A field drug spraying control method and system based on visual recognition

Abstract

The present invention discloses a field pesticide spraying control method and system based on visual recognition, which relates to the field of agricultural extension technology. The method comprises the following steps: placing a debugged target device in a selected field area, collecting image information of the field area below in real time, obtaining the outline of each crop using an image segmentation algorithm, analyzing the image within each outline through a machine learning model, identifying the disease and pest situation of each crop, and synchronously acquiring a status data set. The technical key points are: this technical solution improves the utilization rate of pesticides, reduces pesticide waste and environmental pollution, ensures the healthy growth of crops, and improves the yield and quality of crops. The visual recognition technology is combined with the spraying amount control operation to realize precise targeted spraying operations. Its intelligent and adaptive characteristics also provide strong support for the precise management and sustainable development of modern agriculture.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural extension, and in particular to a field drug spraying control method and system based on visual recognition. Background Art

[0002] Agricultural extension technology refers to the activities that popularize scientific and technological achievements and practical technologies that can be applied to the fields of planting, animal husbandry, forestry, fishery, etc. through experiments, demonstrations, training, guidance and consulting services, and ensure their effective application in the entire process of agricultural production before, during and after production. It is a new agricultural extension technology that uses visual recognition technology to monitor field crops in real time and identify the occurrence of pests and diseases.

[0003] The existing application publication number CN118570678A, entitled "Method, Device, Electronic Device, and Storage Medium for Spraying Growth-Controlling Drugs," states that the growth-controlling drug spraying method includes: determining an undergrowing area of ​​crops in a target plot; converting the target plot into a target array based on the spraying width of a drone and the undergrowing area of ​​crops; the target array includes multiple array points, the distance between each array point equals the spraying width of the drone, and the multiple array points include normal array points and abnormal array points, the abnormal array points being array points within the undergrowing area of ​​crops; based on a reinforcement learning model, determining a spraying path for the drone in the target array, the spraying path must pass through each normal array point, each normal array point only once, and cannot pass through each abnormal array point; controlling the drone to spray the growth-control drug on the crops in the target plot according to the spraying path; the above-mentioned spraying method can improve the accuracy of drone spraying of growth-control drugs and avoid overspraying, but it only uses the drone to spray drugs in a range on crops from a high altitude and cannot accurately operate and treat specific plants.

[0004] In combination with the above documents and existing technologies, when traditionally using equipment with a crawler mechanism to spray pesticides in the field, the method usually adopted is to use a nozzle to spray the pesticides. During spraying, a large range of operations will be carried out, and it is impossible to accurately apply pesticides to individual seedlings or crops. By improving the design of the nozzle, the nozzle can be used to apply pesticides to each seedling, but the specific spraying amount cannot be accurately controlled, and there will still be excessive and uneven application of pesticides. At the same time, for crops that have already been sprayed, if the spraying amount of the pesticide is still set according to the previous standards, there will still be certain problems. Therefore, how to accurately adjust the spraying amount according to the status of individual crops is the core problem that needs to be solved. Summary of the Invention

[0005] (1) Technical problems solved

[0006] In response to the shortcomings of the existing technology, the present invention provides a field pesticide spraying control method and system based on visual recognition. This solution improves the utilization rate of pesticides, reduces pesticide waste and environmental pollution, ensures the healthy growth of crops, and improves the yield and quality of crops. It combines visual recognition technology with spraying rate control operations to achieve precise targeted spraying operations, solving the problems raised in the background technology.

[0007] (2) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] A field drug spraying control method based on visual recognition, the method comprises the following steps:

[0010] The debugged target device is placed in the selected field area, and real-time image information of the field area below is collected. The image segmentation algorithm is used to obtain the outline of each crop. The image within each outline is analyzed through the machine learning model to identify the pest and disease status of each crop, and the status data set is simultaneously acquired;

[0011] In the direction of the target device's movement, the pest and disease situation and status datasets are preprocessed to build a corresponding crop status assessment model, using the presence of pests and diseases as a constraint condition.

[0012] Input the pre-processed pest and disease situation and status data set, and output the corresponding crop status index;

[0013] Mark crops with pests and diseases, calculate the estimated spraying amount based on the status index, and send spraying instructions;

[0014] Phase 1: Receive spraying instructions and spray the crops below according to the estimated spraying amount;

[0015] The second stage: The crops that have been sprayed in real time are subjected to secondary image recognition processing to detect the coverage of the corresponding crops. If the coverage rate does not meet the coverage standard, an early warning signal is issued and the secondary spraying mechanism is triggered to spray the drug on the corresponding crop surface until the coverage rate of the corresponding crops reaches the standard.

[0016] Phase 3: When it is recognized that the corresponding crop has been sprayed before the start of the first phase, a visual inspection operation is performed to determine whether its current state index is lower than the initial state index. If it is not lower than the initial state index, the dosing strategy is adjusted.

[0017] Furthermore, the target equipment includes a spraying robot, two rows of spraying pipes installed below the spraying robot, and a probe module placed between the two spraying pipes.

