the impact on surrounding crops. The robot first determines the specific location and three-dimensional coordinates of each poisonous weed one by one through image recognition and lidar positioning. The system will analyze the location and surrounding environment of the poisonous weeds to determine whether it is suitable for spraying or whether to directly perform mechanical removal. Usually, when the poisonous weeds are scattered and close to the crops, the system will tend to choose physical means to remove the poisonous weeds to reduce the impact of pesticides on crops. In the sporadic distribution processing, the system first uses lidar and image recognition technology to determine the three-dimensional coordinates (x, y, z) of the poisonous weeds and perform precise spatial positioning through three-dimensional point cloud data. The three-dimensional coordinates of the lidar are calculated by the following formula: Among them, x and y are the coordinates of the poisonous weed on the two-dimensional plane. d is the distance between the robot and the target poisonous weed. z is the height of the poisonous weed. This formula helps the robot determine the precise spatial position of each poisonous weed and provides basic data for subsequent operations.

[0076] For poisonous weeds that are difficult to remove directly or are in a remote location, the system will activate the laser transmitter on the robotic arm for precise cutting. Laser cutting technology can directly cut off the stems of poisonous weeds or burn their leaves without damaging surrounding crops. The temperature and intensity of the cutting process are automatically adjusted by the system according to the type of poisonous weed to ensure the accuracy and efficiency of the cutting. The power P of the laser cutting also needs to be determined according to the volume V of the poisonous weed and the cutting depth d. c Adjustment, the specific formula is: P = k 1 ×V×d c ; Where P represents the power of the laser during cutting (watts), V represents the volume of the poisonous weed in cubic centimeters, and d c k is the laser cutting depth in centimeters. 1is the laser intensity coefficient set according to the material of the poisonous weed. Through this formula, the system can adjust the laser power in real time according to the volume and cutting depth of the poisonous weed to ensure accurate cutting. The specific steps of the laser operation are as follows: the robotic arm accurately aligns the position of the poisonous weed according to the data of the laser radar, the laser transmitter monitors the burning status of the target through thermal imaging, and adjusts the power and duration of the laser in real time to ensure that the poisonous weed is completely cut. After the cutting is completed, the system will detect the state of the roots of the poisonous weed and decide whether further spraying of pesticides is needed to prevent recurrence.

[0077] For poisonous weeds with shallow roots or easy to pull out, the robot will first use the mechanical claw to pull them out one by one. The mechanical claw accurately locates the roots of the poisonous weeds through three-dimensional coordinate data, and gradually applies force to pull out the poisonous weeds. The pulling force F of the mechanical claw should be based on the root depth d of the poisonous weeds. r and the root width of the poisonous weeds w r Calculate: F = k 2 ×d r × r ; where F is the force required to remove the poisonous weeds, in Newtons, d r is the root depth of the poisonous weeds, in centimeters, w r is the root width in centimeters, k 2 is the soil hardness coefficient, which depends on the softness of the soil. The formula ensures that the force applied by the mechanical claw is sufficient to remove the poisonous weeds while avoiding unnecessary damage to the soil. The specific steps of the removal operation are as follows: the mechanical arm moves above the poisonous weed, the mechanical claw opens and slowly clamps the stem or root of the poisonous weed, ensuring that enough root area is captured to prevent the root system from remaining in the soil. The mechanical claw applies force and automatically adjusts the speed and force of the removal according to the root depth of the poisonous weed to ensure that the removal process does not damage the surrounding soil structure.

[0078] After the poisonous weeds are pulled out, if the system detects that the roots of the poisonous weeds are relatively developed or difficult to completely remove, the robot will spray the pesticide precisely in that area. The spraying is only targeted at the area where the poisonous weeds are pulled out to reduce the impact on the surrounding environment. For poisonous weeds that are still stubborn after being pulled out or cut, the robot needs to spray precisely. The amount of pesticide required V s According to the root area of ​​the poisonous weeds A r and root residual R r Calculation formula: V s =k 3 ×A r ×R r ; Where: V s Indicates the amount of pesticide sprayed, in milliliters, A r is the area of ​​the root in square centimeters, R r is the root residue, which indicates the part of the poisonous weed root that has not been completely removed, ranging from 0 to 1, k 3is the utility coefficient of the pesticide. This formula can help the robot intelligently control the amount of pesticide used, reduce the waste of pesticides and ensure that the roots are thoroughly treated. The spraying steps are as follows: the robot uses the pesticide spraying device to spray at a fixed point. The nozzle will adjust the spraying range according to the size of the identified poisonous weed roots to ensure that the pesticide evenly covers the root area. During the spraying process, the robot uses sensors to detect the spraying coverage and the amount of pesticide used in real time to ensure accurate control of the amount of pesticide and avoid waste.

