Intelligent breeding phenotype collection and mobile measurement platform

Abstract

The application relates to the technical field of agricultural information detection, and particularly discloses an intelligent breeding phenotype collection and mobile measurement platform, which comprises an image collection unit, a positioning and attitude sensing unit and a central processing unit carried on a mobile device. The central processing unit is used for performing the following operations: collecting a plant image sequence through a multispectral camera, and performing waveband selection and pretreatment based on plant leaf reflectance spectrum characteristics; performing space-time alignment on the image, and identifying and repairing a blurred area caused by wind; performing image enhancement and noise removal on a low-contrast plant contour area; and integrating position and visual data by using a sensor fusion algorithm to generate integrated pose data. The application effectively solves the problems of data distortion and incomplete coverage caused by wind blur, noise interference, mobile device shaking and path fixation in a dynamic field environment, and improves the precision and efficiency of plant phenotype collection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of agricultural information detection, and particularly to an intelligent breeding phenotype collection and mobile measurement platform. BACKGROUND

[0002] In the field of agricultural science research, accurate monitoring of plant growth and aging processes is of great significance for improving crop yields and optimizing agricultural management. This research not only concerns food security, but also directly affects the economic benefits and sustainable development of agricultural production. Although current technical means have made some progress in plant monitoring, there are still significant shortcomings, especially in long-term tracking in dynamic environments and accurate capture of complex feature changes. Existing methods often fail to meet actual needs, especially in complex field scenarios, where data continuity and accuracy are often challenged.

[0003] Under the conditions of average wind speed 2.5 m / s and illumination variation range 500-1000 lx, the image signal-to-noise ratio of traditional RGB imaging decreases from 28 dB to 16 dB, with a decrease of 42.8%, which cannot effectively separate plant features from environmental noise; the accumulated pose error caused by mobile device motion can reach 0.5-1 meters in a 10-meter moving trajectory, which cannot establish accurate plant growth time series data; fixed path planning cannot adapt to complex field terrain, with coverage rate less than 80%, resulting in systematic omission of data collection.

[0004] Current methods generally lack the ability to continuously identify plant individual identities in natural field environments. Many technologies can obtain plant appearance information at a single time point, but cannot maintain continuous attention to the same plant over its growth cycle, especially when plant morphology changes significantly over time, the accuracy of identification will decrease significantly. This limitation further leads to the inability to accurately correlate data at different time points when analyzing plant aging processes, thereby affecting the judgment of aging feature change trends.

[0005] A deeper technical challenge lies in how to accurately locate individual plants and continuously capture their characteristic changes in dynamic field environments. Field environments are complex and variable; factors such as light and wind can interfere with the accurate identification of plant morphology, especially subtle changes in leaf color or shape, which are often overlooked. This environmental interference makes continuous tracking of plant spatial location extremely difficult. Once the target plant's location deviates at different points in time, the correspondence of all subsequent data will become chaotic. For example, in a field measurement, if the mobile device cannot accurately identify the target plant, it may misclassify data from adjacent plants, leading to completely distorted analysis results of the aging process. How to achieve continuous identification of individual plants in complex field environments and accurately capture subtle changes in their appearance during aging has become a key problem to be solved in existing technologies. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent breeding phenotypic acquisition and mobile measurement platform, which can effectively solve the technical problems of wind ambiguity, noise interference, mobile device vibration and path optimization in the dynamic field environment, and achieve the purpose of high-precision and continuous plant phenotypic acquisition and distribution monitoring.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0008] The intelligent breeding phenotypic acquisition and mobile measurement platform includes an image acquisition unit, a positioning and attitude perception unit, and a central processing unit mounted on a mobile device.

[0009] The central processing unit is used to execute the core processing flow of the system, forming a complete technical closed loop from raw data acquisition to high-precision plant distribution information output.

[0010] Specifically, the execution flow of the central processing unit includes:

[0011] Plant image sequences were acquired in a dynamic field environment using a multispectral camera. High-quality target image sets were obtained through band selection and image preprocessing based on the spectral reflectance characteristics of plant leaves. The band selection was optimized for the vegetation's sensitive spectral range, improving data effectiveness from the source.

