A corn field scene-oriented laser weeding method and system

Abstract

The application relates to a corn field scene-oriented laser weeding method and system. The method comprises the following steps: acquiring corn field operation area images collected by a mobile operation platform and mobile platform state information, performing pixel-level segmentation and weed instance separation on the images to obtain a plurality of single-plant weed regions; extracting candidate striking points according to the contour features and skeleton features of the single-plant weed regions, performing safety screening in combination with the spatial distance between the candidate striking points and corn plant regions, the laser action range and the stability degree of the candidate striking points, and determining target striking points; converting the target striking points into ground operation coordinates, performing motion compensation in combination with the mobile platform state information and system time delay, obtaining laser execution coordinates, and controlling a laser to perform pinpoint striking on single-plant weeds. The application can improve the positioning accuracy and operation safety of laser weeding in a corn field scene.

Description

Technical Field

[0001] This application relates to the fields of visual recognition and laser weed control technology, and in particular to a laser weeding method and system for cornfield scenarios. Background Technology

[0002] Laser weeding is a non-contact weeding technology that uses a laser beam to target the growing point, stem base, or other key tissues of weeds, thereby disrupting their growth. Compared to manual, chemical, and mechanical weeding, laser weeding has advantages such as no chemical residue, a high degree of automation, and minimal soil disturbance, making it suitable for integration with agricultural robots or mobile platforms. The effectiveness of laser weeding largely depends on the accurate identification of crops and weeds and the precise determination of the laser's target location.

[0003] In cornfield scenarios, corn seedlings and some weeds are quite similar in color, size, and morphology. However, factors such as overlapping leaves, plant shading, soil, stubble, shadows, and changes in light intensity further complicate the differentiation between corn plants and weeds. Existing methods for weed identification in fields mainly include traditional image processing methods based on color, texture, and vegetation indices, as well as identification methods based on object detection or image segmentation. Traditional image processing methods are sensitive to changes in lighting and background, making them ill-suited to complex field environments. Object detection methods typically only output bounding boxes, failing to reflect the true contours of weeds. While image segmentation methods can obtain pixel-level regions, they usually require further separation of individual weeds and extraction of laser-induced locations.

[0004] Furthermore, laser weeding requires not only identifying the area where weeds are located, but also determining the stem base, growing point, or other key locations suitable for laser action, and converting the target location in the image into actual working coordinates usable by the laser actuator. During mobile operations, the mobile platform moves continuously, and there is a time interval between image acquisition, image processing, and laser execution, which can easily cause the target to deviate from its original identified position when the laser is applied. Summary of the Invention

[0005] To overcome, to some extent, the problems of insufficient accuracy in weed identification and laser targeting in related technologies, this application provides a laser weeding method and system for cornfield scenarios.

[0006] The proposed solution is as follows:

[0007] According to a first aspect of the embodiments of this application, a laser weeding method for cornfield scenarios is provided, comprising: The system acquires images of the cornfield operation area collected by the mobile operation platform and the status information of the mobile platform, including platform speed, direction of travel, platform attitude, and image acquisition time. The cornfield operation area image is segmented at the pixel level to obtain corn plant areas and weed areas, and the weed areas are further separated into multiple individual weed areas. Based on the contour and skeletal features of each individual weed area, candidate points at the base of the stem, candidate points at the growth point, or candidate points at the lower end of the main skeleton are extracted to form a set of candidate attack points. Safety screening is conducted based on the spatial distance between each candidate strike point and the corn plant area, the laser's effective range, and the stability of the candidate strike points. Candidate strike points that pose a risk of accidentally damaging corn plants are eliminated, their positions are adjusted, or they are replaced by backup candidate strike points to determine the target strike point. Based on the calibration results of the image acquisition device and the spatial pose relationship between the image acquisition device and the laser, the target impact point is converted into ground operation coordinates; Based on the mobile platform status information and the system time delay from the image acquisition time to the laser action time, motion compensation is performed on the ground operation coordinates to obtain the laser execution coordinates; The laser is controlled to strike the corresponding individual weeds based on the laser's execution coordinates.

[0008] Preferably, before performing pixel-level segmentation on the image of the cornfield work area, the method further includes: The distortion correction of the cornfield operation area image is performed according to the calibration parameters of the image acquisition device to eliminate the imaging distortion generated by the image acquisition device; Based on the brightness distribution and color deviation of the cornfield operation area image, white balance correction, illumination normalization, and noise reduction are performed to obtain a corrected image; The corrected image is subjected to color space conversion and size unification to obtain the image to be segmented, and the image to be segmented is used as the input for pixel-level segmentation.

[0009] Preferably, the method further includes: Obtain re-inspection images after pinpoint strikes; The strike result is determined based on the area of ​​residual weeds in the re-inspection image; If the strike result does not meet the preset requirements, the target strike point, motion compensation parameters, or laser execution parameters are adjusted before a supplementary strike is performed.

[0010] Preferably, determining the strike result based on the area of ​​residual weeds in the re-inspection image includes: The re-inspection image is segmented at the pixel level to obtain the residual weed area corresponding to the targeted strike; Determine the pixel area of ​​the residual weed region and compare the pixel area with a preset residual area threshold; When the pixel area is not greater than the preset residual area threshold, the impact result is determined to meet the preset requirements; When the pixel area is greater than the preset residual area threshold, it is determined that the strike result does not meet the preset requirements, and the target strike point is adjusted according to the position of the residual weed area relative to the target strike point; The laser execution parameters are adjusted based on the difference between the pixel area and the preset residual area threshold, and the laser execution coordinates are re-determined by combining the adjusted target impact point, motion compensation parameters and laser execution parameters. Supplementary strikes will be carried out based on the newly determined laser execution coordinates.

[0011] Preferably, the image of the cornfield operation area is segmented at the pixel level to obtain corn plant areas and weed areas, including: The image of the cornfield operation area is input into a lightweight semantic segmentation model. The lightweight semantic segmentation model is used to extract lightweight features from the image of the cornfield operation area to obtain image features at different scales. Feature fusion is performed on image features at different scales, and boundary enhancement and attention enhancement are applied to the fused image features to obtain enhanced image features; Based on the enhanced image features, pixel-level spatial information is recovered, and the region category corresponding to each pixel is determined to obtain the corn plant region, the weed region, and the background interference region. Based on the adjacency relationship between the corn plant area and the weed area, a high-risk boundary area near the corn is determined, and the high-risk boundary area near the corn is used for safe screening of the candidate attack points.

[0012] Preferably, the weed areas are separated into multiple individual weed areas, including: Morphological processing is performed on the weed area to remove noisy areas and fill holes in the weed area; Connectivity analysis was performed on the morphologically treated weed regions to obtain multiple connected weed regions. Based on the area, aspect ratio, and structural continuity of each weed connected region, weed connected regions that do not meet the preset weed region conditions are eliminated. When any connected weed region contains multiple weed targets that are interconnected, the connected weed region is divided into watershed segments or skeleton decompositions to separate the connected weed region into multiple single-weed regions.

