System for accurately removing abnormal plants in rice breeding based on intelligent image recognition

Abstract

The invention discloses a system for accurately removing abnormal plants in rice breeding based on intelligent image recognition, and relates to the field of agricultural intelligent equipment.The system comprises an image acquisition module, an abnormal plant detection module and an impurity removal control module, the preprocessing module is used for preprocessing rice field plant images and transmitting the preprocessed rice field plant images into the abnormal plant detection module, the abnormal plant detection module is used for analyzing and processing the rice field plant images and identifying to-be-processed abnormal plants, and the impurity removal control module is used for receiving the positions of the to-be-processed abnormal plants and controlling the impurity removal structure to remove impurities.

Description

Technical Field

[0001] The present invention relates to the field of agricultural intelligent equipment, and in particular to a precise removal system for foreign plants in rice breeding based on intelligent image recognition. Background Art

[0002] During rice seed production, the presence of weeds seriously affects seed purity. Currently, there are two main methods for removing weeds: manual removal and traditional herbicide spraying. While manual removal can accurately remove weeds, it is extremely inefficient, consumes a lot of manpower and time, and is difficult to meet the needs of large-scale cultivation. Traditional herbicide spraying, while highly efficient, lacks specificity and can damage healthy rice plants. Excessive use of herbicides can also pollute the soil and the environment.

[0003] In order to solve these problems, there is an urgent need for a precise removal system for weeds in rice breeding based on intelligent image recognition that can accurately locate weeds and spray herbicides at fixed points to remove rice weeds. Summary of the Invention

[0004] To address the above issues, the present invention proposes a precise weed removal system for rice breeding based on intelligent image recognition. This system uses a wheeled walking mechanism and a camera deployed at the front end to capture rice images. The system determines the location of weeds through an image recognition algorithm and uses a mobile positioning algorithm to determine the position of the nozzle. The system then sprays herbicides to kill the weeds, achieving an integrated "detection-weed removal" operation. The specific contents are as follows:

[0005] A precise removal system for foreign plants in rice breeding based on intelligent image recognition, comprising:

[0006] Image acquisition module: The image acquisition module is used to collect and pre-process images of rice plants in the paddy field, and transmit the pre-processed images of rice plants to the weed detection module;

[0007] Weed plant detection module: The weed plant detection module is used to process the image features of rice field plants and analyze and identify the weed plants to be processed;

[0008] Weed removal control module: The weed removal control module is used to receive the location of the weeds to be processed and control the weed removal structure to remove weeds.

[0009] Preferably, the specific contents of collecting and preprocessing rice field plant images include:

[0010] During the collection process, the same area was collected at least three times;

[0011] Obtain the current acquisition time and weather conditions, generate an environmental marker sequence, and mark the environmental marker sequence on the rice field plant image;

[0012] Perform quality diagnosis on rice field plant images in the same area and sort them from high to low according to their quality;

[0013] The rice field plant images with unqualified quality are eliminated, and the first half of the sorting process is screened to obtain the preprocessed rice field plant images.

[0014] Preferably, the foreign plant detection module includes:

[0015] Standard model unit: the standard model unit includes standard trait indicators of several varieties;

[0016] Feature extraction unit: The feature extraction unit is used to process defects and extract features from rice field plant images to obtain feature elements, and substitute the feature elements into the analysis and recognition unit. The feature extraction unit is based on YOLO 11 Model extraction;

[0017] Analysis and identification unit: The analysis and identification unit obtains the characteristic elements and the standard trait indicators of the corresponding varieties in the standard model unit for comparison and analysis to obtain the weeds to be processed and traces back to obtain the rice field plant image and weed position where the weeds to be processed are located.

[0018] Preferably, the standard trait indicators of the variety include tiller number indicator, plant height indicator and leaf area indicator;

[0019] The tiller number index and plant height index are respectively provided with corresponding tiller number and plant height values and their fluctuation ranges;

[0020] The leaf surface index is provided with a standard center point and a standard range point, as well as a radian value and a radian value fluctuation range corresponding to the standard range point.

[0021] Preferably, the defect processing for the rice field plant image includes:

[0022] The defect processing includes image enhancement processing and image correction processing;

[0023] The image enhancement processing includes contrast adjustment, color correction and noise removal;

[0024] The specific content of the image correction processing is to obtain an environmental tag sequence, extract and determine the time and weather of the photo based on the environmental tag sequence, and perform light and shadow error processing on the picture based on the time and weather.