[0018] Among them, the spray pipe group includes several cross-nozzle structures, which correspond to the crops planted in the selected field area in an upper and lower relationship during spraying. The cross-nozzle structure consists of a Class A nozzle located in the center and a Class B nozzle located at the four corners; the nozzles of the Class A nozzles are vertically downward, and the nozzles of the Class B nozzles are all directed towards a fixed point below the Class A nozzles, and are distributed in an inclined manner.

[0019] Furthermore, the process of analyzing the image within each contour is as follows:

[0020] Feature extraction: For each crop contour image, extract the required features;

[0021] Among them, the required characteristics include at least: color, texture, shape, size, number and distribution of lesions;

[0022] Machine learning model analysis: The extracted features are input into a pre-trained machine learning model for analysis to identify the characteristics of various pests and diseases. During model training, supervised learning is performed to enable the machine learning model to map the input features to the output of pest and disease density estimates.

[0023] Pest and disease identification and density estimation: Based on the analysis results of the machine learning model, the pest and disease situation of each contour image is identified. If the corresponding crop is diseased, the machine learning model outputs the corresponding disease type and the estimated density of the pest and disease on each crop.

[0024] Furthermore, the status dataset includes the average leaf width value and chlorophyll content of each crop.

[0025] Furthermore, when constructing the corresponding crop status assessment model, the formula is based on:

[0026]

[0027] Where Csi(D, W, C) represents the crop state index calculation function, Csi represents the state index of the corresponding crop, D represents the density estimation value, W represents the average leaf width value, C represents the chlorophyll content, and D represents the crop state index. c represents the critical density value, W_opt represents the average width of the leaf;

[0028] a1, b1, and c1 are all weight coefficients in the linear calculation function, and their value ranges are [0, 1].

[0029] a2, b2, and c2 are all weight coefficients in the exponential calculation function, and their value ranges are [0, 2]. α is the exponential term coefficient, and α>0;

[0030] k1 and k2 represent the weight coefficients of the influence of average leaf width and chlorophyll content on the state index when there is no disease or insect pest, and the value range is [0, 1];

[0031] If indicates a constraint condition.

[0032] Furthermore, the method for calculating the estimated spraying amount based on the state index is as follows:

[0033] P=B0+q*Csi

[0034] Where P represents the estimated spraying amount of the corresponding crop, B0 represents the basic spraying amount, and q represents the proportional coefficient, and q is a positive number greater than 0.

[0035] Furthermore, in the first and second stages, the spraying action is performed using the Class A nozzle in the corresponding cross nozzle structure; in the third stage, the dosing strategy is adjusted using the Class A nozzle and Class B nozzle in the corresponding cross nozzle structure.

[0036] Furthermore, a marker is pre-added to the liquid medicine to be sprayed, and during the secondary image recognition process, the marker image on the crop surface is analyzed to detect the amount of medicine coverage of the corresponding crop, i.e., the marker coverage;

[0037] Among them, color recognition technology is used in the secondary image recognition processing to extract the area that matches the color of the marker, and the extracted area is quantitatively analyzed to obtain the drug coverage.

[0038] Furthermore, the adjusted dosing strategy is implemented as follows:

[0039] Based on the current state index and the initial state index, a proportional adjustment calculation model is established to generate the dosage of the corresponding crop, and the estimated spraying amount corresponding to the initial state index is accumulated to obtain the secondary spraying amount for the crop. The crop is sprayed according to the secondary spraying amount.

[0040] When spraying the crop based on the secondary spraying amount, the distance between the current corresponding crop and the surrounding adjacent crops is detected, the minimum distance value is extracted, and the minimum distance value is input into the rule engine to output the control spacing. In the cross nozzle structure corresponding to the crop, the extension value of each Class B nozzle is executed according to the control spacing.

[0041] Furthermore, the formula used when establishing the proportional regulation calculation model is as follows:

[0042] Jd=C_ratio×C_dif 2 ×ε

[0043] Where Jd represents the dosage of the corresponding crop, C_ratio represents the ratio between the current state index and the initial state index, C_dif represents the difference between the current state index and the initial state index, ∈ represents the error coefficient, and the value range is [0, 1].

[0044] The operation process of the rule engine is as follows:

[0045] Receive the minimum distance and simultaneously collect the current height of the target device. Set in the rule engine:

[0046]

[0047] Where Sr represents the control distance, d_min represents the minimum distance, and H represents the current height of the target device.

[0048] A field drug spraying control system based on visual recognition, the system comprising:

[0049] The visual recognition module places the debugged target device in the selected field area, collects real-time image information of the field area below, uses image segmentation algorithms to obtain the outline of each crop, analyzes the image within each outline through a machine learning model, identifies the pest and disease status of each crop, and simultaneously obtains the status data set;

[0050] The judgment control module pre-processes the pest and disease situation and status data set in the direction of movement of the target device, builds a status assessment model for the corresponding crop, and uses the presence of pests and diseases in the corresponding crop as a constraint condition;

[0051] Input the pre-processed pest and disease situation and status data set, and output the corresponding crop status index;

[0052] Mark crops with pests and diseases, calculate the estimated spraying amount based on the status index, and send spraying instructions;

[0053] The spraying stage control module has a built-in primary spray unit, a recheck coverage unit, and an adjustment and dosing unit; the primary spray unit is used to perform the first stage operation, the recheck coverage unit is used to perform the second stage operation, and the adjustment and dosing unit is used to perform the third stage operation;

[0054] The first stage: receiving spraying instructions and spraying the crops below according to the estimated spraying amount;

[0055] The second stage: The crops that have been sprayed in real time are subjected to secondary image recognition processing to detect the coverage of the corresponding crops. If the coverage rate does not meet the coverage standard, an early warning signal is issued and the secondary spraying mechanism is triggered to spray the drug on the corresponding crop surface until the coverage rate of the corresponding crops reaches the standard.