[0079] For strategies against deep-rooted noxious weeds, perennial noxious weeds usually have well-developed root systems and strong vitality, and require a comprehensive approach. The strategy for treating perennial noxious weeds combines physical removal and chemical spraying. When calculating root depth and root complexity, the following formula can be used: F r =k 4 ×d r × r ; Among them, F r The force required to pull out perennial noxious weeds is measured in Newtons, k 4 is the root system coefficient of perennial poisonous weeds, reflecting the degree of development of their root system. If the root system of perennial poisonous weeds is complex, the calculated force F r If the weeds are large, the robot will prioritize the use of a combination of spraying and pesticides rather than simple physical removal.

[0080] For this type of poisonous weeds, the system usually uses a combination of mechanical removal and spraying. First, the above-ground part of the poisonous weeds is removed by mechanical claws, and then the roots are sprayed to ensure that the poisonous weeds cannot grow again; the system intelligently adjusts the order and frequency of removal and spraying according to the root depth and growth rate of the poisonous weeds; secondary monitoring after spraying: Since perennial poisonous weeds are more likely to recur, the system will regularly monitor the area after spraying and perform secondary treatment according to the regeneration of the poisonous weeds.

[0081] For the strategy of poisonous weeds with shallow root system, the root system of annual poisonous weeds is shallow and their vitality is relatively weak. They can be directly treated by physical removal or laser cutting, and usually no additional spraying is required. a According to the root depth d r and soil softness coefficient k 5 Calculate: F a =k 5 ×d r; Since the roots of annual poisonous weeds are shallow, they can usually be removed without much force, and the need for spraying is also low. For simpler poisonous weeds, the system directly uses mechanical claws to remove them to ensure that the roots are completely removed. The system will scan the area after removal to confirm whether further treatment is required. If the poisonous weeds are sparsely distributed, the robot will give priority to laser cutting, using the laser transmitter to quickly cut the stems of the poisonous weeds to ensure that they cannot continue to grow.

[0082] In a specific embodiment, the poisonous weed diffusion model is obtained by:

[0083] Obtain historical data on the growth of various types of poisonous weeds, including the types, density, location, removal time, and environmental information of the poisonous weeds; establish machine learning models and data regression models, use historical data for data regression and training, and obtain a poisonous weed diffusion model. Data source: The data source used in the model covers multidimensional data such as the distribution data of poisonous weeds, meteorological information (such as temperature, humidity, precipitation), topographic features, and soil properties (such as pH value, water content, nutrient content). These variables are cleaned and normalized, and missing values ​​are supplemented by interpolation to ensure data consistency and model input quality.

[0084] Model selection and construction: According to the needs of poisonous weed spread prediction, a combination of decision tree model and multiple linear regression model was used. The decision tree model (DecisionTree) is used to classify and identify the spread pattern of poisonous weeds, helping to identify the spread trend of different poisonous weed types under various environmental conditions; the multiple linear regression model (Multiple LinearRegression) is used to quantify the relationship between the spread rate of poisonous weeds and environmental variables, helping to calculate the spread rate under different conditions.

[0085] Model training and parameter optimization: The machine learning model uses multiple cross-validations (k-fold cross-validation) to optimize the generalization ability of the model. In the decision tree model, information gain and Gini coefficient are used as splitting criteria to split the data set layer by layer according to environmental characteristics to generate the diffusion patterns of different types of poisonous weeds. The regression model estimates the regression coefficient through the ordinary least squares (OLS) method to optimize the prediction accuracy of the spread rate of poisonous weeds. For the multivariate linear regression model, the diffusion rate formula is:

[0086]

[0087] Among them, y is the spreading speed of poisonous weeds, x i For each environment variable, β i is the regression coefficient, which indicates the weight of the environmental variables on the spread rate of poisonous weeds, β0 The intercept represents the baseline value of the spread rate of noxious weeds.

[0088] Model prediction and output: The output of the machine learning model is mainly the diffusion pattern category of poisonous weeds, showing the diffusion trend of various types of poisonous weeds under different combinations of environmental variables; the regression model outputs the diffusion speed of poisonous weeds under specific environmental conditions and obtains the diffusion direction through vector calculation. By comprehensively analyzing the outputs of both, this patent achieves accurate prediction of the diffusion path, speed and direction of poisonous weeds.