[0013] Preferably, candidate points at the base of the stem, candidate points at the growth point, or candidate points at the lower end of the main framework are extracted based on the contour and skeletal features of each individual weed area to form a set of candidate attack points, including: Extract the outline of each individual weed region, and determine the geometric center point, centroid point, and minimum circumscribed region center point of the corresponding individual weed region based on the outline. The skeleton of each individual weed is extracted to obtain the weed skeleton, and the skeleton end point, skeleton bifurcation point and skeleton node near the ground surface are extracted from the weed skeleton. Based on the main axis direction of the weed skeleton, the top and bottom boundaries of the single weed area, the candidate points at the stem base, the candidate points at the growth point, and the candidate points at the lower end of the main skeleton are determined. Candidate priority is determined according to the order of candidate points at the base of the stem, candidate points at the growth point, candidate points at the lower end of the main skeleton, and the geometric center point, thus forming the set of candidate strike points.

[0014] Preferably, safety screening is performed based on the spatial distance between each candidate impact point and the corn plant area, the laser's effective range, and the stability of the candidate impact points, including: Determine the minimum spatial distance between each candidate strike point and the corn plant area, and expand the corn plant area according to the laser's effective range to obtain a safe corn zone; Determine whether each candidate strike point is located within the cornfield safety area, and determine the degree of positional change of each candidate strike point based on continuously collected images of the cornfield operation area, so as to determine the stability of the candidate strike points; Candidate strike points located outside the corn safety zone and whose stability meets the preset stability requirements are identified as safe candidate strike points. When no safe candidate strike point exists, the candidate strike point is moved away from the corn plant area, or a backup candidate strike point is selected from the set of candidate strike points; if neither the candidate strike point after position adjustment nor the backup candidate strike point meets the safety requirements, the targeted strike on the corresponding individual weed is canceled.

[0015] Preferably, the step of converting the target strike point into ground operation coordinates and performing motion compensation on the ground operation coordinates includes: Based on the internal calibration parameters of the image acquisition device and the external calibration parameters between the image acquisition device and the laser, a coordinate mapping relationship between the image coordinate system and the ground operation coordinate system is established. Based on the coordinate mapping relationship, the image coordinates corresponding to the target impact point are converted into the ground operation coordinates, and the ground operation coordinates are corrected according to the platform attitude. Based on the image acquisition time, the expected laser action time, the platform speed, and the direction of travel, the displacement of the mobile work platform within the system time delay is determined; Based on the displacement, the ground work coordinates after attitude correction are compensated for forward look; when the platform speed or the platform attitude changes, the compensated ground work coordinates are predicted and corrected in combination with the platform acceleration of the mobile work platform to obtain the laser execution coordinates.

[0016] According to a second aspect of the embodiments of this application, a laser weeding system for cornfield scenarios is provided, comprising: Mobile work platform, screw lifting mechanism, image acquisition device, 2D servo gimbal and laser; The mobile operating platform is equipped with a processor and a memory; The processor and memory are connected via a communication bus: The processor is used to call and execute the program stored in the memory; The memory is used to store a program, which is at least used to execute a laser weeding method for a cornfield scenario as described in any of the above.

[0017] The technical solution provided in this application may include the following beneficial effects: This application performs pixel-level segmentation on the image of the cornfield operation area and instance separation of the weed area. It can obtain the area of ​​a single weed even when corn plants, weeds and field background overlap. This avoids the loss of contour information caused by determining the location of weeds only by the target box, and provides an accurate target basis for subsequent plant-by-plant positioning and targeted strikes.

[0018] Furthermore, candidate points at the stem base, growth point, or lower end of the main framework are extracted based on the contour and skeletal features of individual weed areas. This refines the laser's target location from the weed's general area to key tissue sites within the weed, improving the effective utilization of laser energy and increasing the success rate of a single strike. By combining the spatial distance between candidate strike points and the corn plant area, the laser's effective range, and the stability of the candidate strike points for safety screening, and by eliminating, repositioning, or replacing candidate strike points with a risk of accidental damage, the risk of accidental damage when corn plants and weeds are adjacent or partially overlapped can be reduced.

[0019] By using the calibration results of the image acquisition device and the spatial pose relationship between the image acquisition device and the laser, the target strike point is converted into ground operation coordinates, realizing the conversion from the image recognition position to the actual position of the laser. Then, by combining the mobile platform status information and system latency to perform motion compensation on the ground operation coordinates, the target position offset caused by the movement and attitude changes of the mobile operation platform, as well as the latency of image processing and laser execution can be corrected, so that the laser can still be aimed at the corresponding single weed at the actual moment of action.

[0020] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0021] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0022] Figure 1 This is a schematic flowchart of a laser weeding method for a cornfield scenario provided in one embodiment of this application; Figure 2 This is a schematic flowchart of a laser weeding method for a cornfield scenario provided in another embodiment of this application; Figure 3 This is a schematic diagram of the structure of a laser weeding system for cornfield scenarios provided in one embodiment of this application.

[0023] Reference numerals: Mobile work platform-21; Screw lifting mechanism-22; Image acquisition device-23; Two-dimensional servo gimbal-24; Laser-25. Detailed Implementation

[0024] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0025] Example 1 Figure 1 This is a flowchart illustrating a laser weeding method for cornfield scenarios provided in one embodiment of this application. (Refer to...) Figure 1 A laser weeding method for cornfield applications includes: S101. Acquire images of the cornfield operation area and mobile platform status information collected by the mobile operation platform. The mobile platform status information includes platform speed, direction of travel, platform attitude, and image acquisition time. S102. Perform pixel-level segmentation on the cornfield operation area image to obtain corn plant area and weed area, and perform instance separation on the weed area to obtain multiple single weed areas. S103. Based on the contour and skeleton features of each individual weed area, extract candidate points at the base of the stem, candidate points at the growth point, or candidate points at the lower end of the main skeleton to form a set of candidate attack points. S104. Based on the spatial distance between each candidate strike point and the corn plant area, the laser range, and the stability of the candidate strike points, a safety screening is conducted. Candidate strike points that pose a risk of accidental damage to corn plants are eliminated, their positions are adjusted, or they are replaced by backup candidate strike points to determine the target strike point. S105. Based on the calibration results of the image acquisition device and the spatial pose relationship between the image acquisition device and the laser, the target strike point is converted into ground operation coordinates. S106. Based on the mobile platform status information and the system time delay from the image acquisition time to the laser action time, perform motion compensation on the ground operation coordinates to obtain the laser execution coordinates. S107. Based on the laser's execution coordinates, control the laser to precisely strike the corresponding individual weeds.

[0026] For ease of understanding, the following explains some key terms in this embodiment: A mobile work platform configured to perform mobile operations in a cornfield environment. Figure 3 This is a schematic diagram of a laser weeding system for cornfield scenarios provided in one embodiment of this application, with reference to... Figure 3 The mobile work platform can be a vehicle body equipped with a screw jack, image acquisition device, two-dimensional servo gimbal and laser, etc.

[0027] Preferably, the image acquisition device uses an RGB industrial camera, mounted on the front or upper support of the mobile work platform, with the lens facing downwards and forwards of the work area, ensuring its field of view covers the corn rows to be processed. When ambient light is insufficient or changes significantly, a supplementary lighting unit can be installed to illuminate the work area, enhancing image clarity and color consistency.

[0028] The image of the cornfield operation area is defined as field image data containing corn plants and weeds, acquired by an image acquisition device during the movement of a mobile operation platform.

[0029] Mobile platform status information is data used to describe the current motion status of the mobile work platform, including the platform's speed, direction of travel, the platform's own posture, and the specific time point when the image was acquired.

[0030] Pixel-level segmentation is used to classify each pixel in an image and assign it to a specific semantic category, such as corn plants, weeds, or background.

[0031] Instance separation is used to further distinguish multiple weed individuals that are in contact with or overlap each other after weed areas have been identified, so that each individual weed individual is identified as a separate instance.