[0056] Phase 3: When it is recognized that the corresponding crop has been sprayed before the start of the first phase, a visual inspection operation is performed to determine whether its current state index is lower than the initial state index. If it is not lower than the initial state index, the dosing strategy is adjusted.

[0057] (3) Beneficial effects

[0058] The present invention provides a field pesticide spraying control method and system based on visual recognition, which has the following beneficial effects:

[0059] (1) This solution can accurately spray pesticides according to the actual pest and disease conditions and growth status of crops, and automatically adjust the initial spraying amount, thereby improving the utilization rate of pesticides and reducing pesticide waste and environmental pollution. At the same time, by real-time collection and analysis of field image information, it can promptly detect and deal with pest and disease problems, effectively ensuring the healthy growth of crops and improving the yield and quality of crops.

[0060] (2) This plan is implemented in stages;

[0061] First, the spraying amount is estimated based on the crop's pest and disease conditions and growth status, and the spraying action is executed;

[0062] Subsequently, secondary image recognition processing is used to detect the coverage of the pesticide to ensure that the pesticide solution evenly covers the crop surface. If the coverage rate does not meet the standard, the secondary spraying mechanism is triggered until the standard is met.

[0063] Furthermore, if a crop is identified as oversprayed, visual inspection can be used to determine changes in its status index and adjust the dosing strategy accordingly, achieving precise application and enhancing management effectiveness. Furthermore, by detecting the distance between crops and inputting this information into the rules engine, the extension value of Class B sprinklers acting on the area below the crop side can be dynamically adjusted. This prevents the spray from affecting surrounding crops while also allowing for targeted, comprehensive spraying of the target crop, improving efficacy to a certain extent.

[0064] In summary, this technical solution improves pesticide utilization, reduces pesticide waste and environmental pollution, ensures the healthy growth of crops, and improves crop yield and quality. It combines visual recognition technology with spray rate control operations to achieve precise targeted spraying operations. Its intelligent and adaptive characteristics also provide strong support for the precise management and sustainable development of modern agriculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a flow chart of the control method steps in the present invention;

[0066] Figure 2This is one of the schematic diagrams of a single cross nozzle structure in the target device used in the present invention;

[0067] Figure 3 This is the second schematic diagram of a single cross nozzle structure in the target device used in the present invention;

[0068] Figure 4 This is a schematic diagram of a real scene of the target device acting on some crops below the present invention;

[0069] Figure 5 This is a modular schematic diagram of the control system in the present invention. DETAILED DESCRIPTION

[0070] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0071] Example 1:

[0072] See also Figures 1 to 4 ,This embodiment provides a field drug spraying control method and system based on visual recognition;

[0073] This method is an intelligent field pesticide spraying control method that integrates visual recognition, information processing, and precise control. Through advanced visual recognition technology, the system can identify pests and diseases and crop status in the field in real time. Combined with a precise spraying control mechanism, it can achieve targeted and quantitative spraying of pesticides. It can carry out targeted spraying treatment in stages, effectively improving the efficiency of pesticide use, reducing environmental pollution, and improving agricultural production benefits.

[0074] The specific steps of this control method are as follows:

[0075] S1. Place the debugged target device in the selected field area, collect real-time image information of the field area below, use image segmentation algorithms to obtain the outline of each crop, analyze the image within each outline through a machine learning model, identify the pest and disease status of each crop, and synchronously obtain the status data set;

[0076] The target equipment includes a spraying robot, two rows of spray hoses installed below the robot (for spraying pesticides), and a probe module placed between the two spray hoses (for image recognition and analysis). The spraying robot is a crawler-type structure, with two crawlers located in two adjacent ditches in the selected field. Therefore, the spraying robot's movement trajectory completely covers every field below it.

[0077] Crops, such as rice seedlings, are planted in intervals within each field (they are arranged relatively regularly and correspond to the positions of the cross-sprinkler structures above, so that each cross-sprinkler structure can accurately spray the corresponding crops below);

[0078] This embodiment takes the uniform planting of rice seedlings in a field as an example:

[0079] Two rows of spray hoses are arranged vertically along the spray robot's trajectory. The robot is equipped with a medicine tank, a booster pump, and other structures to spray the required liquid medicine through the spray hoses and onto the corresponding seedlings below.