[0012] The target image set is subjected to spatiotemporal alignment processing to identify image blurring areas caused by wind, and these areas are corrected using an image inpainting algorithm to obtain aligned and corrected images. Furthermore, local image enhancement is applied to low-contrast areas of plant outlines, and environmental interference is effectively suppressed by combining noise extraction algorithms to output contrast-enhanced images.

[0013] The sensor fusion algorithm is used to integrate the positioning data and visual data, real-time monitoring of the position and attitude change of the mobile device is generated, and high-precision integrated pose data is generated. Based on the pose data, the image distortion caused by device jitter is corrected through the distortion repair algorithm, and the mobile path deviation is compensated by using the trajectory correction algorithm, and finally the corrected plant position distribution map is formed.

[0014] Further, based on the corrected plant position distribution map, combined with the pre-stored field terrain characteristics, a path planning algorithm is used to generate a high-coverage mobile measurement route; through the region recognition method, the unmeasured region is analyzed, and the mobile path is dynamically adjusted to form an optimized measurement path. At the same time, using the repeated region detection algorithm to identify the covered area, the collection frequency and device collection angle are adaptively adjusted to realize high-precision data collection of specific plant features, and finally the spatially consistent plant distribution information is output.

[0015] The application forms a key technology chain through multispectral imaging and band selection, image alignment and enhancement, multi-sensor fusion pose estimation, trajectory correction and dynamic path planning, and constructs a closed-loop processing system of "perception-correction-decision-execution". From the physical level, environmental interference is suppressed, and the original quality of data collection is guaranteed; through the fusion of positioning and visual information, the plant spatial positioning with centimeter-level precision is realized in the complex field environment; and based on the real-time acquisition data, the measurement path is dynamically optimized, which significantly improves the completeness and collection efficiency of data coverage.

[0016] Compared with the prior art, the application has the following beneficial effects:

[0017] The application fundamentally solves the distortion and incomplete coverage problem of phenotype data collection in a dynamic field environment by constructing a closed-loop technology chain of perception-correction-decision-execution. Through the cooperation of the multispectral camera and the band selection algorithm, the reflection characteristics of the plant leaves are optimized for collection from the data source, and combined with the image alignment, interference detection and adaptive enhancement strategy, the environmental noise such as wind and light change is effectively removed, and the image PSNR is verified to be increased from 15.2dB to more than 22.5dB through the embodiment, ensuring the initial fidelity of the plant contour and feature information. Through the deep fusion of the positioning and attitude perception unit with the visual data, a high-precision integrated pose data model is constructed, so that the distortion repair and trajectory correction can accurately compensate for the jitter and movement deviation of the mobile device, and the plant position mapping error is controlled within centimeters, realizing the qualitative change from "single point image" to "accurate spatial distribution". Based on the corrected distribution map, the path planning and adaptive collection are carried out, the unmeasured area and the repeated area are identified, the measurement path and the collection parameters are dynamically adjusted, the field monitoring coverage is improved, and redundant data collection is avoided, and finally the high-precision distribution information of the specific features of the plants is output, which provides a quantifiable, traceable and spatially consistent decision-making data basis for intelligent breeding. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a system architecture block diagram of the present invention.

[0020] Figure 2 This is a flowchart illustrating the operation of the present invention. Detailed Implementation

[0021] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the embodiments of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0022] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0023] Example 1:

[0024] The preset thresholds involved in this invention are determined based on the following principles: First, historical field experimental data are collected to establish a statistical distribution model; then, the optimal threshold point is determined through ROC curve analysis to ensure a balance between precision and recall. Specific thresholds can be fine-tuned according to the actual crop type and field environment during system deployment.

[0025] See Figure 1 and Figure 2 This embodiment discloses an intelligent breeding phenotypic acquisition and mobile measurement platform, including an image acquisition unit, a positioning and attitude perception unit, and a central processing unit mounted on a mobile device;

[0026] The mobile device is a drone or ground mobile robot equipped with a high-performance embedded processing unit. Its central processing unit includes at least one ARM Cortex-A series multi-core CPU and a GPU supporting parallel computing (such as the NVIDIA Jetson Nano series), with at least 4GB of memory to ensure that the image processing algorithm can run stably at a rate of at least 5 frames per second. The multispectral camera is preferably equipped with 450nm, 550nm, 650nm, 750nm, and 850nm band channels, which cover spectral regions crucial for vegetation analysis, such as chlorophyll absorption, green peak, red edge, and near-infrared plateau.