[0032] The candidate strike point set is defined as a set of multiple locations that may be suitable for laser strikes, initially determined based on the morphological characteristics of a single weed area.

[0033] The target strike point is defined as the precise location that, after safety screening, is ultimately determined for use in a laser-based targeted strike.

[0034] Ground-based coordinates are defined as a coordinate system established in actual physical space with the ground as the reference frame, used to accurately indicate the position where the laser is operating.

[0035] System latency is defined as the time interval between the moment the image is acquired and the moment the laser actually acts on the target. This latency includes the time required for image processing, decision-making, and laser response.

[0036] Motion compensation is used to correct the position of the target impact point in the image coordinate system or ground coordinate system based on the motion state of the mobile work platform and system latency, so as to offset the target position offset caused by the platform motion.

[0037] Precision strike is defined as the precise application of a laser beam to a target point based on calculated laser coordinates, thereby disrupting the growth of weeds.

[0038] In this embodiment, firstly, images of the cornfield work area and the status information of the mobile work platform are acquired. Specifically, an image acquisition device is installed on the mobile work platform and configured to periodically capture images of the cornfield. The mobile platform itself is equipped with sensors, such as a Global Positioning System (GPS) receiver and an Inertial Measurement Unit (IMU), to acquire the platform's speed, direction of travel, and attitude. Simultaneously, the image acquisition device records the moment of image acquisition each time an image is captured. This information is either directly stored in local memory or transmitted to a processing unit for further processing.

[0039] Furthermore, the image of the cornfield operation area is segmented pixel-wise to obtain corn plant regions and weed regions. The weed regions are then separated into individual weed regions. Specifically, the image is input into a pre-trained image segmentation model configured to identify different semantic categories in the image. For example, each pixel in the image is classified as a corn plant, weed, or background using a color thresholding method or a simple machine learning classifier. After obtaining the weed regions, instance separation can be performed using a connected component analysis method.

[0040] Based on this, candidate points are extracted from the stem base, growth point, or lower end of the main framework according to the contour and skeletal features of each individual weed area, forming a set of candidate attack points. Based on these features, rules can be preset, for example, using the lowest point of the contour, a specific node of the framework, or the geometric center point as preliminary candidate attack points.

[0041] Subsequently, a safety screening process is conducted based on the spatial distance between each candidate impact point and the corn plant area, the laser's effective range, and the stability of the candidate impact points. Candidate impact points posing a risk of accidental damage to the corn plant are eliminated, repositioned, or replaced with backup candidate impact points to determine the target impact point. Specifically, the Euclidean distance from each candidate impact point to the nearest corn plant area is calculated. A fixed safety distance threshold is set; if the distance between a candidate impact point and the corn plant is less than this threshold, the point is considered to pose a risk of accidental damage. Risky candidate impact points can be directly eliminated or their positions can be fine-tuned to move them away from the corn plant. The stability of the candidate impact point can be evaluated by simply averaging its position over several consecutive image frames to eliminate misjudgments caused by image jitter or unstable recognition.

[0042] Furthermore, based on the calibration results of the image acquisition device and the spatial pose relationship between the image acquisition device and the laser, the target impact point is converted into ground operation coordinates. Specifically, the internal calibration parameters of the image acquisition device (such as focal length and principal point coordinates) and its external calibration parameters with the laser (such as relative translation and rotation relationships) can be obtained in advance through manual measurement or a simple calibration board method. By applying these calibration parameters and geometric transformation matrices, the target impact point in the image coordinate system is projected onto the actual ground coordinate system.

[0043] Based on this, motion compensation is performed on the ground operation coordinates according to the mobile platform's status information and the system delay from the image acquisition time to the laser application time to obtain the laser execution coordinates. Specifically, the mobile platform's status information (such as platform speed and direction of travel) and the system delay are known. Motion compensation can be achieved through a simple linear extrapolation method, that is, based on the platform's average speed and direction within the system delay, the target's displacement in the ground coordinate system is calculated and added to the transformed ground operation coordinates to predict the target's actual position at the laser application time.

[0044] Finally, the laser is controlled to precisely strike the corresponding individual weeds based on the laser's execution coordinates. Specifically, the laser is configured to receive the calculated laser execution coordinates. By controlling the laser's scanning galvanometer or robotic arm, its laser beam is precisely pointed at the physical location indicated by the execution coordinates, and a brief laser irradiation is performed, thereby destroying the weeds.

[0045] This technical solution performs pixel-level segmentation on the image of the cornfield operation area and separates the weed areas into instances. It can obtain the area of ​​a single weed even when corn plants, weeds and field background overlap. This avoids the loss of contour information caused by determining the location of weeds only by the target box, and provides an accurate target basis for subsequent plant-by-plant positioning and targeted strikes.

[0046] Furthermore, candidate points at the stem base, growth point, or lower end of the main framework are extracted based on the contour and skeletal features of individual weed areas. This refines the laser's target location from the weed's general area to key tissue sites within the weed, improving the effective utilization of laser energy and increasing the success rate of a single strike. By combining the spatial distance between candidate strike points and the corn plant area, the laser's effective range, and the stability of the candidate strike points for safety screening, and by eliminating, repositioning, or replacing candidate strike points with a risk of accidental damage, the risk of accidental damage when corn plants and weeds are adjacent or partially overlapped can be reduced.

[0047] By using the calibration results of the image acquisition device and the spatial pose relationship between the image acquisition device and the laser, the target strike point is converted into ground operation coordinates, realizing the conversion from the image recognition position to the actual position of the laser. Then, by combining the mobile platform status information and system latency to perform motion compensation on the ground operation coordinates, the target position offset caused by the movement and attitude changes of the mobile operation platform, as well as the latency of image processing and laser execution can be corrected, so that the laser can still be aimed at the corresponding single weed at the actual moment of action.

[0048] Example 2 This embodiment further proposes a preprocessing step for the image before pixel-level segmentation of the cornfield operation area image, specifically including: The distortion correction of the cornfield operation area image is performed according to the calibration parameters of the image acquisition device to eliminate the imaging distortion generated by the image acquisition device; Based on the brightness distribution and color deviation of the cornfield operation area image, white balance correction, illumination normalization and noise reduction are performed to obtain the corrected image; The corrected image is subjected to color space conversion and size unification to obtain the image to be segmented, and the image to be segmented is used as the input for pixel-level segmentation.

[0049] Specifically, during the imaging process, image acquisition devices experience radial and tangential distortion due to limitations in lens design and manufacturing, resulting in bent lines and distorted object shapes in the image. Distortion correction aims to utilize pre-obtained internal parameters (such as focal length, principal point coordinates, and distortion coefficients) obtained through camera calibration to construct a mathematical model for geometric transformation of image pixels. This maps the distorted pixel positions back to their true positions under ideal imaging conditions, effectively eliminating these imaging distortions, ensuring the geometric accuracy of the image, and providing a reliable image basis for subsequent precise measurements and positioning.

[0050] Furthermore, the color representation of images from cornfield work areas can vary significantly under different lighting conditions. For example, images may appear generally cooler or warmer under cloudy skies or direct sunlight. White balance correction analyzes the color distribution of the image and adjusts the gain of the red, green, and blue channels to make white areas in the image appear pure white, thereby correcting the overall color cast and restoring the true colors of the scene. Simultaneously, images may exhibit localized overexposure or underexposure. Illumination normalization adjusts the brightness range and contrast of the image to ensure similar visual effects under different lighting conditions, enhancing the image's robustness. Images may also be subject to various noise interferences during acquisition and transmission. Denoising processing applies various filtering algorithms (such as mean filtering, median filtering, Gaussian filtering, etc.) to effectively remove random noise from the image while preserving image details and edge information as much as possible, smoothing the image and improving its clarity.