[0080] like Figures 2 to 4 As shown, the spray pipe group includes several cross nozzle structures, which are composed of a type A nozzle located in the center and type B nozzles located at the four corners. The nozzles of the type A nozzles are vertically downward, and the nozzles of the type B nozzles are all directed to a certain distance below the type A nozzles, and are distributed in an inclined manner (specifically, inclined at 30° to 60°). The area covered during spraying is referred to Figure 4 It can be seen that the liquid medicine is used to fully cover the bottom of the seedling leaves;

[0081] In the same nozzle structure, the distance between any Class B nozzle and any Class A nozzle can be adjusted synchronously through the telescopic rod, which ensures to a certain extent that the liquid sprayed from the Class B nozzle will not act on the adjacent seedlings;

[0082] Before using the image segmentation algorithm to obtain the outline of each crop:

[0083] The field image information needs to be preprocessed, including steps such as cropping, scaling, and denoising, to improve image quality. Then, an image segmentation algorithm, such as a deep learning algorithm like the Unet model, is used to segment the preprocessed field image information to obtain the outline of each crop. The image segmentation algorithm can automatically identify the crop area in the field image information and separate it from the background.

[0084] The process of analyzing the image within each contour through the machine learning model is:

[0085] Feature extraction: For each crop outline image, useful features are extracted. In addition to color, texture, and shape, these features also include the size, number, and distribution of lesions. These complex features are crucial for estimating pest and disease density.

[0086] Among them, color, texture, and shape are specific not only to the crop but also to the presence of disease spots;

[0087] Machine learning model analysis: The extracted features are input into a pre-trained machine learning model (such as a convolutional neural network (CNN)) for analysis. By learning from a large amount of labeled pest and disease image data, the machine learning model can accurately identify the characteristics of various pests and diseases. The model training also requires a large amount of image data with pest and disease density annotations. Through supervised learning, the model learns how to map the input features to the output of pest and disease density estimates.

[0088] The image data containing pest and disease density annotations can be obtained by manual annotation or by using other sensors (such as lidar); the machine learning model used in this embodiment is a CNN model;

[0089] Pest and disease identification and density estimation: Based on the analysis results of the CNN model, the pest and disease situation of each contour image is identified; if the corresponding crop is diseased, the model will output the corresponding disease type and the estimated density of the disease and pest on each crop;

[0090] Among them, the pest and disease situation shows:

[0091] the presence or absence of diseases, and if present, the type of disease and estimated density of pests on each crop;

[0092] The status data set includes the average leaf width and chlorophyll content of each crop. The average leaf width represents the data obtained by summing and averaging the maximum width of all detected leaves of the corresponding crop under the current detection environment. The chlorophyll content is obtained through non-contact spectral detection, which uses visible light and near-infrared spectral analysis technology to infer the chlorophyll content by measuring the spectral information of light after passing through or reflecting from the leaves.

[0093] S2. Preprocess the pest and disease condition and status dataset in the direction of motion of the target device, construct a corresponding crop status assessment model, use the presence of pests and diseases on the corresponding crop as a constraint, input the preprocessed pest and disease condition and status dataset, and output the corresponding crop status index;

[0094] Mark crops with pests and diseases, calculate the estimated spraying amount based on the status index, and send spraying instructions;

[0095] Among them, when the target device continues to move, its movement trajectory will completely cover the selected field area;

[0096] The contents of preprocessing the pest and disease situation and status dataset are as follows:

[0097] Normalize the pest and disease situation and status data set to facilitate subsequent calculations and processing;

[0098] When constructing the status assessment model for the corresponding crop, the formula is as follows:

[0099]

[0100] Where Csi(D, W, C) represents the crop state index calculation function, Csi represents the state index of the corresponding crop, D represents the density estimation value, W represents the average leaf width value, C represents the chlorophyll content, and D represents the crop state index. c Indicates the critical density value, which is obtained by calculating the average value of the corresponding historical data. It is used to distinguish between linear calculation and exponential calculation. W_opt represents the average width of the blade;

[0101] a1, b1, and c1 are all weight coefficients in the linear calculation function, and their value ranges are [0, 1].

[0102] a2, b2, and c2 are all weight coefficients in the exponential calculation function, and their value ranges are [0, 2]. α is the exponential term coefficient, and α>0;

[0103] k1 and k2 represent the weight coefficients of the influence of average leaf width and chlorophyll content on the state index when there is no disease or insect pest, and the value range is [0, 1];

[0104] if indicates a constraint condition;

[0105] The weight coefficient is determined using the coefficient of variation method, which assigns weights to each indicator based on the degree of variation between the current value of each evaluation indicator and the target value. If the numerical difference of an indicator is large, it can clearly distinguish the evaluated objects, indicating that the indicator has rich discrimination information, and thus it should be given a larger weight. Conversely, if the numerical difference of each evaluated object on a certain indicator is small, then the ability of this indicator to distinguish the evaluated objects is weak, and thus it should be given a smaller weight. This method directly uses the information contained in each indicator to calculate the weight of the indicator, so it is objective.

[0106] Function logic description:

[0107] When D≤D c When the pest and disease density is low, a linear calculation method is used, and the state index Csi is a linear combination of the density estimate D, the deviation degree of the average leaf width value (1-W / W_opt), and the degree of insufficient chlorophyll content (1-C); when D>D cWhen D = 0 (high pest and disease density), the exponential calculation method is used. The state index Csi is an exponential function of the density estimate D, and is also affected by the average leaf width and chlorophyll content. When D = 0 (no pests and diseases), the state index Csi is mainly affected by the deviation degree of the average leaf width (1-W / W_opt) and the insufficient chlorophyll content (1-C). The square function is used to calculate its nonlinear contribution.