[0089] Specifically, the evaluation of the spatial distribution of poisonous weeds needs to rely on the environmental information and poisonous weed data automatically recorded by the robot during the weeding process. The acquisition of this information is completed by a series of sensors, image processing systems and GPS positioning modules. When the robot performs the weeding task, it automatically records the specific location, type, size and surrounding environmental conditions of each poisonous weed. The specific process is: the robot is equipped with a variety of sensors (such as optical sensors, temperature sensors, humidity sensors, etc.) and GPS modules to monitor the geographical location, soil moisture, light intensity, ambient temperature and other environmental information of the poisonous weeds in real time. Combining visible light cameras and thermal infrared cameras, the robot can automatically identify the type and size of poisonous weeds and record the precise geographical coordinates of the poisonous weeds through GPS. Poisonous weed location information: GPS will record the longitude, latitude, altitude and other data of the poisonous weeds. Whenever the robot completes a poisonous weed removal task, the system will store the location information of the poisonous weeds in the central database to ensure the accurate positioning and subsequent analysis of the poisonous weeds. Environmental information collection: The surrounding environmental information, such as soil moisture, light intensity, temperature, etc., will also be synchronously collected and stored by sensors. These data will provide key references for subsequent diffusion analysis and decision-making.

[0090] The robot will visualize all recorded location information and environmental data through the data processing module to generate a spatial distribution map of poisonous weeds. In the distribution map, each area will be marked with the distribution density of different types of poisonous weeds, the size of the poisonous weeds, and the surrounding environmental conditions. The generation of the poisonous weed distribution map is dynamic and real-time. The robot can update the distribution map according to each weeding task and adjust the subsequent treatment plan at any time. Representation of poisonous weed density: The distribution density of poisonous weeds will be displayed on the distribution map through color gradients or density maps. The darker the color, the more concentrated and continuous the growth of poisonous weeds, while the lighter the color, the more scattered the distribution. Such a graphical representation can help the robot identify the key areas for weeding and effectively manage weeding resources.

[0091] Through the poisonous weed distribution map, the robot can not only grasp the current distribution of poisonous weeds, but also use time series data to analyze the spread trend of poisonous weeds. Based on the historical poisonous weed growth data, the system can analyze the spread direction and speed of poisonous weeds, predict high-risk areas that may appear in the future, and take timely measures to deal with them. Diffusion direction assessment: The robot will combine historical environmental changes and poisonous weed growth data to predict the direction in which poisonous weeds will spread. If the soil moisture in certain areas is high, poisonous weeds may spread to these areas. The distribution map will show the possible paths of the spread of poisonous weeds and mark the direction of spread with arrows. Diffusion speed analysis: By comparing the current and historical growth rates of poisonous weeds, the system can estimate the speed of poisonous weed spread and use this data to further adjust the weed control strategy. The calculation of the spread speed of poisonous weeds helps to set isolation belts in advance or conduct preventive weed control.

[0092] The evaluation of diffusion patterns is based on the recording and analysis of historical data on poisonous weeds. The robot will build a diffusion model of poisonous weeds through multiple removal records, and predict future diffusion trends based on environmental conditions. The system will record detailed data of each poisonous weed removal, including the type, density, location and removal time of the poisonous weeds. These data will be used to generate a historical model of the spread of poisonous weeds and evaluate the spread range and direction of poisonous weeds in different periods. Through machine learning and data regression models, the robot can analyze the diffusion patterns of poisonous weeds under different environmental conditions in the past, apply these patterns to the current environment, and predict the spread of poisonous weeds in the future. This can identify areas where high-density poisonous weeds may appear in advance, so that preventive measures can be taken. The robot can also adjust the diffusion model according to changes in the environment. For example, if there is heavy rainfall in the future, the spread of poisonous weeds may accelerate, and the system will automatically update the diffusion model and make timely adjustments.

[0093] In a specific embodiment, the preventive measures include: establishing isolation zones in high-risk areas by physical or chemical means, regularly testing the isolation effects of the isolation zones, and performing secondary weeding operations inside and outside the isolation zones.

[0094] Specifically, according to the direction of the spread of poisonous weeds, the robot can establish isolation belts in high-risk areas in advance. Isolation belts can be established by physical means (such as laser cutting) or chemical means (such as spraying pesticides) to prevent the further spread of poisonous weeds. Based on the diffusion direction analysis, the robot uses laser equipment or robotic arms to establish isolation belts at the edge of the high-density area of ​​poisonous weeds. This can effectively prevent the roots or seeds of poisonous weeds from spreading outward and prevent the poisonous weeds from spreading to other areas. The system will regularly detect the effectiveness of the isolation belt. If it is found that there is a trend of poisonous weeds breaking through the isolation belt, the robot will further reinforce the isolation belt, or perform secondary weeding operations inside and outside the isolation belt in a targeted manner.

[0095] In a specific embodiment, it also includes subsequent processing, and the subsequent processing strategy mainly includes regular weeding, automatic reporting, alarm, automatic adjustment of weeding plan, subsequent monitoring and feedback, data storage and knowledge accumulation, etc. Through these strategies, the system can ensure that poisonous weeds will not regenerate, and take timely countermeasures when abnormal situations occur. Specifically:

[0096] Regarding regular weeding, based on the generated spatial distribution map of poisonous weeds and the analysis of their diffusion trend, the robot will set regular weeding tasks. In particular, for areas with high-density distribution and areas with obvious diffusion trends, the system will perform weeding tasks more frequently to ensure that poisonous weeds do not recur or spread. The robot can set the weeding frequency based on the distribution density of poisonous weeds. For example, in areas with high-density poisonous weeds, the robot may clear them once a week, while in areas with low density or slower diffusion, the robot will set a longer clearing cycle. The system will continuously monitor the growth of poisonous weeds in each area. If the poisonous weeds begin to recur during the scheduled weeding cycle, the robot will automatically adjust the weeding plan and take more weeding measures.