[0025] Preferably, the specific content of performing light and shadow error processing on the image according to time and weather is:

[0026] Defining images as cloudy, rainy, and sunny labels according to weather conditions, wherein the cloudy, rainy, and sunny labels are respectively matched with equalization processing parameters, wherein the equalization processing parameters include gamma correction parameters and color adjustment parameters;

[0027] The processing method for the cloudy day label includes: using the color adjustment parameters under the cloudy day label to adjust the color of the image to obtain a cloudy day image;

[0028] Use the gamma correction parameters under the cloudy tag to correct the cloudy image to obtain the cloudy corrected image;

[0029] The processing method for rainy day labels includes: using the deconvolution algorithm to perform inverse operations on the image to obtain a rainy day corrected image;

[0030] The processing method for the sunny day tag includes: obtaining the gamma correction parameters and image time under the sunny day tag;

[0031] Determine the solar altitude angle of the current image based on the image time, and determine the light intensity based on the solar altitude angle;

[0032] According to the light intensity, configure the adjustment factor for the gamma correction parameter under the sunny label to obtain the gamma adjustment parameter;

[0033] The image is corrected according to the gamma adjustment parameters to obtain a sunny day corrected image.

[0034] Preferably, the analysis and identification unit obtains the characteristic elements and compares them with the standard trait indicators of the corresponding varieties in the standard model unit to obtain the specific content of the offal plants to be processed:

[0035] Determine the actual number of tillers, actual plant height and leaf outline among the characteristic elements;

[0036] Confirm whether the actual number of tillers and actual plant height are within the fluctuation range;

[0037] If yes, it will be retained; if no one of the items is present, it will be determined as a weed to be processed;

[0038] The leaf contour is fixed with the standard center point as the fixed point. If the standard range point falls on the leaf contour, it is retained. If the standard range point does not fall on the leaf contour, it is determined to be a weed to be treated;

[0039] Calculate the radian value of the leaf surface contour. If the radian value of the leaf surface contour does not exceed the radian value fluctuation range, collect the actual leaf surface contour into the standard model unit.

[0040] If the curvature value of the leaf surface contour exceeds the curvature value fluctuation range, it is determined to be a weed to be treated;

[0041] If a flower bud is identified in the process of acquiring the characteristic elements by the analysis and identification unit, the spider carrying the flower bud is determined to be a weed to be processed.

[0042] Preferably, the specific content of controlling the impurity removal structure to remove impurities is:

[0043] Construct the camera coordinate system, pixel coordinate system and impurity removal structure coordinate system;

[0044] Perform camera calibration to obtain the camera's intrinsic and extrinsic parameters to form a projection matrix;

[0045] Obtain the pixel coordinates of the weeds to be processed and convert the pixel coordinates into camera coordinates based on the projection matrix;

[0046] The camera coordinate system and the impurity removal structure coordinate system are in a nested relationship. The camera coordinates are substituted into the impurity removal structure coordinate system and the position distance of the impurities to be processed in the impurity removal structure is evaluated;

[0047] Calculate the number of steps that the impurity removal structure takes to move to the position of the weeds to be treated and sort them from low to high. Select the impurity removal structure that ranks first in the step number sorting for impurity removal.

[0048] An electronic device includes a memory and a processor, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, the content of the precise removal system of foreign plants in rice breeding based on intelligent image recognition is implemented.

[0049] A storage medium stores computer-executable instructions, which, when loaded and executed by a processor, implement the contents of the precise removal system for foreign plants in rice breeding based on intelligent image recognition.

[0050] In summary, the precise removal system for foreign plants in rice breeding based on intelligent image recognition of the present invention has the following advantages over conventional technologies:

[0051] 1. This device combines image recognition technology with precise positioning and spraying. It can accurately locate weeds and spray herbicides at specific locations, avoiding accidental spraying. This represents a significant innovation in precision removal strategies. Compared to manual removal, this device can quickly and automatically complete weed removal, significantly improving work efficiency, reducing weed removal costs, and significantly increasing the economic benefits of rice seed production. Furthermore, due to precise positioning and spraying, soil and environmental pollution are reduced, resulting in significant social benefits.

[0052] 2. The device integrates image recognition, mobile positioning, herbicide spraying, and control systems, with each system working closely together. Image recognition provides the location of weeds, mobile positioning accurately moves the nozzle, the spraying system operates on demand, and the control system coordinates the entire system. The multi-system collaborative working mode is innovative in the field of agricultural weed removal equipment. Through advanced image recognition technology and precise mobile positioning mechanisms, it can accurately locate weeds and spray herbicides directly over them, effectively avoiding accidental spraying of normal surrounding rice plants. This greatly improves the accuracy of weed removal and helps improve the seed purity of rice seed production.