[0108] For example:

[0109] Assume there is a set of data: D = 8 / square meter (pests and diseases exist), W = 10cm, C = 2.0mg / g, D c =5 pieces / m2, W_opt=12cm, a1, b1, c1 are 1, 0.5, 0.3 respectively, a2, b2, c2 are 2, 0.5, 0.3 respectively, α=0.5, k1, k2 are 0.4 and 0.6 respectively;

[0110] When pests and diseases exist (D=8>D c =5):

[0111] Csi(8, 10, 2.0)≈4.48+0.083-0.3=4.263 (keep three decimal places);

[0112] When there are no pests and diseases (D=0):

[0113] Csi(0, 10, 2.0)≈0.011+0.6=0.611 (keep three decimal places);

[0114] It should be noted that:

[0115] In the construction of the corresponding crop status assessment model, the larger the crop status index is, the worse the corresponding crop status is, and the more pesticides are required;

[0116] The method for calculating the estimated spraying amount based on the status index is as follows:

[0117] P=B0+q*Csi

[0118] Where P represents the estimated spraying rate for the corresponding crop, B0 represents the basic spraying rate, and q represents the proportional coefficient. The proportional coefficient q is positively correlated with the state index Csi, and q is a positive number greater than 0. It should be noted that even if the crop is in a completely healthy state, a certain amount of pesticide may be required for prevention or maintenance, so the basic spraying rate is selected.

[0119] By adopting the above technical solutions, intelligent and precise crop disease and insect pest control effects have been achieved;

[0120] The solution first uses image segmentation algorithms and machine learning models to identify pests and diseases and estimate their density in field crops. It also obtains crop status data, such as average leaf width and chlorophyll content. It then constructs a crop status assessment model, outputting a crop status index based on the pest and disease status data set. This index serves as the basis for spraying dosage decisions.

[0121] This technical solution solves the problems of over-application and uneven application in traditional spraying methods. It enables precise spraying based on the actual pest and disease conditions and growth status of crops, improving pesticide utilization and reducing pesticide waste and environmental pollution. Furthermore, by collecting and analyzing field image information in real time, it can promptly detect and address pest and disease problems, effectively ensuring healthy crop growth and improving crop yield and quality. Furthermore, the solution can automatically adjust spray dosage based on specific conditions, such as pest and disease conditions, providing strong support for the sustainable development of modern agriculture.

[0122] S3, Phase 1:

[0123] Receive spraying instructions and spray the crops below according to the estimated spraying amount;

[0124] Among them, the structure used when executing the spraying action is:

[0125] lie in Figure 4 The cross nozzle structure in front of the middle arrow corresponds to a cross nozzle structure located directly above the corresponding crop. Only the valve connecting the liquid medicine and the type A nozzle is opened, and the type A nozzle sprays the liquid medicine on the surface of the corresponding crop, thereby achieving the spraying action. The corresponding spraying amount is also operated according to the estimated spraying amount P of the corresponding crop.

[0126] Phase 2:

[0127] Perform secondary image recognition processing on crops that have been sprayed in real time to detect the amount of drug coverage on the corresponding crops. When the drug coverage rate reaches the preset coverage standard, no response will be taken;

[0128] When the drug coverage rate does not meet the coverage standard, an early warning signal is issued and the secondary spraying mechanism is triggered to spray the drug on the corresponding crop surface until the drug coverage of the corresponding crop is detected to meet the standard;

[0129] Among them, the surface of the crop that has been sprayed in real time is covered with liquid medicine;

[0130] Add markers to the drug solution. During the secondary image recognition process, only the marker image on the crop surface is analyzed to detect the drug coverage of the corresponding crop, i.e. the marker coverage.

[0131] It should be noted that this marker needs to have a significant visual feature, such as a specific color or fluorescence, and be harmless to the seedlings and the environment. For example, a food-grade food coloring can be used as a marker. Addition method: When preparing the drug, simply mix an appropriate amount of the marker evenly into the drug.

[0132] Specifically, color recognition technology is used in the secondary image recognition processing:

[0133] Principle: Color recognition technology based on color spaces (such as RGB and HSV) can detect and analyze specific colors in images. Application: A high-resolution color camera (i.e., a probe module) is used to capture images of the crop surface. A color recognition algorithm can be used to extract areas that match the marker color. Advantages: Intuitive and easy to implement, suitable for situations where the marker color contrasts sharply with the crop surface.

[0134] Apply the above color recognition technology to extract the area of ​​the marker;

[0135] By quantitatively analyzing the extracted marker areas, its drug coverage can be obtained;

[0136] The preset coverage standard is usually set at 95% coverage;

[0137] When the drug coverage rate does not reach 95%, an early warning signal is issued, which is received, recorded and stored by the remote end. This indicates that there is an abnormality or other problem in the spray pipe group of the corresponding target equipment, and it needs to be checked and repaired after the operation is completed.