[0097] Regarding automatic reporting, when the system detects that a certain type of poisonous weed is spreading too fast, or the density of poisonous weeds in a certain area suddenly increases, the robot will automatically generate a detailed report and sound an alarm to remind relevant personnel to take timely weed control measures. The robot uses the sensors, cameras and GPS positioning system equipped to continuously monitor the type, quantity, location and growth status of poisonous weeds in the field. All data is uploaded to the central processing unit in real time. The system has a built-in anomaly detection algorithm that can analyze real-time data and compare it with historical data. If it is found that the spread rate of a certain poisonous weed exceeds the set threshold, or the density of poisonous weeds in a certain area is significantly higher than the average, the algorithm will determine it as an abnormal situation. The system will also combine environmental data such as temperature, humidity, rainfall, etc. to analyze whether there are external factors causing abnormal growth of poisonous weeds.

[0098] Once an anomaly is detected, the system automatically generates a detailed report containing the following:

[0099] Anomaly type: Clearly indicate whether it is excessive diffusion or a sudden increase in density.

[0100] Types of poisonous weeds: scientific names, common names, biological characteristics and degree of harm of the poisonous weeds involved.

[0101] Affected area: precise GPS coordinates, size of the affected area, description of surrounding environment.

[0102] Data charts: including poisonous weed quantity change curve, density heat map, diffusion trend map, etc.

[0103] Environmental conditions: recent weather conditions, soil moisture, temperature and other environmental data.

[0104] Analysis of possible causes: Preliminary speculation based on the data, such as recent rainfall leading to increased soil moisture.

[0105] The report is presented in an easy-to-understand format, including text descriptions, charts, map annotations, etc. The diverse information display methods facilitate quick understanding and decision-making.

[0106] Regarding alarms, in order to ensure timely delivery of information, the system uses a variety of alarm methods:

[0107] Sound Alarm: The robot itself emits a sound prompt to alert nearby staff.

[0108] Visual Alerts: Warning messages flash on the robot display, or abnormal areas are highlighted on the map.

[0109] Remote Notification: Alerts and reports are sent to farm managers, technicians or responsible persons via SMS, email or dedicated mobile applications.

[0110] Depending on the severity of the anomaly, alerts are divided into different levels, such as general alerts, emergency alerts, etc. The higher the severity, the higher the frequency and significance of the alerts.

[0111] Regarding the automatic adjustment of weeding plans, the system will automatically increase the priority of weeding tasks in affected areas based on abnormal situations, and the robot will go to these areas first for treatment. For poisonous weeds that spread too quickly, the system will adjust the frequency of weeding, which may increase from once a week to once a day until the problem is under control. Depending on the type of poisonous weeds and environmental conditions, the system may adjust the weeding method, for example, selecting more effective laser parameters, or combining physical and chemical means for comprehensive management.

[0112] Regarding follow-up monitoring and feedback, after taking weed control measures, the system will continue to monitor the affected areas and evaluate the effectiveness of the measures. If the density of noxious weeds decreases and the spread is controlled, the system will record a success story. If the measures are not effective, the system will continue to issue alarms and may upgrade the alarm level to indicate the need for manual intervention or more powerful measures. All new data and analysis results will be updated to the central database to provide richer information for future decision-making.

[0113] Regarding data storage and knowledge accumulation, the system will archive all abnormal events, treatment measures and results data to form a complete historical record. Using the accumulated data, the system can optimize the growth and spread model of poisonous weeds, improve prediction accuracy and response speed. If the system is connected to the Internet, these data and experience can be shared with other farms or research institutions to promote technological progress within the industry.

[0114] The embodiment of the present invention also discloses a weed control device based on computer vision, such as Figure 2 As shown, it includes: a quadruped robot chassis 1, a mechanical arm module, a sensor module, and a layered combination central processing unit; the mechanical arm module, the sensor module, and the layered combination central processing unit are all fixedly installed on the quadruped robot chassis 1;

[0115] The robotic arm module includes: a laser emitting robotic arm 304, a pesticide spraying robotic arm 305, a robotic claw, a liftable robotic arm platform, and a liftable working platform;

[0116] The sensor modules include: visible light camera, thermal infrared camera, lidar, temperature sensor, humidity sensor, rainfall sensor, GPS module;

[0117] The layered combined central processor includes a real-time control layer, an AI processing layer, and a communication and recording layer; the real-time control layer obtains positioning information based on the data uploaded by the lidar and GPS modules, and performs operation control; the AI ​​processing layer performs image processing and AI recognition on the data uploaded by the visible light camera and thermal infrared camera, determines the weeding strategy executed by the robot and controls the execution of the robotic arm module; the communication and recording layer is used to communicate and record data with the cloud platform, as well as to process environmental information collection and generate spatial distribution maps.