[0051] Furthermore, original images are typically represented in the RGB color space. For better image analysis, especially for plant identification, it's sometimes necessary to convert images to other color spaces, such as HSV, Lab, or YCrCb. Different color spaces represent color information differently. For example, the HSV space separates color and brightness information, helping to maintain the stability of color features under varying lighting conditions, thus enabling better extraction of plant color features. Simultaneously, to adapt to the input requirements of subsequent pixel-level segmentation models and improve processing efficiency and model generalization ability, image resizing is required, adjusting all images to a preset uniform resolution or size.

[0052] Example 3 Figure 2 This is a flowchart illustrating a laser weeding method for cornfield scenarios provided in another embodiment of this application, referred to below. Figure 2 The methods also include: S108. Obtain re-inspection images after pinpoint strikes; S109. Determine the strike results based on the areas of residual weeds in the re-inspection images; S110. If the strike result does not meet the preset requirements, adjust the target strike point, motion compensation parameters or laser execution parameters and then perform a supplementary strike.

[0053] Specifically, after the initial targeted strike on the target weeds, the system will use an image acquisition device to obtain an image of the struck area again, i.e., a re-inspection image.

[0054] This step aims to provide an immediate or near real-time assessment of the laser strike effect to determine whether the weeds have been effectively cleared. The image acquisition device can be the same as the one used for the initial image acquisition, or it can be a specialized auxiliary device for re-inspection. Acquisition is typically performed after the laser strike action is completed and the mobile work platform has moved to a suitable position.

[0055] After acquiring the re-inspection image, the system will analyze the image to identify whether there are areas of weeds that have not been completely removed, i.e., areas with residual weeds.

[0056] By quantitatively analyzing the morphology, size, and location of the remaining weed areas and comparing them with the preset weed control effect requirements, the final result of this targeted weed control can be determined. For example, a residual area threshold can be set; if the area of ​​the remaining weed area is less than this threshold, the control result is considered to meet the requirements; otherwise, it is considered not to meet the requirements.

[0057] When the system determines that the initial targeted strike has failed to achieve the preset weeding effect, a supplementary strike mechanism will be activated to ensure that the weeds are completely removed.

[0058] At this point, the system will correct the position of the original target strike point based on the specific situation of the residual weed area in the re-inspection image, so that it is more accurately aligned with the residual part; at the same time, considering the real-time status of the mobile operation platform and system latency, the motion compensation parameters may be fine-tuned to ensure that the laser beam can be accurately projected onto the corrected strike point.

[0059] In addition, laser execution parameters can be adjusted, such as increasing laser energy, extending the action time, or changing the laser mode, to improve the weeding effect. By adjusting these parameters, the system can perform secondary or multiple additional strikes on residual weeds until the preset weeding requirements are met.

[0060] Specifically, the strike results are determined based on the areas of residual weeds in the re-examination images, including: The re-inspection images are segmented at the pixel level to obtain the residual weed areas corresponding to the targeted strikes; Determine the pixel area of ​​the residual weed region and compare the pixel area with a preset residual area threshold; When the pixel area is not greater than the preset residual area threshold, the strike result is determined to meet the preset requirements; When the pixel area is greater than the preset residual area threshold, it is determined that the strike result does not meet the preset requirements, and the target strike point is adjusted according to the position of the residual weed area relative to the target strike point. The laser execution parameters are adjusted based on the difference between the pixel area and the preset residual area threshold. The laser execution coordinates are then re-determined by combining the adjusted target impact point, motion compensation parameters, and laser execution parameters. Supplementary strikes will be carried out based on the newly determined laser execution coordinates.

[0061] Specifically, when performing pixel-level segmentation on the re-inspection image to obtain the residual weed region corresponding to the targeted strike, this step aims to accurately identify and distinguish the residual weeds that may exist in the image after the targeted strike. Pixel-level segmentation is an image processing technique that aims to classify each pixel in an image into a predefined category, such as "residual weeds" or "background." In this way, the precise boundary and shape information of the residual weed region can be obtained. Implementation methods may include, but are not limited to: using deep learning models (such as semantic segmentation networks) to train and infer on the re-inspection image, or using traditional image processing algorithms that combine color and texture features. For example, first, color space conversion is performed to enhance the contrast between weeds and the background; then, threshold segmentation and morphological operations (such as opening and closing operations) are applied to remove noise and connect broken regions; finally, independent residual weed regions are extracted through connected component analysis.

[0062] When determining the pixel area of ​​the residual weed region and comparing it with a preset residual area threshold, a quantitative evaluation is required after the residual weed region is identified. The pixel area is typically determined by counting the number of pixels contained within the residual weed region. This number of pixels directly reflects the actual size of the residual weeds. Subsequently, the calculated pixel area is compared with a preset "residual area threshold." This threshold represents the maximum acceptable area of ​​residual weeds, and its setting can be based on agronomic standards, weed control efficiency requirements, or optimized through experimental data.

[0063] When the pixel area is not greater than the preset residual area threshold, the strike result is determined to meet the preset requirements. This step defines the criteria for determining the success of laser targeted strikes. When the pixel area of ​​the residual weed area is less than or equal to the preset residual area threshold, it indicates that the first strike has cleared the target weeds to an acceptable degree, or the residual portion is so small that it does not affect crop growth. In this case, the system determines that the current strike result is satisfactory, and no further supplementary strikes are required.

[0064] When the pixel area exceeds the preset residual area threshold, it is determined that the strike result does not meet the preset requirements, and the target strike point is adjusted according to the position of the residual weed area relative to the target strike point. When the pixel area of ​​the residual weed area exceeds the preset residual area threshold, it indicates that the first strike failed to completely remove the target weeds, and a supplementary strike is required. To improve the efficiency and accuracy of the supplementary strike, the original target strike point needs to be adjusted. The adjustment is based on the spatial position of the residual weed area relative to the original target strike point. For example, the geometric center or centroid of the residual weed area can be calculated and used as the new target strike point, or the target strike point can be shifted towards the center of the residual weed area to ensure that the supplementary strike can accurately cover the remaining weeds.

[0065] When adjusting the laser execution parameters based on the difference between the pixel area and the preset residual area threshold, and redetermining the laser execution coordinates by combining the adjusted target impact point, motion compensation parameters, and laser execution parameters, in addition to adjusting the impact position, it is also necessary to optimize the laser's impact intensity and method based on the severity of the residual weeds (i.e., the difference between the pixel area and the threshold). For example, if the residual area is much larger than the threshold, it may be necessary to increase the laser power, extend the action time, or change the laser scanning mode to ensure more thorough removal. These adjustments to the laser execution parameters are dynamic and designed to match actual weed control needs. Subsequently, the adjusted target impact point, updated motion compensation parameters (considering that the platform may continuously move), and adjusted laser execution parameters are combined to recalculate the laser execution coordinates for supplementary impacts. This process ensures that supplementary impacts are not only accurately positioned but also optimized in terms of energy and method.