[0053] 3. The present invention realizes the automation and intelligence of the entire process from weed identification to removal. Operators only need to perform simple settings and monitoring, reducing manual intervention and errors. This highly automated and intelligent control is unique among similar devices. The control system realizes the automation and intelligence of the entire removal process. Operators only need to perform simple parameter settings and monitoring, reducing the complexity and errors of manual operation and improving the stability and reliability of the device.

[0054] 4. While improving removal efficiency and reducing labor costs, precise spraying can significantly reduce the use of herbicides, alleviate environmental pollution, and take into account both agricultural production efficiency and environmental protection needs, providing a new direction for agricultural weed removal.

[0055] The technical method of the present invention is further described in detail below through the accompanying drawings and examples. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a diagram showing the steps of controlling an impurity removal structure to remove impurities in a precise impurity removal system for rice breeding based on intelligent image recognition according to the present invention;

[0057] Figure 2 This is a schematic diagram of the modules of the system for accurately removing foreign plants in rice breeding based on intelligent image recognition according to the present invention;

[0058] Figure 3 It is the overall structure of the weed removal system of the present invention.

[0059] Reference numerals

[0060] 1. Electronically controlled nozzle; 2. Variable controller; 3. Stepper motor; 4. Industrial camera; 5. Lifting rod; 6. Water pump; 7. Medicine box; 8. Drive motor; 9. Special wheel for paddy fields; 10. Steering module; 11. Industrial router; 12. Signal amplifier; 13. I / O controller; 14. Industrial computer; 15. Lithium battery. DETAILED DESCRIPTION

[0061] The technical method of the present invention is further described below through the accompanying drawings and embodiments. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and values described in these embodiments do not limit the scope of this application.

[0062] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present disclosure, its application, or uses.

[0063] Technologies, systems, and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, they should be considered part of the specification.

[0064] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0065] Unless otherwise defined, technical or scientific terms used in the present invention shall have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

[0066] The present invention provides a precise removal system for foreign plants in rice breeding based on intelligent image recognition, comprising:

[0067] Image acquisition module: The image acquisition module is used to collect and pre-process images of rice plants in the paddy field, and transmit the pre-processed images of rice plants to the weed detection module;

[0068] Weed plant detection module: The weed plant detection module is used to process the image features of rice field plants and analyze and identify the weed plants to be processed;

[0069] Weed removal control module: The weed removal control module is used to receive the location of the weeds to be processed and control the weed removal structure to remove weeds.

[0070] Preferably, the specific contents of collecting and preprocessing rice field plant images include:

[0071] During the collection process, the same area was collected at least three times;

[0072] Obtain the current acquisition time and weather conditions, generate an environmental marker sequence, and mark the environmental marker sequence on the rice field plant image;

[0073] Perform quality diagnosis on rice field plant images in the same area and sort them from high to low according to their quality;

[0074] The rice field plant images with unqualified quality are eliminated, and the first half of the sorting process is screened to obtain the preprocessed rice field plant images.

[0075] Preferably, the foreign plant detection module includes:

[0076] Standard model unit: the standard model unit includes standard trait indicators of several varieties;

[0077] Feature extraction unit: The feature extraction unit is used to process defects and extract features from rice field plant images to obtain feature elements, and substitute the feature elements into the analysis and recognition unit. The feature extraction unit is based on YOLO 11 Model extraction;

[0078] Analysis and identification unit: The analysis and identification unit obtains the characteristic elements and the standard trait indicators of the corresponding varieties in the standard model unit for comparison and analysis to obtain the weeds to be processed and traces back to obtain the rice field plant image and weed position where the weeds to be processed are located.

[0079] Preferably, the standard trait indicators of the variety include tiller number indicator, plant height indicator and leaf area indicator;

[0080] The tiller number index and plant height index are respectively provided with corresponding tiller number and plant height values and their fluctuation ranges;

[0081] The leaf surface index is provided with a standard center point and a standard range point, as well as a radian value and a radian value fluctuation range corresponding to the standard range point.

[0082] Preferably, the defect processing for the rice field plant image includes:

[0083] The defect processing includes image enhancement processing and image correction processing;

[0084] The image enhancement processing includes contrast adjustment, color correction and noise removal;

[0085] The specific content of the image correction processing is to obtain an environmental tag sequence, extract and determine the time and weather of the photo based on the environmental tag sequence, and perform light and shadow error processing on the picture based on the time and weather.