[0118] In a specific embodiment, the liftable work platform is fixedly mounted on the upper part of the quadruped robot chassis 1, and the mechanical claw is fixedly mounted on the liftable work platform, so that the mechanical claw moves up and down when the liftable work platform moves;

[0119] The lifting robot arm platform is fixedly mounted on the lower part of the quadruped robot chassis 1, and the pesticide spraying robot arm 305 and the laser emitting robot arm 304 are both fixedly mounted on the bottom of the lifting robot arm platform, so that the pesticide spraying robot arm 305 and the laser emitting robot arm 304 move up and down when the lifting robot arm platform moves;

[0120] The laser radar is fixedly installed on the front end of the quadruped robot chassis 1 through a laser radar bracket; the visible light camera and the thermal infrared camera are both installed on the quadruped robot chassis 1 through slide rails.

[0121] In a specific embodiment, the quadruped robot chassis 1 includes a fuselage, a drive motor and four legs that can move independently. The fuselage is composed of an upper cover, a lower cover and a baffle to form an overall frame to provide a supporting structure. The drive motor is fixedly installed on the fuselage through a motor base and a flange. The drive motor drives the legs to move through a connecting rod.

[0122] Specifically, the layered combination central processing unit adopts a layered central processing unit design. In the real-time control layer, Texas Instruments TMS320F28388D is used. This is a high-performance microcontroller designed for complex multi-axis motion control and high-precision sensor acquisition tasks, which can meet the needs of real-time operation of robots. In the AI ​​processing layer, NVIDIA Jetson AGX Orin is selected. This platform has a computing power of 275TOPS and can efficiently handle complex AI reasoning tasks such as image recognition and poisonous weed classification to ensure the accuracy and speed of robot decision-making. Qualcomm QCS8250 is recommended for the communication and recording layer. This processor has 5G connectivity, supports high-speed data upload and edge AI computing, and can efficiently handle environmental information collection and spatial distribution map generation. This architecture will greatly improve the overall performance of the system and ensure the efficient operation of the weeding robot in complex environments.

[0123] The camera uses Sony IMX477 as a visible light camera and FLIR Boson 320 as a thermal infrared camera. Sony IMX477 has a 12-megapixel resolution, excellent dynamic range and color reproduction capabilities, and can accurately capture the details of vegetation and the environment, providing high-quality visible light images for identification. The FLIR Boson 320 is a high-performance thermal infrared camera that supports 320×256 resolution and can detect temperature differences in plants, providing thermal infrared images for poisonous weed identification. Through image fusion, the texture and color information of visible light is combined with the temperature characteristics of thermal infrared, greatly improving the recognition accuracy.

[0124] The Velodyne Puck VLP-16 laser radar is a lightweight, high-precision laser radar with 16 channels and 360-degree panoramic scanning capability. The maximum measurement distance can reach 100 meters, which is suitable for large-scale environmental perception. Its high-frequency data acquisition (maximum 300,000 point cloud data per second) can provide detailed terrain and obstacle information for the weeding robot. At the same time, its compact design and low power consumption are also very suitable for the robot platform. Combined with the task of identifying poisonous weeds, it can help the robot accurately understand the three-dimensional structure of the ground, and integrate with the data of visible light and thermal infrared cameras to further improve the recognition efficiency and navigation capabilities.

[0125] The robot arm is a four-degree-of-freedom robot arm, such as Figure 6-7As shown in the figure, it has four independent joints: base rotation, shoulder joint up and down movement, elbow joint extension and wrist joint rotation. It can flexibly perform various actions and is suitable for precise operation of the robot. Three types of four-degree-of-freedom robotic arms with different functions are installed on the weeding robot, and corresponding treatment methods are adopted according to different types of poisonous weeds. The first type of robotic arm is equipped with a laser transmitter, which is specially used to deal with more stubborn poisonous weeds. It ensures thorough removal by precise laser cutting or burning of the roots of poisonous weeds. The second type of robotic arm is equipped with a pesticide spraying device, which is suitable for dealing with residual poisonous weeds or large areas of poisonous weed growth after laser treatment. By spraying pesticides at multiple angles, it can accurately cover large areas or densely distributed poisonous weeds to prevent them from spreading. The third type of robotic arm is equipped with a mechanical claw, which is mainly used for poisonous weeds with shallow roots or easy to pull out. It is cleaned by grabbing and directly pulling out. These three types of robotic arms provide a variety of treatment methods according to the different characteristics of poisonous weeds, ensuring that the robot can efficiently complete the weeding task in a complex environment.