[0066] Finally, supplementary strikes are performed based on the redefined laser execution coordinates. After all the parameters have been fine-tuned and the coordinates redefined, the laser will perform a second or more supplementary strikes on the remaining weeds based on these latest laser execution coordinates. This step is the final execution stage of the entire closed-loop feedback control system, designed to thoroughly remove weeds that were not completely removed in the first strike, thereby achieving the desired weed control effect.

[0067] Example 4 This embodiment further proposes to perform pixel-level segmentation on the cornfield operation area image to obtain corn plant areas and weed areas, including: The image of the cornfield operation area is input into the lightweight semantic segmentation model. The lightweight semantic segmentation model performs lightweight feature extraction on the cornfield operation area image to obtain image features at different scales. Feature fusion is performed on image features at different scales, and boundary enhancement and attention enhancement are applied to the fused image features to obtain enhanced image features; Based on the enhanced image features, pixel-level spatial information is recovered, and the region category corresponding to each pixel is determined to obtain the corn plant region, weed region, and background interference region. The high-risk boundary area of ​​corn neighbors was determined based on the adjacency relationship between corn plant areas and weed areas, and this high-risk boundary area of ​​corn neighbors was used for safe screening of candidate strike points.

[0068] Preferably, the lightweight semantic segmentation model in this embodiment includes a lightweight encoder, a multi-scale feature fusion module, a boundary enhancement branch, an attention enhancement unit, and a decoding and restoration structure. The lightweight encoder reduces the number of parameters and computational cost; the multi-scale feature fusion module simultaneously captures the scale features of both small and large weeds; the boundary enhancement branch enhances the distinguishing ability between weeds and corn leaf edges; the attention enhancement unit highlights key difference regions between weeds and crops; and the decoding and restoration structure restores the spatial details of the segmentation mask.

[0069] The model output categories include at least the corn plant area, weed area, soil background area, and shadow or stubble interference area. For each pixel in the image... The model outputs its category label. : ; in, Represents pixels Belongs to the The probability of a class This represents the total number of categories.

[0070] The set of pixels representing weeds can be represented as: ; in Labels for weeds.

[0071] The feature fusion step effectively integrates image features extracted from different scales to aggregate multi-level information. Fusion methods can include simple feature concatenation, element-wise addition, or more complex pyramid feature networks (FPN) structures. Through feature fusion, the model can simultaneously utilize high-resolution detail information and low-resolution semantic information, thereby improving the precision and accuracy of the segmentation results.

[0072] The boundary enhancement aims to improve the model's ability to recognize object boundaries, especially when the boundary between corn plants and weeds is blurred or overlapping. This can be achieved by introducing a boundary loss term into the loss function or by designing a dedicated boundary detection module, allowing the model to focus more on the classification accuracy of boundary pixels during training.

[0073] The attention enhancement mechanism allows the model to dynamically focus on regions or feature channels in the image that are more important for the segmentation task. For example, through channel attention mechanisms (such as SE-Net) or spatial attention mechanisms (such as CBAM), the model can adaptively adjust the weights of different feature channels or spatial locations, thereby suppressing background noise, highlighting the key features of corn plants and weeds, and further improving the robustness and accuracy of segmentation.

[0074] The features processed through feature fusion, boundary enhancement, and attention enhancement are referred to as enhanced image features, which contain richer and more discriminative information. These features not only integrate multi-scale information but also enhance boundary details and highlight key regions, providing a solid foundation for subsequent accurate pixel-level classification.

[0075] This embodiment further proposes to separate the weed areas into instances, obtaining multiple single-plant weed areas, including: Morphological treatment was performed on the weed areas to remove noisy areas and fill holes in the weed areas; Connectivity analysis was performed on the morphologically treated weed regions to obtain multiple connected weed regions. Based on the area, aspect ratio, and structural continuity of each connected weed area, weed connected areas that do not meet the preset weed area conditions are removed. When any connected weed region contains multiple weed targets that are interconnected, the connected weed region is divided into watershed segments or skeleton decompositions to separate the connected weed region into multiple single-weed regions.

[0076] Specifically, morphological processing can employ opening operations to eliminate isolated noise points or small connections in the image. For example, by performing an erosion followed by dilation operation, tiny noise regions smaller than a preset structuring element can be effectively removed, thereby improving the purity of the weed region. Simultaneously, closing operations can be used to fill small holes or breaks within the weed region. For instance, by performing a dilation followed by erosion operation, the interior of the weed region can be made more complete, avoiding errors in subsequent analysis caused by holes. The shape and size of the structuring elements for these morphological operations can be optimized based on the actual morphological characteristics of the weeds and the image resolution to achieve the best denoising and hole-filling effects.

[0077] Subsequently, connected component analysis was performed on the morphologically processed weed region, resulting in multiple connected weed regions. Connected component analysis is a technique that identifies all interconnected sets of pixels in an image as independent regions. Through this step, the morphologically processed weed region can be divided into several independent, unconnected sets of pixels, each set being considered a potential connected weed region. This lays the foundation for subsequent fine-tuning, enabling the system to further determine and separate each independent connected region.

[0078] Based on this, connected weed regions that do not meet the preset weed region conditions are eliminated according to their area, aspect ratio, and structural continuity. For example, an area threshold can be set to eliminate connected regions that are too small (possibly residual noise) or too large (possibly non-weed objects or abnormal adhesion of multiple weeds). Simultaneously, by calculating the aspect ratio of the minimum bounding rectangle of the connected regions, regions with abnormal shapes can be identified and eliminated.

[0079] Furthermore, by extracting the skeleton of connected regions and evaluating their continuity, it is possible to determine whether the region is a complete object conforming to the morphology of a weed, thereby excluding regions with discontinuous or fragmented skeletons. These preconditions are determined based on statistical analysis and accumulated experience of the morphological characteristics of common weeds in cornfields, aiming to improve the accuracy of weed identification and reduce false identification.

[0080] Furthermore, when any connected region of weeds contains multiple intertwined weed targets, watershed segmentation or skeleton decomposition is performed on this connected region to separate it into multiple individual weed regions. Watershed segmentation can be applied when connected component analysis reveals that the area, aspect ratio, or skeleton features of a region indicate that it may contain multiple intertwined weeds. Watershed segmentation treats the gradient magnitude of the image as terrain height, and by finding local minima (catchment basins) and simulating the water flow convergence process, it separates intertwined objects at the "watershed."

[0081] Another alternative method is skeletal decomposition. This method extracts the skeletal structure of the clump of weeds and analyzes the bifurcation points and connection paths on the skeletal structure. Cutting is then performed at these key connection points, thus decomposing a complex skeletal structure into multiple independent skeletals, each corresponding to a single weed. Using either of these methods, multiple weed targets that were originally clumped together can be effectively and precisely separated into independent weed regions, ensuring that each region corresponds to an individual weed plant as closely as possible.

[0082] This embodiment further proposes to extract candidate points at the base of the stem, candidate points at the growth point, or candidate points at the lower end of the main framework based on the contour and skeletal features of each individual weed area, forming a set of candidate attack points, including: Extract the outline of each individual weed region, and determine the geometric center point, centroid point, and minimum circumscribed region center point of the corresponding individual weed region based on the outline. The skeleton of each individual weed is extracted to obtain the weed skeleton, and the skeleton end point, skeleton bifurcation point and skeleton node near the ground surface are extracted from the weed skeleton. Based on the main axis direction of the weed skeleton, the top and bottom boundaries of the area of ​​a single weed, candidate points at the base of the stem, candidate points at the growth point, and candidate points at the lower end of the main skeleton are determined. Candidate priorities are determined according to the order of candidate points at the base of the stem, candidate points at the growth point, candidate points at the lower end of the main framework, and the geometric center point, forming a set of candidate attack points.