[0138] When the secondary spraying mechanism is triggered:

[0139] The structure used refers to Figure 4 As shown, another set of spray pipes behind the arrow opens the corresponding Class A nozzles flush with the Class A nozzles in front. After the target device moves, the corresponding Class A nozzles in the rear will be above the corresponding crops, and then a second spraying action will be performed. At this time, the probe module remains open until it detects that the drug coverage rate reaches the preset coverage standard;

[0140] Phase 3:

[0141] Under the condition that it is recognized that the corresponding crop has been sprayed before the start of the first stage, a visual inspection operation is performed to determine whether its current state index is lower than the initial state index;

[0142] If it is not lower than that, the dosing strategy is adjusted, proving that the initial spraying action is invalid;

[0143] If it is lower, no response will be made, proving that the initial spraying action is effective;

[0144] If it is recognized that the corresponding crop has been sprayed before the start of the first stage, it means that the current spraying operation is the second or more spraying operations in the same period (for example, the seedling period);

[0145] The visual inspection operation is performed taking into account the crops that have been marked previously;

[0146] The adjusted dosing strategy implemented is as follows:

[0147] S301: Establishing a proportional adjustment calculation model based on the current state index and the initial state index to generate a dosage for the corresponding crop, and accumulating the estimated spraying amount corresponding to the initial state index to obtain a secondary spraying amount for the crop, and completing the spraying treatment for the crop based on the secondary spraying amount;

[0148] The formula used when establishing the proportional regulation calculation model is as follows:

[0149] Jd=C_ratio×C_dif 2 ×ε

[0150] Where Jd represents the dosage of the corresponding crop, C_ratio represents the ratio between the current state index and the initial state index, C_dif represents the difference between the current state index and the initial state index, ∈ represents the error coefficient, and the value range is [0, 1];

[0151] Logic explanation: The ratio between the current state index and the initial state index reflects the degree of change of the state index. The difference between the current state index and the initial state index is then squared. This square term emphasizes the impact of the magnitude of the change in the state index on the dosage. That is, the greater the change in the state index, the faster the dosage increases (because the square term amplifies the change). Finally, the error coefficient is used to consider the errors in the actual measurement and model calculation to adjust the final result and improve the accuracy of the calculation result.

[0152] The following table shows an example:

[0153]

[0154] In the above example, the error coefficient is a variable. Through this formula and table example, we can more intuitively understand the relationship between the dosage and the change of the state index, and adjust the error coefficient according to the actual situation to adapt to different application scenarios;

[0155] S302: When spraying the crop according to the secondary spraying amount, the distance between the crop and the surrounding crops is detected, the minimum distance is extracted, and the minimum distance is input into the rule engine to output the control spacing. The extension value of each Class B sprinkler in the cross sprinkler structure above the crop is implemented according to the control spacing;

[0156] Among them, when the crop is sprayed according to the secondary spraying amount, all types of nozzles in the cross nozzle structure used are as follows Figure 2 In the state shown, the type A nozzle sprays the crop from above, and the type B nozzle sprays the crop from the side. Without affecting other crops, the upper and lower surfaces of the crop leaves and the roots can be infiltrated with the liquid to a certain extent.

[0157] Specifically, during secondary spraying, in order to accurately spray the liquid and avoid affecting surrounding crops, we need to detect the distance between the target crop and adjacent crops. Specifically, we use image recognition or sensor technology to obtain several distance values ​​between the crop border and the adjacent crop borders.

[0158] To extract the minimum of these distance values, a traversal algorithm is used. This involves comparing all detected distance values ​​one by one, initially setting a larger value as a candidate for the minimum. Each distance value is then compared with this candidate value in turn. If the current distance value is less than the candidate value, the candidate value is updated to the current distance value. After traversing all distance values, the candidate value becomes the desired minimum distance value. This algorithm is simple and efficient, quickly and accurately extracting the minimum distance between crops, providing key parameters for subsequent spraying processes and ensuring that the spray solution is accurately sprayed on the target crops.

[0159] The operation process of the rule engine is as follows:

[0160] Receive the minimum distance and simultaneously collect the current height of the target device. Set in the rule engine:

[0161]

[0162] Where Sr represents the control distance, d_min represents the minimum distance, and H represents the current height of the target device;

[0163] The control spacing Sr should be a balance between the minimum distance d_min and the target device height H. Specifically:

[0164] When d_min is small (i.e., the distance between crops is very close), Sr should be relatively small to avoid spraying the liquid onto adjacent crops, thereby reducing the extension value of the Class B nozzle. When H is large (i.e., the equipment is very high), the liquid may be more easily affected by external factors such as wind during the spraying process and drift. Therefore, Sr should also be relatively small to ensure that the liquid can be sprayed accurately on the target crops.

[0165] Specifically, in the same cross nozzle structure, the four B-type nozzles are all away from the central A-type nozzle. The distance away is the adjustment spacing, refer to Figure 3 shown.

[0166] By adopting the above technical solution, accurate and efficient crop spraying effect is achieved, solving the problems of excessive spraying, uneven spraying and inability to adjust the spraying amount according to the actual state of the crop in the traditional spraying process;

[0167] The solution first estimates the spraying amount based on the crop's pest and disease conditions and growth status, and then executes the spraying action;

[0168] Subsequently, secondary image recognition processing is used to detect the coverage of the pesticide to ensure that the pesticide solution evenly covers the crop surface. If the coverage rate does not meet the standard, the secondary spraying mechanism is triggered until the standard is met.