[0027] wherein the central processing unit is configured to perform the following operations:

[0028] Step 1: Collecting a plant image sequence in a field dynamic environment by a multispectral camera, obtaining a target image set after preprocessing and screening by using a waveband selection algorithm based on the reflectance spectrum characteristics of plant leaves;

[0029] Step 2: For the target image set, using an image alignment algorithm to perform time and space alignment processing on consecutive frame images, identifying blur areas caused by wind through an interference detection algorithm, and determining the aligned and corrected images;

[0030] Step 3: If the local contrast of the plant contour in the aligned and corrected images is lower than a preset threshold, performing local processing on the low-contrast area by an image enhancement algorithm, and removing environmental noise effects by combining a noise extraction algorithm, to obtain a contrast-enhanced image;

[0031] Step 4: According to the contrast-enhanced image, using a sensor fusion algorithm to integrate position data and visual data, monitoring the position of the mobile device in real time, recording the tilt changes of the mobile device in space through attitude monitoring, and determining the integrated pose data;

[0032] Step 5: For the integrated pose data, using a distortion correction algorithm to correct the image distortion caused by the shaking of the mobile device, adjusting the deviation of the mobile device during movement through a trajectory correction algorithm, and obtaining a corrected plant position distribution map;

[0033] Step 6: According to the corrected plant position distribution map, using a path planning algorithm to generate a mobile route with high coverage based on the field terrain characteristics, analyzing the unmeasured field area through a region recognition method, and determining the optimized measurement path and final plant distribution data;

[0034] Step 7: If there are repeated areas in the optimized measurement path, identifying the covered areas through a repeated area detection algorithm and adjusting the acquisition frequency, combining a feedback mechanism to dynamically adjust the acquisition angle of the mobile device, and obtaining high-precision distribution information for specific plant characteristics.

[0035] In this embodiment, the specific steps of Step 1 are as follows:

[0036] The plant image sequence is acquired in the field dynamic environment by a multispectral camera, the original image data is extracted according to the plant leaf characteristics, and the target image set after preprocessing and screening is obtained; the original image sequence is denoised by using image preprocessing technology, and the preliminary filtering is carried out according to the environmental interference factors, and the denoised preprocessing result image is obtained; according to the influence of the dynamic environment in the preprocessing result image, the image segmentation technology is applied to separate the plant leaf area and the background area, the clear boundary of the leaf area is determined, and the segmented segmentation result image is obtained; according to the leaf reflection characteristic data in the segmentation result image, if abnormal fluctuations exist in the reflection data, the mean value smoothing technology is used to correct the abnormal area, and the third image set after correction is obtained; the key band data corresponding to the band selection algorithm is extracted through the plant leaf characteristics in the third image set, whether the key band data meets the preset threshold range is judged, and the fourth image set after screening is obtained; according to the image acquisition process data in the fourth image set, the influence degree of the field dynamic environment on the image quality is analyzed, the final image set output by this sequence is taken as the input of step 2; according to the leaf reflection analysis result in the final image set, the original data collected by the multispectral camera is combined, the reflection characteristic distribution of the plant leaf under different bands is obtained, and the health state information of the plant leaf is judged.

[0037] In specific implementation, when acquiring plant image sequence in the field dynamic environment by a multispectral camera, a multispectral camera equipped with 5 band channels (450nm, 550nm, 650nm, 750nm, 850nm) can be used, which is installed on a UAV platform, scans the target farmland in the air at a height of 10 meters at a frequency of 2 frames per second, covers an area of about 500 square meters, and acquires image data containing visible light and near-infrared bands.

[0038] After acquisition, by analyzing the reflectivity of the plant leaf at 450nm, 550nm, 650nm, 750nm, and 850nm bands, the NDVI value is calculated based on the PROSAIL model or the USGS vegetation spectrum library, the threshold value 0.3 (suitable for common crops such as wheat and corn) is set, and the effective vegetation area is automatically screened. The NDVI threshold value 0.3 is based on the public vegetation spectrum library (such as PROSAIL) and 1000 sets of field measured data (covering 3 crops such as wheat and corn), the NDVI of healthy vegetation is >0.3, and the NDVI of non-vegetation is ≤0.2, and the screening accuracy is verified by experiment to be 92.3%.