[0083] The process involves extracting the outline of each individual weed region and determining the geometric center point, centroid point, and minimum circumscribed region center point of the corresponding individual weed region based on the outline, with the aim of obtaining the overall morphology and location information of the individual weed region.

[0084] Contour extraction can be achieved by combining edge detection algorithms (such as the Canny operator and the Sobel operator) with contour tracking techniques to obtain the boundary pixel set of the weed region. Based on the extracted contour, the geometric center point can be obtained by calculating the average of the coordinates of all pixels on the contour; the centroid can be obtained by calculating the zeroth and first moments of the image, which better reflects the center of the region's mass distribution; the center point of the smallest bounding region can be determined by calculating the center point of the smallest rectangle, smallest circle, or smallest ellipse enclosing the contour. These center points provide a basic geometric reference for the subsequent selection of the strike point.

[0085] The skeleton of each individual weed is extracted to obtain the weed skeleton, and the skeleton end point, skeleton bifurcation point and skeleton node near the ground surface are extracted from the weed skeleton to reveal the internal structure and growth path of the weed.

[0086] Skeleton extraction typically employs morphological thinning algorithms or distance transform methods to simplify the weed region into skeleton lines of single-pixel width. From the resulting weed skeleton, the end point is a point on the skeleton with only one adjacent pixel, usually corresponding to the leaf tip or root end; the branching point is a point on the skeleton with three or more adjacent pixels, representing the branching points of the weed structure; and the skeleton nodes near the ground surface refer to nodes within a certain distance from the bottom of the weed region, these nodes are usually associated with the stem base or root collar of the weed.

[0087] Based on the main axis direction of the weed skeleton, the top and bottom boundaries of the single weed area, the candidate points at the stem base, growth points, and lower end of the main skeleton are determined to accurately lock onto effective targets for laser strikes.

[0088] The principal axis direction of the weed skeleton can be determined by fitting the skeleton point set using principal component analysis (PCA) or least squares method, reflecting the main growth direction of the weed. Stem base candidate points are typically nodes in the weed skeleton that intersect with or are closest to the bottom boundary, located at the lower end of the principal axis direction. Growth point candidate points are usually located near the top end of the weed skeleton, or at the uppermost bifurcation point along the principal axis direction; these points are key sites for photosynthesis and growth in the weed. Lower end candidate points of the principal skeleton are the skeleton nodes extending downwards along the weed's principal axis direction to the very bottom, providing a robust, near-surface striking point.

[0089] Candidate priorities are determined by ranking candidate points at the base of the stem, growth points, lower ends of the main framework, and geometric center points to form the set of candidate attack points, with the aim of establishing a preferred attack strategy.

[0090] Priority settings typically follow the principle of maximizing the weed-killing effect.

[0091] Candidate sites at the base of the stem have the highest priority because they directly connect the root system and the above-ground parts, and their destruction can effectively prevent weed regeneration.

[0092] Candidate growth points are areas where weeds are actively growing; destroying them can inhibit their growth, so they are of secondary priority.

[0093] Candidate points at the lower end of the main skeleton are considered robust alternatives to candidate points at the base of the stem, and are given secondary priority.

[0094] The geometric center point, centroid point, or minimum circumscribed region center point are used as general center points, and are the lowest priority alternatives when a more precise vulnerability point cannot be determined.

[0095] Through this prioritization, the system can select the most effective strike point based on the actual situation.

[0096] Geometric center point It can be obtained from the center of the smallest bounding rectangle: ; centroid coordinates It can be represented as: ; in, The number of pixels in the weedy area. ) is the first in the weedy area The coordinates of each pixel.

[0097] This embodiment further proposes a safety screening method based on the spatial distance between each candidate impact point and the corn plant area, the laser's effective range, and the stability of the candidate impact points, including: Determine the minimum spatial distance between each candidate strike point and the corn plant area, and expand the corn plant area according to the laser's effective range to obtain the safe corn zone. Determine whether each candidate strike point is located within the safe area of ​​the cornfield, and determine the degree of positional change of each candidate strike point based on continuously collected images of the cornfield operation area, so as to determine the stability of the candidate strike points; Candidate strike points located outside the corn safety zone and whose stability meets the preset stability requirements are identified as safe candidate strike points. When no safe candidate strike point exists, the candidate strike point is moved away from the corn plant area, or a backup candidate strike point is selected from the set of candidate strike points; if neither the candidate strike point after the position adjustment nor the backup candidate strike point meets the safety requirements, the targeted strike on the corresponding individual weed is cancelled.

[0098] Specifically, when determining the minimum spatial distance between each candidate strike point and the corn plant area, the pixel distance from each candidate strike point in the image to the nearest corn plant area boundary can be calculated and converted into an actual physical distance based on the image resolution. The laser's effective range refers to the area that the laser's energy can effectively cover during a targeted strike, typically determined by the laser's power, focusing characteristics, and the characteristics of the target. Expanding the corn plant area based on the laser's effective range is to provide a safety margin during safety screening, ensuring that the actual impact of the laser strike does not reach the corn plant. For example, if the laser's effective range is a circular area of ​​radius R, the boundary of the corn plant area will be extended outward by a distance R, forming a "corn safety zone." Any candidate strike point falling within this expanded corn safety zone will be considered at risk of accidentally damaging the corn plant.

[0099] Let the corn-covered area be Candidate strike points are The final strike point is The minimum Euclidean distance from the candidate strike point to the cornfield can be expressed as: ; in, These are the pixels in the corn region.

[0100] Let the safe distance threshold be Then when At that time, it was believed that the candidate strike point was located outside the safe area of ​​cornfields; when At that time, it was believed that the candidate strike point was located within the safe area of ​​cornfields.

[0101] After determining whether each candidate strike point is located within the cornfield safety area, it is necessary to determine the degree of positional change of each candidate strike point based on continuously acquired images of the cornfield operation area to determine the stability of the candidate strike points. The continuously acquired images of the cornfield operation area refer to a sequence of multiple frames continuously acquired at a certain frequency during the movement of the mobile operation platform. By analyzing these continuous images, the positions of the same weed instance and its candidate strike points at different times can be tracked. The degree of positional change of the candidate strike point refers to the stability of the positional offset or trajectory of the same candidate strike point between different frames in the continuous image sequence. For example, its stability can be quantified by calculating the standard deviation or average displacement of the centroid position of the candidate strike point in a continuous P-frame image. If the degree of positional change is small, the candidate strike point is considered stable; otherwise, it is unstable. Its purpose is to evaluate the reliability of the candidate strike points before actual laser strikes, avoiding strike failures or accidental damage caused by platform jitter, image recognition errors, etc., leading to instability of the strike points.

[0102] A candidate impact point is only identified as a safe candidate impact point if it is located outside the corn safety zone and meets a preset stability requirement. The preset stability requirement is a threshold set for the degree of change in the position of the candidate impact point. For example, it can be set to ensure that the pixel displacement of the candidate impact point does not exceed a certain preset value in several consecutive frames of images.