[0086] Different light intensities can cause variations in the overall brightness of an image. In strong light, the image may be overexposed, resulting in a loss of plant details, particularly texture and color information in brighter areas. In low-light environments, the image may appear dark and contain excessive shadows, affecting the accurate assessment of plant morphology and characteristics.

[0087] Preferably, the specific content of performing light and shadow error processing on the image according to time and weather is:

[0088] Defining images as cloudy, rainy, and sunny labels according to weather conditions, wherein the cloudy, rainy, and sunny labels are respectively matched with equalization processing parameters, wherein the equalization processing parameters include gamma correction parameters and color adjustment parameters;

[0089] The processing method for the cloudy day label includes: using the color adjustment parameters under the cloudy day label to adjust the color of the image to obtain a cloudy day image;

[0090] Use the gamma correction parameters under the cloudy tag to correct the cloudy image to obtain the cloudy corrected image;

[0091] The processing method for rainy day labels includes: using the deconvolution algorithm to perform inverse operations on the image to obtain a rainy day corrected image;

[0092] On sunny days, there's ample but intense light, which can cause the image's contrast to be too high. Background elements like blue sky and white clouds can create a sharp contrast with the plants, detracting from the focus on the plants themselves. Furthermore, direct sunlight can cause some plants to have severe glare on their surfaces.

[0093] The processing method for the sunny day tag includes: obtaining the gamma correction parameters and image time under the sunny day tag;

[0094] Determine the solar altitude angle of the current image based on the image time, and determine the light intensity based on the solar altitude angle;

[0095] Different lighting angles can cause uneven distribution of light and shadow on a plant's surface, creating shadows and highlights. For example, side lighting can cause a distinct shadow on one side of a plant, while highlighting the other. This enhances the plant's three-dimensionality but also makes subsequent feature extraction difficult. Top lighting can highlight details at the top of a plant, but create deeper shadows at the bottom, affecting the interpretation of the plant's overall form.

[0096] According to the light intensity, configure the adjustment factor for the gamma correction parameter under the sunny label to obtain the gamma adjustment parameter;

[0097] The image is corrected according to the gamma adjustment parameters to obtain a sunny day corrected image.

[0098] This processing method can reduce the difficulty of shadow removal and highlight suppression and even avoid subsequent operations.

[0099] YOLO 11 Inheriting the excellent characteristics of the YOLO series of algorithms, fast and accurate, it demonstrates excellent performance in various complex visual tasks through its advanced neural network architecture and optimized training strategy. 11 The introduction of YOLO represents the latest progress in current target detection technology and provides a strong starting point for this invention.11 The progress of YOLO also means that it has better adaptability and scalability, which is crucial for the present invention to meet the challenges in specific visual tasks. However, the present invention also realizes that YOLO 11 The model has some shortcomings in practical applications. For example, the model has a large size, which makes it difficult to deploy on resource-constrained platforms (such as mobile devices). This challenge is particularly important for the research of this invention, because the goal of this invention is to develop an efficient and practical object detection system. At the same time, when dealing with different scenes of rice weeds detection, the sample labels of the dataset are unbalanced, and YOLO 11 There is still room for improvement in the detection accuracy of 11 It performs well in many aspects, but in certain areas, such as visual inspection of agricultural products, it still needs further optimization and adjustment. To solve these problems, USCA-YOLO 11 The model was improved in three aspects: The C3k2_Ghost module was introduced to replace the C3k2 module in the original network, reducing the model's computational complexity and memory consumption while increasing its inference speed. An adaptive threshold focal loss function was used to balance the training difficulty of positive and negative samples. The FocalModulation module was introduced to replace the SPPF module in the original network, leveraging an attention mechanism to focus on key areas in the image, further improving the detection of hybrid rice offspring in complex environments.

[0100] Lightweighting is a key goal in optimizing deep neural network models, especially when deploying them on resource-constrained devices. This paper proposes a lightweight improvement to the C3k2 feature extraction module. C3k2 is improved using the Ghost module, a dynamic convolution algorithm designed to increase the number of parameters with virtually no additional floating-point operations (FLOPs).