[0169] Furthermore, if the system detects that a crop has been oversprayed, it uses visual inspection to determine changes in its status index and adjusts the dosing strategy accordingly, achieving precise application. Furthermore, during the secondary spraying process, the system dynamically adjusts the extension value of Class B sprinklers by detecting the distance between crops and inputting this information into a rules engine to prevent the spray from affecting surrounding crops.

[0170] This technical solution improves the utilization rate of pesticides, reduces pesticide waste and environmental pollution, ensures the healthy growth of crops, and improves the yield and quality of crops; at the same time, its intelligent and adaptive characteristics also provide strong support for the precise management and sustainable development of modern agriculture.

[0171] Example 2:

[0172] See also Figure 5 Based on Example 1, this embodiment further provides a field drug spraying control system based on visual recognition, which includes the following:

[0173] The visual recognition module places the debugged target device in the selected field area, collects real-time image information of the field area below, uses image segmentation algorithms to obtain the outline of each crop, analyzes the image within each outline through a machine learning model, identifies the pest and disease status of each crop, and simultaneously obtains the status data set;

[0174] The judgment control module pre-processes the pest and disease situation and status data set in the direction of movement of the target device, builds a status assessment model for the corresponding crop, and uses the presence of pests and diseases in the corresponding crop as a constraint condition;

[0175] Input the pre-processed pest and disease situation and status data set, and output the corresponding crop status index;

[0176] Mark crops with pests and diseases, calculate the estimated spraying amount based on the status index, and send spraying instructions;

[0177] The spraying stage control module has a built-in primary spray unit, a recheck coverage unit, and an adjustment and dosing unit; the primary spray unit is used to perform the first stage operation, the recheck coverage unit is used to perform the second stage operation, and the adjustment and dosing unit is used to perform the third stage operation;

[0178] The first stage: receiving spraying instructions and spraying the crops below according to the estimated spraying amount;

[0179] The second stage: The crops that have been sprayed in real time are subjected to secondary image recognition processing to detect the coverage of the corresponding crops. If the coverage rate does not meet the coverage standard, an early warning signal is issued and the secondary spraying mechanism is triggered to spray the drug on the corresponding crop surface until the coverage rate of the corresponding crops reaches the standard.

[0180] Phase 3: When it is recognized that the corresponding crop has been sprayed before the start of the first phase, a visual inspection operation is performed to determine whether its current state index is lower than the initial state index. If it is not lower than the initial state index, the dosing strategy is adjusted.

[0181] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0182] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0183] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A field drug spraying control method based on visual recognition, characterized in that: The steps of this method are: The debugged target device is placed in the selected field area, and real-time image information of the field area below is collected. The image segmentation algorithm is used to obtain the outline of each crop. The image within each outline is analyzed through the machine learning model to identify the pest and disease status of each crop, and the status data set is simultaneously acquired; In the direction of the target device's movement, the pest and disease situation and status datasets are preprocessed to build a corresponding crop status assessment model, using the presence of pests and diseases as a constraint condition. Input the preprocessed pest and disease situation and status data set, and output the corresponding crop status index; the status data set includes the average leaf width and chlorophyll content of each crop; when constructing the corresponding crop status assessment model, the formula based on it is: ； Where Csi (D, W, C) represents the crop state index calculation function, Csi represents the state index of the corresponding crop, D represents the density estimation value, W represents the average leaf width value, and C represents the chlorophyll content. represents the critical density value, W_opt represents the average width of the leaf; Both are weight coefficients in the linear calculation function, and their value ranges are [0, 1]; are all weight coefficients in the exponential calculation function, and their value range is [0, 2]. α is the exponential term coefficient, and α>0; Indicates the weight coefficient of the influence of leaf average width and chlorophyll content on the state index when there is no disease or insect pest, and the value range is [0, 1]; if indicates a constraint condition; Mark crops with pests and diseases, calculate the estimated spraying amount based on the status index, and send spraying instructions; Phase 1: Receive spraying instructions and spray the crops below according to the estimated spraying amount; The second stage: The crops that have been sprayed in real time are subjected to secondary image recognition processing to detect the coverage of the corresponding crops. If the coverage rate does not meet the coverage standard, an early warning signal is issued and the secondary spraying mechanism is triggered to spray the drug on the corresponding crop surface until the coverage rate of the corresponding crops reaches the standard. Phase 3: When it is recognized that the corresponding crop has been sprayed before the start of the first phase, a visual inspection operation is performed to determine whether its current state index is lower than the initial state index. If it is not lower than the initial state index, the dosing strategy is adjusted.

2. The method for controlling field pesticide spraying based on visual recognition according to claim 1, wherein: The target equipment includes a spraying robot, two rows of spray hoses installed below the robot, and a probe module placed between the two spray hoses. Among them, the spray pipe group includes several cross-nozzle structures, which correspond to the crops planted in the selected field area in an upper and lower relationship during spraying. The cross-nozzle structure consists of a Class A nozzle located in the center and a Class B nozzle located at the four corners; the nozzles of the Class A nozzles are vertically downward, and the nozzles of the Class B nozzles are all directed towards a fixed point below the Class A nozzles, and are distributed in an inclined manner.