[0103] If, after the above screening, no safe candidate attack point still meets the safety and stability requirements, the system will take further measures. One approach is to adjust the current candidate attack point's position away from the corn plant area. For example, it can be moved in small steps away from the nearest corn plant area until it meets the conditions outside the safe zone. This adjustment needs to ensure that the adjusted point can still effectively combat weeds. Another approach is to select backup candidate attack points from the candidate attack point set. When forming the candidate attack point set, multiple different types of candidate points are usually extracted (such as stem base, growing point, lower end of the main skeleton, etc.) and assigned different priorities. When the highest priority candidate point does not meet the safety requirements, a backup point with the next lower priority can be selected from the set for evaluation. If neither the candidate attack point after position adjustment nor the backup candidate attack point meets the safety requirements, the system will cancel the targeted attack on the corresponding individual weed to avoid accidentally damaging the corn plant, reflecting the highest priority of crop protection.

[0104] This embodiment further proposes converting the target strike point into ground operation coordinates and performing motion compensation on the ground operation coordinates, including: Based on the internal calibration parameters of the image acquisition device and the external calibration parameters between the image acquisition device and the laser, establish the coordinate mapping relationship between the image coordinate system and the ground operation coordinate system; Based on the coordinate mapping relationship, the image coordinates corresponding to the target impact point are converted into ground operation coordinates, and the ground operation coordinates are corrected according to the platform attitude. Based on the image acquisition time, the expected laser action time, the platform speed, and the direction of travel, the displacement of the mobile work platform within the system time delay is determined; The ground operation coordinates after attitude correction are compensated based on the displacement. When the platform speed or attitude changes, the compensated ground operation coordinates are predicted and corrected based on the platform acceleration of the mobile operation platform to obtain the laser execution coordinates.

[0105] Specifically, the internal calibration parameters of an image acquisition device refer to its inherent properties, such as focal length, principal point coordinates, radial distortion coefficient, and tangential distortion coefficient. These parameters can be obtained through offline calibration to eliminate geometric distortions generated during the imaging process and ensure the accuracy of image coordinates.

[0106] The external calibration parameters between the image acquisition device and the laser refer to their relative positions and orientations in three-dimensional space, typically represented by a rotation matrix and a translation vector. These parameters can be obtained through methods such as hand-eye calibration and are used to transform points in the image acquisition device's coordinate system to the laser's coordinate system, and then correlated to the ground operation coordinate system.

[0107] The coordinate mapping relationship is a mathematical model established based on the aforementioned internal and external calibration parameters. It can accurately project the two-dimensional pixel coordinates on the image plane to the actual physical coordinates in the three-dimensional ground space. This typically involves the principles of perspective projection and the calculation of coordinate transformation matrices, enabling the target points in the image to be accurately located on the ground.

[0108] When converting the image coordinates corresponding to the target impact point into ground operation coordinates, the established coordinate mapping relationship is used to convert the target impact point (e.g., pixel coordinates) identified in the image into three-dimensional coordinates in the ground operation coordinate system through inverse projection or direct projection calculation.

[0109] In this embodiment, a planar homography transformation method is preferably used to convert pixel coordinates to ground coordinates.

[0110] Let the number of image pixels be 1. The actual ground coordinates are Then we have: ; in, for Homography matrix.

[0111] When expanded, it can be represented as: ; ; in, These are elements of the homography matrix.

[0112] After mapping, the impact point is determined by image coordinates. Convert to actual work coordinates .

[0113] Platform attitude refers to the real-time spatial attitude of a mobile work platform during operation, including pitch, roll, and yaw angles. This attitude information is typically acquired in real time by an inertial measurement unit (IMU) or other attitude sensors.

[0114] Attitude correction adjusts the transformed ground coordinates according to the real-time attitude of the mobile platform. For example, if the platform tilts, the calculated ground coordinates will be inaccurate without attitude correction. By transforming the ground coordinates using a rotation matrix (based on the platform's attitude), the influence of platform attitude changes on the ground coordinate positioning of the strike point can be eliminated, ensuring that the strike point is always located at the expected ground position.

[0115] To address the dynamic nature of mobile work platforms, it is necessary to determine the movement and displacement of the mobile work platform within the system latency.

[0116] The image acquisition time refers to the precise timestamp at which the image acquisition device obtains the image of the current cornfield operation area.

[0117] The predicted laser action time refers to the time when the laser beam actually acts on the target impact point, as predicted by the system based on factors such as processing time and laser response time.

[0118] System delay is the time difference between the image acquisition time and the expected laser action time. Platform speed and direction of travel refer to the average linear velocity and direction of motion of the mobile work platform within the system delay.

[0119] This information is typically provided by sensors such as GNSS (Global Navigation Satellite System) receivers, encoders, or visual odometry. Movement displacement is the displacement vector of the mobile work platform on the horizontal plane, calculated based on system time delay, platform speed, and direction of travel.

[0120] Forward compensation is performed on the attitude-corrected ground work coordinates based on the displacement. Forward compensation refers to adjusting the attitude-corrected ground work coordinates along the platform's travel direction in advance, taking into account the movement of the mobile work platform within the system delay, when determining the laser execution coordinates.

[0121] Specifically, the ground coordinates of the target impact point are moved forward (or backward, depending on the relative direction of motion) by a distance equal to the displacement of the platform, to ensure that when the laser emits the laser at the expected moment of action, the laser beam can accurately hit the expected location of the weeds, rather than the location of the weeds at the moment of image acquisition.

[0122] Set the platform at time The speed is The unit vector of the direction of motion is The total latency for image processing and laser execution is The platform has latency The displacement compensation amount within can be expressed as: ; If the platform moves on a two-dimensional plane, it can be written as: ; in, , The platform is respectively shaft and The velocity component in the axial direction.

[0123] The actual working coordinates after compensation are: .

[0124] Furthermore, when the platform speed or platform attitude changes, the compensated ground operation coordinates are predicted and corrected by combining the platform acceleration of the mobile operation platform to obtain the laser execution coordinates.

[0125] Changes in platform speed or attitude refer to nonlinear or significant changes in the motion state (speed magnitude or direction) or spatial attitude (pitch, roll, yaw) of a mobile work platform during operation.

[0126] Platform acceleration refers to the acceleration vector of a mobile work platform as its speed or attitude changes. Acceleration information can be acquired in real time by an IMU or other motion sensors.

[0127] Predictive correction builds upon forward-looking compensation by further considering the impact of platform acceleration on future position. When platform speed or attitude changes, forward-looking compensation based solely on constant speed may not be accurate enough. By introducing platform acceleration, kinematic models (e.g., uniformly accelerated linear motion models or more complex motion models) can be used to more accurately predict the platform's precise position at the expected laser impact moment. This allows for a secondary correction of the compensated ground coordinates, yielding the final laser execution coordinates. This correction effectively addresses the dynamic changes of the platform in complex terrain or during variable-speed travel, further improving strike accuracy.

[0128] Example 5 Figure 3 This is a schematic diagram of a laser weeding system for cornfield scenarios provided in one embodiment of this application, with reference to... Figure 3 A laser weeding system for cornfield applications includes: Mobile operating platform 21, screw lifting mechanism 22, image acquisition device 23, two-dimensional servo gimbal 24, and laser 25; The mobile operating platform is equipped with a processor and a memory; The processor and memory are connected via a communication bus: The processor is used to call and execute programs stored in memory. A memory for storing a program, which is at least used to execute a laser weeding method for a cornfield scenario as described in any of the above embodiments.

[0129] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0130] It should be noted that in the description of this application, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this application, unless otherwise stated, "a plurality of" means at least two.

[0131] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the function involved, as will be understood by those skilled in the art to which embodiments of this application pertain.

[0132] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0133] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0134] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0135] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.