[0126] The platform 3 of the liftable robot arm provides the robot arm with height adjustment function through the central lifting column, so that it can operate flexibly at different heights, greatly expanding the working range of the robot arm. The platform design structure is sturdy, especially when the two robot arms are working at the same time, it can still maintain stable support to ensure the accuracy of operation. The central lifting column may be hydraulically driven, which can smoothly raise and lower the platform, allowing the robot arm to easily transition from the high task area to the ground operation, adapting to a wide range of operation scenarios from high to low. The lateral bracket firmly fixes the robot arm to ensure stability during the lifting process, avoiding shaking or loss of accuracy of the robot arm when performing tasks. Figure 5 As shown, the liftable mechanical arm platform 3 includes:

[0127] Control box 301: the core control module of the system, used to manage and regulate the working status of each component;

[0128] Central lifting column 302: the main lifting support structure, which can adjust the height vertically to meet different operating requirements;

[0129] Horizontal bracket 303: connects and supports the horizontal platform of the robot arm to ensure the balance of the structure;

[0130] Laser launch robot 304: a robot with four degrees of freedom for flexibly controlling the direction of the jet and detection head;

[0131] Pesticide spraying robot arm 305: another four-axis robot arm, used in conjunction with 304 to provide more flexible operation support;

[0132] Laser transmitter 306: a laser device used for precise positioning and target identification to ensure the accuracy of the operation;

[0133] Pesticide spray head 307: A spray head used for pesticide spraying, which can accurately control the spraying range and amount to achieve fixed-point operation.

[0134] The liftable work platform 4 adopts a scissor-type structure, which realizes height adjustment through the extension and retraction of hinges and connecting rods. It has good stability, can carry a large weight and remain stable during the lifting process. The top of the platform is designed as a flat area, which can be used to carry various equipment, such as robotic arms, sensors, cameras or other tools to meet different work needs. The top is also equipped with a ring rail, which allows equipment or tools to be inserted into it and slide along the track, providing a more flexible operation method, suitable for surrounding operations or circular operations, ensuring full coverage of the work area. The platform is driven by an electric system to ensure the smoothness and safety of the lifting process and ensure efficient and stable operation. The box above the platform contains a control system, a power unit and a power supply system, which are responsible for the operation of the entire lifting platform and the robotic arm. The control module controls the various joints of the robotic arm and the lifting and lowering of the platform through circuit boards and embedded systems. The power unit includes a hydraulic device to provide power for the lifting platform. At the same time, the power supply system ensures stable power supply to all components, and exchanges information and monitors the operating status with the external control system through the communication module. Figure 4 As shown, the escalable working platform 4 includes:

[0135] Bottom control panel 401: supports the bottom platform of the entire lifting structure, providing basic support and control interface for the system;

[0136] Upper lifting plate 402: a mobile platform located at the top of the structure, used to carry and support lifting operations;

[0137] Circular track 403: A guide rail designed for the circular movement of components, which can provide continuous track support on the horizontal plane to ensure smooth circular movement of components;

[0138] Movable round rod 404: a round rod installed in the guide rail, which can move freely to support the lifting action;

[0139] Connecting block 405: a key component used to connect different rods to ensure the stability of the structure;

[0140] Three-ring connecting rod 406: a rod having three ring-shaped connecting points, used to connect and coordinate the overall movement of the platform.

[0141] The poisonous weed storage box 5 is specially designed for storing poisonous weeds removed by the mechanical claw. The main body is a square or rectangular block with a large opening, which is convenient for the mechanical claw to smoothly put the poisonous weeds in. The box body is sturdy and can withstand repeated use and a certain weight. The four buckles at the bottom are designed to firmly fix the box body on the platform or robot to ensure that the storage box will not shift or fall off during the operation. This buckle structure is easy to install and disassemble quickly, so that the storage box can be easily replaced or emptied, while ensuring the stability and efficiency of the operation process.

[0142] The structure of the quadruped robot chassis 1 is composed of multiple key components to ensure its flexible movement and stable operation. The upper and lower end covers of the fuselage constitute the overall frame, provide structural support, and protect the core components inside. The inner and outer baffles are used to separate and fix the internal components, making their layout more compact and reasonable, and preventing mutual interference during operation. The motor is the core driving part. The inner and outer motors drive different joints respectively. Each leg has a separate motor to ensure the independent movement of each leg. It is firmly installed on the fuselage through the motor base and flange connection to ensure that the motor can work smoothly during movement. The motor protection shell is used to prevent the influence of external dust, moisture, etc. on the motor and extend the life of the motor. The thigh connecting rod and the rear thigh connecting rod are responsible for transmitting the rotational force of the motor to the robot legs, providing the main power when walking. They coordinate movement to enable the robot to adjust its gait naturally. The calf connecting rod further transmits the movement of the thigh to the foot, ensuring the continuity and smoothness of the entire leg movement. The foot is the part of the robot that contacts the ground. It is usually designed with anti-skid devices to ensure the robot's stability when walking on different terrains. It can also effectively distribute the pressure of the ground to avoid sliding or tilting. This complete design enables the robot to move smoothly and flexibly in complex terrains and has strong environmental adaptability. Figure 2-3 As shown, the quadruped robot chassis 1 serves as the core frame of the entire robot, supporting and connecting other components to ensure the stability and overall coordination of the system. The robot main body of the quadruped robot chassis 1 includes:

[0143] Upper body cover 101: a cover located on the upper part of the robot body, providing protection and support for internal components;

[0144] The lower end cover 102 of the body is located at the bottom of the robot body, providing support and closing the bottom structure to enhance the overall stability;

[0145] External fuselage baffle 103: a protective plate installed on the side of the fuselage to protect the internal structure from the external environment;

[0146] First fuselage inner baffle 104: a first layer of baffle located inside the fuselage, used to fix and support internal components to ensure their stable position;

[0147] Second inner fuselage baffle 105: a second layer of baffle located inside the fuselage, further enhancing the stability and shock resistance of the internal structure.

[0148] The quadruped robot chassis 1 is provided with four legs, which are fixedly connected to the main body of the quadruped robot chassis 1 through the leg power structure 2. The specific structure of the legs is as follows Figure 8 As shown, including:

[0149] External motor 201: an external motor used to provide power to drive the movement of the robot's legs;

[0150] Motor flange 202: a connection flange between the motor and the leg structure to ensure that the motor is securely installed;

[0151] Connecting rod 203: a rod connecting the motor and the leg bracket to transmit the driving force of the motor;

[0152] Internal motor 204: a motor located inside the leg structure, which assists the external motor in precise control;

[0153] Motor protection shell 205: a shell used to protect the internal and external motors from external damage;

[0154] Motor base 206: supports and fixes the base of the motor to ensure the stability of the motor during movement;

[0155] First thigh link 207: one of the main connecting rods of the robot leg, providing support and movement functions;

[0156] Second thigh link 208: another main leg connecting rod, cooperating with the first thigh link 207 to provide stable support for the leg;

[0157] Thigh rear link 209: a rear link connecting the thigh portion and the base, providing support and resilience;

[0158] Calf connecting rod 210: a connecting rod used for the calf part to help maintain the stability of the calf structure;

[0159] Robotic foot 211: The end component of the robot's leg that contacts the ground and completes the walking movement.

[0160] Various sensors, including visible light cameras, thermal infrared cameras, LiDAR, temperature sensors, humidity sensors, rainfall sensors, GPS modules, etc., provide the robot with comprehensive environmental perception and poisonous weed identification capabilities. Visible light cameras capture the appearance details of poisonous weeds, and thermal infrared cameras detect the state of plants through temperature differences, especially in low light conditions. LiDAR generates three-dimensional maps to ensure accurate navigation and obstacle avoidance. Temperature and humidity sensors and rainfall sensors monitor environmental conditions and help analyze the environmental factors for the growth of poisonous weeds, thereby optimizing treatment strategies. The collaborative work of these sensors enables the robot to efficiently identify, manage, and treat poisonous weeds, ensuring the smooth completion of the mission.

[0161] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0162] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for removing poisonous weeds based on computer vision, characterized in that: The specific steps are as follows: Collect and preprocess image data of different types of poisonous weeds to establish a poisonous weed dataset; Constructing a YOLOv9 model and using the poisonous weed dataset for training and testing to obtain a poisonous weed recognition model; Acquire image information, positioning information and environmental information of each position of the grassland, use the poisonous weed recognition model to identify the poisonous weed information in the image information, and determine the three-dimensional position of each poisonous weed according to the positioning information, and then generate a spatial distribution map according to the three-dimensional position of the poisonous weed and the environmental information of the corresponding position; Based on the distribution characteristics and biological characteristics of poisonous weeds at each location in the spatial distribution map, poisonous weeds are treated according to differentiated treatment strategies; high-risk areas in the spatial distribution map are determined using a poisonous weed diffusion model, and preventive measures are taken.

2. A method for removing poisonous weeds based on computer vision according to claim 1, characterized in that: The collecting of different types of poisonous weed image data specifically includes: selecting the types of poisonous weeds that need to be identified and controlled; selecting different time and lighting shooting conditions according to the types of poisonous weeds, and shooting visible light images and infrared images of the poisonous weeds from multiple different angles; The preprocessing includes: geometrically correcting the visible light image and the infrared image to ensure that the two are spatially aligned, applying a filtering algorithm to remove noise in the image, and performing edge detection and contrast enhancement on the image; using a method based on a self-supervised transformer to fuse the visible light image and the infrared image; and annotating the fused image to obtain the poisonous weed dataset.