[0101] The C3K2 module is the latest YOLO 11The model's feature extraction module. Based on the CSPNet structure, it processes the input feature map by splitting it into two parts and using a bottleneck module for multi-scale feature extraction. However, traditional bottleneck structures typically contain multiple convolutional layers, requiring a large amount of computation. The introduction of the GhostBottleneck structure effectively solves this problem. Compared with the traditional bottleneck structure, GhostBottleneck proposes a more efficient feature extraction method, which is based on GhostConv. GhostConv uses fewer convolutional kernels to generate partial feature maps, which are then expanded using inexpensive linear transformation operations to generate more diverse feature representations, significantly reducing the computational workload while maintaining performance. Finally, the two parts of the feature map are spliced together to obtain the final output feature map. This design concept enables GhostBottleneck to maintain or even improve the performance of the model while significantly reducing the number of parameters and computations. By replacing the Bottleneck structure in the C3k2 module with the GhostBottleneck structure, the present invention can greatly reduce the computational cost of the model's feature extraction stage, thereby achieving a lightweight model.

[0102] This paper uses a fully adaptive threshold focal loss function to replace Yolo11's original loss function VFLLoss, which is based on the characteristics of the dataset used in this paper. Adaptive Threshold Focal Loss (ATFL) is a loss function that dynamically adjusts the loss weight. It aims to reduce the influence of easy-to-classify samples and increase the focus on difficult-to-classify samples, thereby improving model performance in object detection and segmentation tasks, especially in the case of class imbalance.

[0103] Focal loss function formula:

[0104] FL(p t )=-α t (1-p t ) γ log(p t );

[0105] where p t is the probability predicted by the model, α t is a balancing factor, usually used to balance positive and negative samples, and γ is an adjustment factor, used to adjust the weight between easy-to-classify and difficult-to-classify samples.

[0106] Adaptive threshold focal loss function

[0107]

[0108] Where μ represents the pixel-level feature, τ is the threshold, and k is a parameter that controls the slope, which is usually a positive number.

[0109] In the target detection task, the efficient extraction and fusion of multi-scale features are crucial for identifying objects of different sizes and shapes. SPP and its efficient version SPPF use fixed-size pooling kernels to examine the input feature map and extract multi-scale features, effectively improving the model's ability to detect targets of different sizes. However, SPP and SPPF usually use simple concatenation operations to integrate features extracted from different pooling layers, which may not fully utilize the complementary information between multi-scale features, thereby limiting the performance of the model. The basic principle of FocalModulation is to replace the self-attention module and use the focal modulation mechanism to capture long-range and contextual information in the image.

[0110] Focal Contextualization is a component of Focal Modulation. Focal Contextualization uses a series of depth-wise convolutional layers to encode visual contextual information at different scales. These layers can capture visual features from near to far, allowing the network to understand image content at different levels. In this way, the network is able to maintain sensitivity to local details while aggregating contextual information and enhance awareness of global structure.

[0111] Preferably, the analysis and identification unit obtains the characteristic elements and compares them with the standard trait indicators of the corresponding varieties in the standard model unit to obtain the specific content of the offal plants to be processed:

[0112] Determine the actual number of tillers, actual plant height and leaf outline among the characteristic elements;

[0113] Confirm whether the actual number of tillers and actual plant height are within the fluctuation range;

[0114] If yes, it will be retained; if no one of the items is present, it will be determined as a weed to be processed;

[0115] The leaf contour is fixed with the standard center point as the fixed point. If the standard range point falls on the leaf contour, it is retained. If the standard range point does not fall on the leaf contour, it is determined to be a weed to be treated;

[0116] Calculate the radian value of the leaf surface contour. If the radian value of the leaf surface contour does not exceed the radian value fluctuation range, collect the actual leaf surface contour into the standard model unit.

[0117] If the curvature value of the leaf surface contour exceeds the curvature value fluctuation range, it is determined to be a weed to be treated;

[0118] If a flower bud is identified in the process of acquiring the characteristic elements by the analysis and identification unit, the spider carrying the flower bud is determined to be a weed to be processed.

[0119] Preferably, the specific content of controlling the impurity removal structure to remove impurities is:

[0120] Construct the camera coordinate system, pixel coordinate system and impurity removal structure coordinate system;

[0121] Perform camera calibration to obtain the camera's intrinsic and extrinsic parameters to form a projection matrix;

[0122] Obtain the pixel coordinates of the weeds to be processed and convert the pixel coordinates into camera coordinates based on the projection matrix;

[0123] The camera coordinate system and the impurity removal structure coordinate system are in a nested relationship. The camera coordinates are substituted into the impurity removal structure coordinate system and the position distance between the impurities to be processed and the impurity removal structure is evaluated.