[0136] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0137] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A laser weeding method for cornfield scenarios, characterized in that, include: The system acquires images of the cornfield operation area collected by the mobile operation platform and the status information of the mobile platform, including platform speed, direction of travel, platform attitude, and image acquisition time. The cornfield operation area image is segmented at the pixel level to obtain corn plant areas and weed areas, and the weed areas are further separated into multiple individual weed areas. Based on the contour and skeletal features of each individual weed area, candidate points at the base of the stem, candidate points at the growth point, or candidate points at the lower end of the main skeleton are extracted to form a set of candidate attack points. Safety screening is conducted based on the spatial distance between each candidate strike point and the corn plant area, the laser's effective range, and the stability of the candidate strike points. Candidate strike points that pose a risk of accidentally damaging corn plants are eliminated, their positions are adjusted, or they are replaced by backup candidate strike points to determine the target strike point. Based on the calibration results of the image acquisition device and the spatial pose relationship between the image acquisition device and the laser, the target impact point is converted into ground operation coordinates; Based on the mobile platform status information and the system time delay from the image acquisition time to the laser action time, motion compensation is performed on the ground operation coordinates to obtain the laser execution coordinates; The laser is controlled to strike the corresponding individual weeds based on the laser's execution coordinates.

2. The method according to claim 1, characterized in that, Before performing pixel-level segmentation on the image of the cornfield work area, the method further includes: The distortion correction of the cornfield operation area image is performed according to the calibration parameters of the image acquisition device to eliminate the imaging distortion generated by the image acquisition device; Based on the brightness distribution and color deviation of the cornfield operation area image, white balance correction, illumination normalization, and noise reduction are performed to obtain a corrected image; The corrected image is subjected to color space conversion and size unification to obtain the image to be segmented, and the image to be segmented is used as the input for pixel-level segmentation.

3. The method according to claim 1, characterized in that, The method further includes: Obtain re-inspection images after pinpoint strikes; The strike result is determined based on the area of ​​residual weeds in the re-inspection image; If the strike result does not meet the preset requirements, the target strike point, motion compensation parameters, or laser execution parameters are adjusted before a supplementary strike is performed.

4. The method according to claim 3, characterized in that, The strike results are determined based on the areas of residual weeds in the re-examined images, including: The re-inspection image is segmented at the pixel level to obtain the residual weed area corresponding to the targeted strike; Determine the pixel area of ​​the residual weed region and compare the pixel area with a preset residual area threshold; When the pixel area is not greater than the preset residual area threshold, the impact result is determined to meet the preset requirements; When the pixel area is greater than the preset residual area threshold, it is determined that the strike result does not meet the preset requirements, and the target strike point is adjusted according to the position of the residual weed area relative to the target strike point; The laser execution parameters are adjusted based on the difference between the pixel area and the preset residual area threshold, and the laser execution coordinates are re-determined by combining the adjusted target impact point, motion compensation parameters and laser execution parameters. Supplementary strikes will be carried out based on the newly determined laser execution coordinates.

5. The method according to claim 1, characterized in that, The image of the cornfield operation area is segmented at the pixel level to obtain the corn plant area and the weed area, including: The image of the cornfield operation area is input into a lightweight semantic segmentation model. The lightweight semantic segmentation model is used to extract lightweight features from the image of the cornfield operation area to obtain image features at different scales. Feature fusion is performed on image features at different scales, and boundary enhancement and attention enhancement are applied to the fused image features to obtain enhanced image features; Based on the enhanced image features, pixel-level spatial information is recovered, and the region category corresponding to each pixel is determined to obtain the corn plant region, the weed region, and the background interference region. Based on the adjacency relationship between the corn plant area and the weed area, a high-risk boundary area near the corn is determined, and the high-risk boundary area near the corn is used for safe screening of the candidate attack points.

6. The method according to claim 1, characterized in that, The weed areas were separated into multiple individual weed areas, including: Morphological processing is performed on the weed area to remove noisy areas and fill holes in the weed area; Connectivity analysis was performed on the morphologically treated weed regions to obtain multiple connected weed regions. Based on the area, aspect ratio, and structural continuity of each weed connected region, weed connected regions that do not meet the preset weed region conditions are eliminated. When any connected weed region contains multiple weed targets that are interconnected, the connected weed region is divided into watershed segments or skeleton decompositions to separate the connected weed region into multiple single-weed regions.

7. The method according to claim 1, characterized in that, Based on the contour and skeletal features of each individual weed area, candidate points are extracted from the stem base, growth point, or lower end of the main skeletal structure to form a set of candidate attack points, including: Extract the outline of each individual weed region, and determine the geometric center point, centroid point, and minimum circumscribed region center point of the corresponding individual weed region based on the outline. The skeleton of each individual weed is extracted to obtain the weed skeleton, and the skeleton end point, skeleton bifurcation point and skeleton node near the ground surface are extracted from the weed skeleton. Based on the main axis direction of the weed skeleton, the top and bottom boundaries of the single weed area, the candidate points at the stem base, the candidate points at the growth point, and the candidate points at the lower end of the main skeleton are determined. Candidate priority is determined according to the order of candidate points at the base of the stem, candidate points at the growth point, candidate points at the lower end of the main skeleton, and the geometric center point, thus forming the set of candidate strike points.

8. The method according to claim 1, characterized in that, Safety screening is conducted based on the spatial distance between each candidate impact point and the corn plant area, the laser's effective range, and the stability of the candidate impact points, including: Determine the minimum spatial distance between each candidate strike point and the corn plant area, and expand the corn plant area according to the laser's effective range to obtain a safe corn zone; Determine whether each candidate strike point is located within the cornfield safety area, and determine the degree of positional change of each candidate strike point based on continuously collected images of the cornfield operation area, so as to determine the stability of the candidate strike points; Candidate strike points located outside the corn safety zone and whose stability meets the preset stability requirements are identified as safe candidate strike points. When no safe candidate strike point exists, the candidate strike point is moved away from the corn plant area, or a backup candidate strike point is selected from the set of candidate strike points; if neither the candidate strike point after position adjustment nor the backup candidate strike point meets the safety requirements, the targeted strike on the corresponding individual weed is canceled.

9. The method according to claim 1, characterized in that, The step of converting the target strike point into ground operation coordinates and performing motion compensation on the ground operation coordinates includes: Based on the internal calibration parameters of the image acquisition device and the external calibration parameters between the image acquisition device and the laser, a coordinate mapping relationship between the image coordinate system and the ground operation coordinate system is established. Based on the coordinate mapping relationship, the image coordinates corresponding to the target impact point are converted into the ground operation coordinates, and the ground operation coordinates are corrected according to the platform attitude. Based on the image acquisition time, the expected laser action time, the platform speed, and the direction of travel, the displacement of the mobile work platform within the system time delay is determined; Based on the displacement, the ground work coordinates after attitude correction are compensated for forward look; when the platform speed or the platform attitude changes, the compensated ground work coordinates are predicted and corrected in combination with the platform acceleration of the mobile work platform to obtain the laser execution coordinates.

10. A laser weeding system for cornfield scenarios, characterized in that, include: Mobile work platform, screw lifting mechanism, image acquisition device, 2D servo gimbal and laser; The mobile operating platform is equipped with a processor and a memory; The processor and memory are connected via a communication bus: The processor is used to call and execute the program stored in the memory; The memory is used to store a program, which is at least used to execute the laser weeding method for cornfield scenarios as described in any one of claims 1-9.

Images