[0039] After obtaining the original image data, the original image sequence is generated, but due to the complex field environment, the image contains light changes, wind moving shadows and other interference, the peak signal-to-noise ratio (PSNR) standard is used to calculate the image quality, and the original image sequence is disturbed by wind (2.5 m / s) and light changes, and the average PSNR is 15.2 dB (the calculation reference is the plant image collected in a dark room without interference as a reference frame). The adaptive median filtering algorithm is used for preprocessing, and the filtering window size is dynamically adjusted according to the image resolution (the default is 5x5 pixels), and the condition for triggering enhanced filtering is that the standard deviation of the pixel gray scale in the window is greater than 10 (the threshold is determined based on the statistical analysis of typical noise images in the field). The PSNR of the processed image is improved to 22.5 dB, and the leaf edge detail retention rate is 89%. The experimental data is derived from 3 groups of parallel field tests (500 images per group), which ensures the accuracy of subsequent analysis.

[0040] Through the above process, a complete chain is formed from data acquisition to preprocessing. If further analysis of the health status of the leaf is required, the processed image set can be input into the CNN model (structure: input layer → 3 convolution layers → 2 pooling layers → 2 fully connected layers → output layer). The model training data set contains 10,000 labeled images (covering 10 common crop diseases and pests, 1,000 images for each disease, image resolution 1920x1080, and labeling accuracy up to pixel level). The Adam optimizer (learning rate 0.001, iteration number 500) is used for training, and the test set (2000 independent images) recognition accuracy is 85.3%, and the recall rate is 82.1%, thereby realizing the closed-loop logic from collection to application.

[0041] In the present embodiment, the specific steps of step 2 are as follows:

[0042] For the initial image, the image alignment algorithm is used to process the continuous frame image. On the basis of time alignment and space alignment, the inter-frame consistency correction is completed to obtain the aligned image group. Through the aligned image group, the disturbance detection algorithm is applied to analyze the wind influence in the dynamic environment, identify the specific position of the blurred area, and determine the disturbed image range. According to the disturbed image range, the region segmentation technology is used to separate the blurred area and the clear area, and the segmented image subset is obtained. For the segmented image subset, the pixel interpolation technology is used for repair processing of the blurred area to obtain the repaired image combination. According to the repaired image combination, the contrast adjustment method is applied to optimize the image details and highlight the plant target area to determine the optimized image data set. Through the optimized image data set, if local area blur is still detected, a preset threshold is used for screening to remove the image parts that do not meet the standard to obtain the final available image set. For the final available image set, data normalization processing is used to unify the image format to obtain the standardized image set suitable for subsequent analysis.

[0043] In specific implementation, after obtaining the original image sequence in the field dynamic environment, the image alignment algorithm is used for time and space alignment processing of the continuous frame image. The specific method is as follows:

[0044] The feature point-based registration algorithm is used, for example, the ORB algorithm is used, the number of feature points is dynamically adjusted (the default maximum is 500, if the number of detected stable feature points is less than 100, the detection threshold is automatically reduced to increase the number of feature points), the Hamming distance is used for feature matching, the error tolerance is set to 1.5 pixels (this value is set based on the statistical registration error of field images in the pre-experiment), the key point detection threshold is set to 0.02 to ensure that not less than 200 feature points are extracted, and then the random sample consensus (RANSAC) algorithm is used to calculate the inter-frame transformation matrix, the error tolerance is set to 1.5 pixels, the continuous frame images are aligned to the same coordinate system, the time interval error of the processed image is controlled within 0.1 seconds, and the spatial offset error is less than 2 pixels, which lays a foundation for subsequent analysis.

[0045] For the aligned image, the disturbance detection is performed to identify the blurred area caused by wind. The gradient analysis-based method is used to calculate the gradient amplitude of the local area of the image. The gradient threshold determination method based on statistics is used: 100 frames of field plant images under different wind grades (0-5 m / s) are collected, the average gradient value of the clear area (no wind blur) is calculated as 28.5, the average gradient value of the blurred area (wind speed ≥2 m / s) is calculated as 8.2, the gradient statistics of the clear and blurred areas are combined, the gradient threshold is initially set to 10, and the local variance feature of the image can be adaptively fine-tuned to reduce the interference value of light change. Under this threshold, the accuracy rate of blurred area identification is 90.1% (experimental samples: 200 frames of random wind disturbance images).