[0124] Calculate the number of steps that the impurity removal structure takes to move to the position of the weeds to be treated and sort them from low to high. Select the impurity removal structure that ranks first in the step number sorting for impurity removal.

[0125] First, the camera is calibrated to obtain the intrinsic and extrinsic parameters of the camera so that the pixel coordinates in the image can be converted to the actual world coordinates.

[0126] Internal reference:

[0127]

[0128] where f x and f y is the focal length, c x and c y are the coordinates of the principal point.

[0129] The external parameters include the rotation matrix R and the translation vector t, expressed as [R|t], where R is a 3*3 rotation matrix and t is a 3*1 translation matrix.

[0130] In general, the intrinsic and extrinsic parameters of a camera are combined to form a matrix, usually called a projection matrix or camera matrix, which is expressed as:

[0131] P = K[R|t];

[0132] This projection matrix can project points in three-dimensional space onto the two-dimensional image plane, realizing the conversion from the camera coordinate system to the pixel coordinate system.

[0133] Image processing and position detection:

[0134] Weed detection is performed on the captured image. A detection zone is set up on the image, corresponding to the position directly below the spray boom. When a target appears in the detection zone, the pixel coordinates of the target are obtained and converted into world coordinates through calibration parameters.

[0135] Assume that it is known that the foreign plants are in the image (x t ,y t ) coordinates, converting them into camera coordinates (X c , Y c , Z c ).

[0136] First, the pixel coordinates (x t ,y t ) is converted to normalized plane coordinates (u, v). This conversion process can be achieved by the following formula:

[0137]

[0138] Then convert the normalized plane coordinates into camera coordinates. This conversion process can be achieved by the following formula:

[0139]

[0140] Then the camera coordinates (X c , Y c , Z c ) is converted to nozzle coordinates (X p , Y p , Z p ,1),This conversion process can be achieved by the following formula:

[0141]

[0142] In this way, the position of the weeds in the nozzle can be determined, and the movement of the nozzle can be controlled to remove the weeds.

[0143] Compare the converted actual world coordinates with the current nozzle position to calculate the distance or position the nozzle needs to move.

[0144] Because the y and z axes of the nozzle have been set to fixed values, we only need to compare the x axis. Assuming that the actual position of the nozzle is X pi The position of the obtained hybrid is X p , the judgment process is:

[0145] X dist =X pi -X p ;

[0146] If X dist >0, then move X to the left dist ; Otherwise, move X to the rightdist .

[0147] Horizontal movement control:

[0148] The stepper motor controls the horizontal movement of the print head to move it to the target position, ensuring the precision and accuracy of the movement.

[0149] Real-time feedback and adjustments:

[0150] During the movement, the position X of the nozzle is monitored and recorded in real time. pi , provide feedback and adjustments based on actual conditions to ensure that the print head moves accurately to the target position.

[0151] An electronic device includes a memory and a processor, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, the content of the precise removal system of foreign plants in rice breeding based on intelligent image recognition is implemented.

[0152] A storage medium stores computer-executable instructions, which, when loaded and executed by a processor, implement the contents of the precise removal system for foreign plants in rice breeding based on intelligent image recognition.

[0153] It consists of two parts: a mobile platform and a weed removal device. The mobile platform is driven by a motor, and its wheels are controlled by a remote control to move in the field. The weed removal device consists of an industrial camera, an industrial computer, an I / O controller, a lifting rod, an electronically controlled sprinkler, a variable controller, a stepper motor, a medicine box, and a water pump. The industrial computer uses an industrial-grade microcontroller (MCU) or programmable logic controller (PLC). The industrial camera is located at the front end and can be one or more. The electronically controlled sprinkler is located below the industrial camera and its opening and closing are controlled by electrical signals. The vertical displacement is controlled by the lifting rod, and the horizontal displacement is controlled by the stepper motor. The industrial computer and I / O controller are installed in the control room to complete the weed detection and weed removal control. The medicine box and water pump provide the sprinkler with liquid medicine at a certain pressure, and the battery powers the entire device.

[0154] The nozzles are fine atomizers, mounted below the mobile unit, capable of spraying herbicides at precise angles and ranges, reaching directly above weeds. A herbicide tank stores the herbicide and is connected to a water pump via a pipe. The pump draws the herbicide from the tank and delivers it to the nozzles through a pipe.