3. A method for removing poisonous weeds based on computer vision according to claim 2, characterized in that: The method based on self-supervised transformer combines convolutional neural network and transformer model to efficiently fuse visible light image and infrared image of poisonous weeds, including: independent encoders extract features of the two images respectively, the fusion layer fuses features by enhancing multimodal synergy effect through high-order space and channel interaction, and then the decoder restores the fused features into a fused image.

4. The method for removing poisonous weeds based on computer vision according to claim 1, characterized in that: The image information includes: visible light image, infrared image, three-axis shooting angle and focal length of the poisonous weed; The positioning information includes: GPS positioning data of the shooting location and laser radar three-dimensional point cloud data; The environmental information includes: soil moisture, light intensity, temperature, humidity, and precipitation; The poisonous weed information includes: poisonous weed type, size, and state.

5. The method for removing poisonous weeds based on computer vision according to claim 1, characterized in that: The distribution characteristics include concentrated continuous distribution and sporadic distribution, which are determined according to the density threshold of poisonous weeds in the area; the biological characteristics include root depth and growth years; the differentiated treatment strategies specifically include: Concentrated and continuous poisonous weeds: Cover large areas of poisonous weeds with pesticides and use laser technology to establish isolation zones to prevent spread; determine the amount of pesticides and spraying paths based on the density and distribution of poisonous weeds; Scattered poisonous weeds: precise cutting with laser transmitters, or removal with mechanical claws and local spraying; Toxic weeds with deep roots: comprehensive treatment through mechanical claw removal, deep spraying and heat treatment; Poisonous weeds with shallow roots: Use mechanical claws to remove them directly.

6. The method for removing poisonous weeds based on computer vision according to claim 1, characterized in that: The poisonous weed diffusion model is generally obtained in the following manner: Obtain historical data on the growth of various types of poisonous weeds, the historical data including the type, density, location, removal time and environmental information of the poisonous weeds; establish a machine learning model and a data regression model, use the historical data for data regression and training, and obtain a poisonous weed diffusion model.

7. The method for removing poisonous weeds based on computer vision according to claim 1, characterized in that: The preventive measures include: establishing isolation zones in the high-risk areas by physical or chemical means, regularly testing the isolation effects of the isolation zones, and performing secondary weeding operations inside and outside the isolation zones.

8. A poisonous weed removal device based on computer vision, using a poisonous weed removal method based on computer vision as claimed in any one of claims 1 to 7, characterized in that: include: Quadruped robot chassis, robotic arm module, sensor module, and layered combined central processing unit; The mechanical arm module, the sensor module, and the layered combined central processing unit are all fixedly mounted on the quadruped robot chassis; The robotic arm module includes: a laser emitting robotic arm, a pesticide spraying robotic arm, a robotic claw, a liftable robotic arm platform, and a liftable working platform; The sensor module includes: a visible light camera, a thermal infrared camera, a laser radar, a temperature sensor, a humidity sensor, a rainfall sensor, and a GPS module; The layered combined central processor includes a real-time control layer, an AI processing layer, and a communication and recording layer; the real-time control layer obtains positioning information based on the data uploaded by the laser radar and the GPS module, and performs operation control; the AI ​​processing layer performs image processing and AI recognition on the data uploaded by the visible light camera and the thermal infrared camera, determines the weeding strategy executed by the robot and controls the execution of the robotic arm module; the communication and recording layer is used to communicate with the cloud platform, record data, and process environmental information collection and spatial distribution map generation.

9. The computer vision-based weed control device according to claim 8, characterized in that: The liftable working platform is fixedly mounted on the upper part of the chassis of the quadruped robot, and the mechanical claw is fixedly mounted on the liftable working platform, so that the mechanical claw moves up and down when the liftable working platform moves; the liftable mechanical arm platform is fixedly mounted on the lower part of the chassis of the quadruped robot, and the pesticide spraying mechanical arm and the laser emitting mechanical arm are both fixedly mounted on the bottom of the liftable mechanical arm platform, so that the pesticide spraying mechanical arm and the laser emitting mechanical arm move up and down when the liftable mechanical arm platform moves; the laser mine is fixedly mounted on the front end of the chassis of the quadruped robot through a laser radar bracket; the visible light camera and the thermal infrared camera are both mounted on the chassis of the quadruped robot through slide rails.

10. The computer vision-based weed control device according to claim 8, characterized in that: The quadruped robot chassis includes a fuselage, a drive motor and four legs that can move independently. The fuselage is composed of an upper cover, a lower cover and a baffle to form an overall frame to provide a supporting structure. The drive motor is fixedly installed on the fuselage through a motor base and a flange. The drive motor drives the legs to move through a connecting rod.

Images