[0046] The blurred areas are preliminarily corrected, preferably using a deep learning restoration model based on U-Net (pre-trained on ImageNet and a custom field blurred image dataset), or a backup bilinear interpolation method (radius 4 pixels), combined with the pixel values of the adjacent clear areas, to fill in the missing areas and ensure that the restored areas naturally connect with the surrounding textures, forming a preliminary corrected image set. The gradient threshold is dynamically adjusted based on real-time wind speed data: when the wind speed is <2 m / s, the threshold is 10, and when the wind speed is ≥2 m / s, the threshold is 8. To ensure logical integrity, the corrected image set can also be correlated with the field wind speed data (real-time collected by sensors, assuming an average wind speed of 2.5 meters per second) for dynamic adjustment of the blur detection threshold. If the wind speed exceeds 3 meters per second, the gradient threshold is reduced to 8 to adapt to more complex environmental disturbances, forming a complete processing chain from alignment to correction to environmental adaptation.

[0047] Further, in actual use, the image is scanned and processed using interference detection technology based on wind disturbance data in a dynamic environment to obtain the distribution position of the blurred area and determine the preliminary affected range.

[0048] The specific steps are as follows:

[0049] According to the preliminary affected range, the plant contour image is divided into blocks using a region division tool to obtain the divided image units. For the divided image units, if the proportion of the blurred area in some units exceeds the preset threshold, they are marked as high repair level and determined as high-priority repair units. Using the data of the high-priority repair units, the blurred areas are locally filled using a pixel reconstruction method to obtain the reconstructed image units. According to the reconstructed image units, a consistency checking tool is applied to smooth the boundaries between adjacent units to obtain the smoothed image combination. For the smoothed image combination, if there are still local areas that do not meet the preset clarity standard, the unqualified areas are removed through a secondary screening tool to obtain the final available image units. Using the final available image units, a data integration method is used to recombine all the units to determine the complete plant target image.

[0050] In actual implementation, in a dynamic field environment, the position of the blurred area is analyzed based on wind disturbance data, wind speed data is collected using an environmental sensor, and it is assumed that the real-time wind speed is 3.2 meters per second. Combined with image processing technology, an edge detection-based algorithm such as the Canny edge detection method is used, with a low threshold of 20 and a high threshold of 60, to extract the edge information of the plant target in the image. Then, the edge continuity interruption area is analyzed to determine the potential blurred area, and the proportion of the area with an interruption length exceeding 5 pixels is about 12% of the total image area, which is used to locate the blurred position affected by the wind.

[0051] The affected range of the plant contour image is evaluated, a region segmentation algorithm is used to divide the image into multiple 10x10 pixel grid cells, the proportion of blurred pixels in each cell is calculated, and if the proportion exceeds 30%, it is marked as a high-impact area. At the same time, the morphological feature database of the plant target is analyzed to determine whether the blurred area covers the key parts of the plant, such as the center of the leaf. If the coverage ratio exceeds 20%, the priority is raised. The statistical results show that the high-impact area accounts for about 18% of the plant contour image.

[0052] The repair priority is determined based on the influence of the blurred area on the plant target recognition, and a priority scoring model is constructed. The scoring formula is:

[0053] ;

[0054] wherein, is the repair priority score,

[0055] is the proportion of blurred areas (0~1),

[0056] is the key part coverage ratio (0~1),

[0057] is the wind direction weight (0.05 when the wind direction deviation angle is greater than 45°, otherwise 0),

[0058] Weight coefficient: (for quantifying the wind direction impact as a score addition).

[0059] Assuming that the blurred proportion of a certain area is 35% and the key part coverage ratio is 25%, the priority score is 31. If the score exceeds 30, it is listed as a priority repair object. At the same time, the field wind direction data is associated, and when the wind direction deviation angle is greater than 45 degrees, an additional 5% priority weight is added to ensure that the repair order adapts to environmental changes, forming a complete analysis chain from blurred positioning to impact evaluation to priority sorting.