3. The method for controlling field pesticide spraying based on visual recognition according to claim 2, wherein: The process of analyzing the image within each contour is: Feature extraction: For each crop contour image, extract the required features; Among them, the required characteristics include at least: color, texture, shape, size, number and distribution of lesions; Machine learning model analysis: The extracted features are input into a pre-trained machine learning model for analysis to identify the characteristics of various pests and diseases. During model training, supervised learning is performed to enable the machine learning model to map the input features to the output of pest and disease density estimates. Pest and disease identification and density estimation: Based on the analysis results of the machine learning model, the pest and disease situation of each contour image is identified. If the corresponding crop is diseased, the machine learning model outputs the corresponding disease type and the estimated density of the pest and disease on each crop.

4. The method for controlling field pesticide spraying based on visual recognition according to claim 1, wherein: The method for calculating the estimated spraying amount based on the status index is as follows: ； Where P represents the estimated spraying amount of the corresponding crop, B0 represents the basic spraying amount, and q represents the proportional coefficient, and q is a positive number greater than 0.

5. The method for controlling field pesticide spraying based on visual recognition according to claim 4, characterized in that: In the first and second stages, the spraying action is performed using the Class A nozzle in the corresponding cross nozzle structure; in the third stage, the Class A and Class B nozzles in the corresponding cross nozzle structure are used when adjusting the dosing strategy.

6. The method for controlling field pesticide spraying based on visual recognition according to claim 1, characterized in that: Markers are pre-added to the spray solution. During the secondary image recognition process, the marker image on the crop surface is analyzed to detect the amount of drug coverage on the corresponding crop, i.e., the marker coverage. Among them, color recognition technology is used in the secondary image recognition processing to extract the area that matches the color of the marker, and the extracted area is quantitatively analyzed to obtain the drug coverage.

7. The method for controlling field pesticide spraying based on visual recognition according to claim 5, characterized in that: The adjusted dosing strategy implemented is as follows: Based on the current state index and the initial state index, a proportional adjustment calculation model is established to generate the dosage of the corresponding crop, and the estimated spraying amount corresponding to the initial state index is accumulated to obtain the secondary spraying amount for the crop. The crop is sprayed according to the secondary spraying amount. The formula used when establishing the proportional regulation calculation model is as follows: ； Where Jd represents the dosage of the corresponding crop, C_ratio represents the ratio between the current state index and the initial state index, C_dif represents the difference between the current state index and the initial state index, and ϵ represents the error coefficient, which ranges from [0, 1]. When spraying the crop based on the secondary spraying amount, the distance between the current crop and the surrounding crops is detected, the minimum distance value is extracted, and the minimum distance value is input into the rule engine to output the control spacing. In the cross nozzle structure above the crop, the extension value of each Class B nozzle is executed according to the control spacing; The operation process of the rule engine is as follows: Receive the minimum distance and simultaneously collect the current height of the target device. Set in the rule engine: ； Where Sr represents the control distance, d_min represents the minimum distance, and H represents the current height of the target device.

8. A field drug spraying control system based on visual recognition, characterized in that: The system includes: The visual recognition module places the debugged target device in the selected field area, collects real-time image information of the field area below, uses image segmentation algorithms to obtain the outline of each crop, analyzes the image within each outline through a machine learning model, identifies the pest and disease status of each crop, and simultaneously obtains the status data set; The judgment control module pre-processes the pest and disease situation and status data set in the direction of movement of the target device, builds a status assessment model for the corresponding crop, and uses the presence of pests and diseases in the corresponding crop as a constraint condition; Input the preprocessed pest and disease situation and status data set, and output the corresponding crop status index; the status data set includes the average leaf width and chlorophyll content of each crop; when constructing the corresponding crop status assessment model, the formula based on it is: ； Where Csi (D, W, C) represents the crop state index calculation function, Csi represents the state index of the corresponding crop, D represents the density estimation value, W represents the average leaf width value, and C represents the chlorophyll content. represents the critical density value, W_opt represents the average width of the leaf; Both are weight coefficients in the linear calculation function, and their value ranges are [0, 1]; are weight coefficients in the exponential calculation function, and their value ranges are [0, 2]. α is the exponential term coefficient, and α>0; Indicates the weight coefficient of the influence of leaf average width and chlorophyll content on the state index when there is no disease or insect pest, and the value range is [0, 1]; if indicates a constraint condition; Mark crops with pests and diseases, calculate the estimated spraying amount based on the status index, and send spraying instructions; The spraying stage control module has a built-in primary spray unit, a recheck coverage unit, and an adjustment and dosing unit; the primary spray unit is used to perform the first stage operation, the recheck coverage unit is used to perform the second stage operation, and the adjustment and dosing unit is used to perform the third stage operation; The first stage: receiving spraying instructions and spraying the crops below according to the estimated spraying amount; The second stage: The crops that have been sprayed in real time are subjected to secondary image recognition processing to detect the coverage of the corresponding crops. If the coverage rate does not meet the coverage standard, an early warning signal is issued and the secondary spraying mechanism is triggered to spray the drug on the corresponding crop surface until the coverage rate of the corresponding crops reaches the standard. Phase 3: When it is recognized that the corresponding crop has been sprayed before the start of the first phase, a visual inspection operation is performed to determine whether its current state index is lower than the initial state index. If it is not lower than the initial state index, the dosing strategy is adjusted.

Images