[0155] This intelligent agricultural machinery device consists of five functional units, each of which operates in coordination through precise connections: the mobile unit includes an agricultural vehicle frame, four wheels, four motors, and a hydraulic steering gear. Each wheel is connected to the corresponding motor via a drive shaft. The motor is connected to the power supply system on the first floor of the control room via a power line and is controlled by a signal line and an industrial computer. The hydraulic steering gear controls the steering system via hydraulic pipelines. The control unit is located in the rear double-layer control room. The industrial computer installed on the second floor is connected to the industrial router via an Ethernet cable to build a communication network. At the same time, it is connected to the serial relay via the RS485 interface for command control. All equipment is powered by a first-floor power supply system. The image acquisition unit, located above the front-end spray boom, includes a high-definition camera, which transmits images directly to an industrial computer via a high-speed data cable. The impurity removal unit consists of a stepper motor and a sprinkler mounted on the boom. The stepper motor is connected to a serial relay via a control line. The sprinkler is connected to a pre-mounted water pump via a high-pressure pipeline, forming a spraying system. The water pump is connected to the spray tank via an inlet pipe and powered by the power supply system. The power supply unit, located on the first floor of the control room, consists of a battery pack, a power management system, and a charging port. It supplies power to the entire system via the main power bus and is equipped with an emergency stop switch for emergency response. All connections are waterproof and dustproof to ensure stable operation in farmland environments.

[0156] During weed removal, an industrial camera captures real-time video of the rice field and transmits it to an industrial computer. When a weed is detected, the weed detection algorithm sends its location to the weed removal control algorithm. Based on the weed's location, the algorithm determines the appropriate sprinkler head and the stepper motor's movement distance. The sprinkler head moves directly over the weed and sends an electrical signal to the I / O controller to activate the sprinkler head. The sprinkler head then sprays herbicide on the weeds, removing them.

[0157] After the image recognition and positioning system determines the position of the weeds, the control system calculates the distance and direction the mobile motor needs to move based on the coordinate information of the weeds, drives the mobile motor to move along the track, and allows the nozzle installed on the mobile device to move accurately to the top of the weeds.

[0158] The industrial computer runs the weed detection and removal control algorithm and is connected to the industrial camera, stepper motor, and I / O controller. The I / O controller controls the opening and closing of each nozzle. It receives weed location information from the weed detection algorithm and, based on a pre-programmed algorithm, coordinates the operation of the motors and the opening and closing of the nozzles, ensuring the accuracy and efficiency of the entire removal process. The control system also features data recording and storage, recording information such as the location, time, and amount of herbicide applied for each weed removal operation, facilitating subsequent analysis and management of weed distribution in the rice fields.

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical method of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical method to deviate from the spirit and scope of the technical method of the present invention.

Claims

1. A precise removal system for foreign plants in rice breeding based on intelligent image recognition, characterized in that: include: Image acquisition module: The image acquisition module is used to collect and pre-process images of rice plants in the paddy field, and transmit the pre-processed images of rice plants to the weed detection module; Weed plant detection module: The weed plant detection module is used to process the image features of rice field plants and analyze and identify the weed plants to be processed; Weed removal control module: The weed removal control module is used to receive the location of the weeds to be processed and control the weed removal structure to remove weeds.

2. The precise removal system for foreign plants in rice breeding based on intelligent image recognition according to claim 1, characterized in that: The specific contents of collecting and preprocessing rice field plant images include: During the collection process, the same area was collected at least three times; Obtain the current acquisition time and weather conditions, generate an environmental marker sequence, and mark the environmental marker sequence on the rice field plant image; Perform quality diagnosis on rice field plant images in the same area and sort them from high to low according to their quality; The rice field plant images with unqualified quality are eliminated, and the first half of the sorting process is screened to obtain the preprocessed rice field plant images.

3. The precise removal system for foreign plants in rice breeding based on intelligent image recognition according to claim 2, characterized in that: The foreign plant detection module includes: Standard model unit: the standard model unit includes standard trait indicators of several varieties; Feature extraction unit: The feature extraction unit is used to process defects and extract features from rice field plant images to obtain feature elements, and substitute the feature elements into the analysis and recognition unit. The feature extraction unit is based on YOLO 11 Model extraction; Analysis and identification unit: The analysis and identification unit obtains the characteristic elements and the standard trait indicators of the corresponding varieties in the standard model unit for comparison and analysis to obtain the weeds to be processed and traces back to obtain the rice field plant image and weed position where the weeds to be processed are located.

4. The precise removal system for foreign plants in rice breeding based on intelligent image recognition according to claim 3, characterized in that: The standard trait indicators of the varieties include tiller number index, plant height index and leaf area index; The tiller number index and plant height index are respectively provided with corresponding tiller number and plant height values and their fluctuation ranges; The leaf surface index is provided with a standard center point and a standard range point, as well as a radian value and a radian value fluctuation range corresponding to the standard range point.

5. The precise removal system for foreign plants in rice breeding based on intelligent image recognition according to claim 4, characterized in that: The content of defect processing for rice field plant images is as follows: The defect processing includes image enhancement processing and image correction processing; The image enhancement processing includes contrast adjustment, color correction and noise removal; The specific contents of the image correction processing are: Obtain an environmental tag sequence, extract and determine the time and weather of the photo based on the environmental tag sequence, and perform light and shadow error processing on the photo based on the time and weather.

6. The precise removal system for foreign plants in rice breeding based on intelligent image recognition according to claim 5, characterized in that: The specific content of light and shadow error processing of pictures according to time and weather is as follows: Defining images as cloudy, rainy, and sunny labels according to weather conditions, wherein the cloudy, rainy, and sunny labels are respectively matched with equalization processing parameters, wherein the equalization processing parameters include gamma correction parameters and color adjustment parameters; The processing method for the cloudy day label includes: using the color adjustment parameters under the cloudy day label to adjust the color of the image to obtain a cloudy day image; Use the gamma correction parameters under the cloudy tag to correct the cloudy image to obtain the cloudy corrected image; The processing method for rainy day labels includes: using the deconvolution algorithm to perform inverse operations on the image to obtain a rainy day corrected image; The processing method for the sunny day tag includes: obtaining the gamma correction parameters and image time under the sunny day tag; Determine the solar altitude angle of the current image based on the image time, and determine the light intensity based on the solar altitude angle; According to the light intensity, configure the adjustment factor for the gamma correction parameter under the sunny label to obtain the gamma adjustment parameter; The image is corrected according to the gamma adjustment parameters to obtain a sunny day corrected image.

7. The system for accurately removing foreign plants in rice breeding based on intelligent image recognition according to claim 5, characterized in that: The analysis and identification unit obtains the characteristic elements and the standard trait indicators of the corresponding varieties in the standard model unit for comparison and analysis to obtain the specific content of the offal plants to be processed: Determine the actual number of tillers, actual plant height and leaf outline among the characteristic elements; Confirm whether the actual number of tillers and actual plant height are within the fluctuation range; If yes, it will be retained; if one of the items is not, it will be determined as a weed to be treated; The leaf contour is fixed with the standard center point as the fixed point. If the standard range point falls on the leaf contour, it is retained. If the standard range point does not fall on the leaf contour, it is determined to be a weed to be treated; Calculate the radian value of the leaf surface contour. If the radian value of the leaf surface contour does not exceed the radian value fluctuation range, collect the actual leaf surface contour into the standard model unit. If the curvature value of the leaf surface contour exceeds the curvature value fluctuation range, it is determined to be a weed to be treated; If a flower bud is identified in the process of acquiring the characteristic elements by the analysis and identification unit, the spider carrying the flower bud is determined to be a weed to be processed.

8. The system for accurately removing foreign plants in rice breeding based on intelligent image recognition according to claim 5, characterized in that: The specific contents of controlling the impurity removal structure to remove impurities are: Construct the camera coordinate system, pixel coordinate system and impurity removal structure coordinate system; Perform camera calibration to obtain the camera's intrinsic and extrinsic parameters to form a projection matrix; Obtain the pixel coordinates of the weeds to be processed and convert the pixel coordinates into camera coordinates based on the projection matrix; The camera coordinate system and the impurity removal structure coordinate system are in a nested relationship. The camera coordinates are substituted into the impurity removal structure coordinate system and the position distance of the impurities to be processed in the impurity removal structure is evaluated; Calculate the number of steps that the impurity removal structure takes to move to the position of the weeds to be treated and sort them from low to high. Select the impurity removal structure that ranks first in the step number sorting for impurity removal.

9. An electronic device, characterized in that: It includes a memory and a processor, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, it implements the content of the precise removal system of foreign plants in rice breeding based on intelligent image recognition as described in any one of claims 1 to 8.

10. A storage medium, characterized in that: The storage medium stores computer-executable instructions, which, when loaded and executed by the processor, implement the contents of the precise removal system for foreign plants in rice breeding based on intelligent image recognition as claimed in any one of claims 1 to 7.

Images