[0060] In this embodiment, the specific steps of step 3 are as follows:

[0061] Through preliminary analysis on the image set, the pixel distribution of the plant contour region is detected, whether the local contrast is lower than the preset threshold is judged, and a preliminary contrast evaluation result is obtained; if the preliminary contrast evaluation result shows that the local contrast is lower than the preset threshold, an image enhancement algorithm is used for local adjustment on the low-contrast region to generate an enhanced plant contour image; for the enhanced plant contour image, a noise extraction algorithm is used to identify and separate the environmental noise to obtain a denoised intermediate image data; according to the denoised intermediate image data, whether there is residual noise interference in the plant image is detected, if there is residual noise, further removal is performed through iterative filtering processing to obtain a noise-free plant image; after obtaining the noise-free plant image, edge details of the plant contour are sharpened to determine a final clear image; through verification on the pixel distribution of the final clear image, the contour integrity of the plant image is judged, and a plant image group meeting the expectation is obtained.

[0062] In specific implementation, when processing the preliminary corrected image set, if it is detected that the local contrast of the plant contour is lower than the preset threshold, for example, the set threshold is 0.3 (the contrast value range is 0 to 1), the image enhancement algorithm is automatically triggered.

[0063] For the low-contrast region, a local histogram equalization algorithm (CLAHE) is used, the window size is adaptively set according to the typical size of the plant leaf (the default is 8x8 pixels), the clipping value of the contrast enhancement is dynamically adjusted based on the local contrast distribution of the image (the default is 2.0), the gray scale difference between the plant contour and the background in the enhanced region is enhanced by calculating the gray scale distribution in each window, and the analysis result shows that the contrast can be improved by about 30%.

[0064] In combination with the noise extraction algorithm, the image is decomposed by using a wavelet transform method, the decomposition level is set to 3 layers, the noise features in the high-frequency components are extracted, and the environmental noise is filtered out by using a wavelet soft threshold method, the threshold is set based on the estimation of the noise level of the image background region, and the analysis shows that this process can reduce about 15% of the random noise interference in the image, and ensures that the plant edge details are not blurred.

[0065] The enhanced region and the denoising result are integrated, an enhanced image is generated by using an image fusion algorithm (based on weighted average, the weights are 0.7 and 0.3), the signal-to-noise ratio (PSNR, taking the image of the same scene collected in a windless static environment as a reference) of the fused image is improved to more than 25dB, meeting the subsequent analysis requirements. The above process forms a complete logical chain from contrast detection to enhancement and then to denoising, which is closely related to the business target of plant image processing, and provides reliable data support for subsequent feature extraction.

[0066] In the embodiment, the specific steps of step 4 are as follows:

[0067] By combining the plant image with the position information, a fusion method is used to track the mobile device position in real time, and preliminary positioning data of the mobile device in the current environment is obtained; according to the preliminary positioning data, the mobile device position is corrected in combination with visual information, and a more accurate spatial position distribution is obtained; for the spatial position distribution, posture monitoring data of the mobile device in the moving process is obtained, the specific situation of the inclination change is recorded, and dynamic adjustment information of the mobile device in the space is determined; the key points of the inclination change are extracted from the dynamic adjustment information, an analysis model established in advance is used to evaluate the spatial consistency, and the stability performance of the mobile device in the moving process is judged; if the stability performance is lower than a preset threshold, the position information and the visual information are processed again through a data fusion algorithm, and an optimized mobile device position state is obtained; through the optimized mobile device position state, in combination with the posture monitoring data in the moving process, the spatial consistency is continuously tracked, and the final mobile device dynamic distribution result is determined; after obtaining the final mobile device dynamic distribution result, the coverage range of the plant image is analyzed in association, and spatial corresponding data of the mobile device and the target region is obtained.

[0068] In a specific implementation, when processing the enhanced image, the position data and the visual data are integrated through a positioning fusion method, specifically, a fusion algorithm based on extended Kalman filter (EKF) is used, the position data (GPS / Beidou dual-mode, update frequency 10Hz, static positioning accuracy ±0.1 meters) and the visual data (ORB feature point matching, sampling rate 10Hz, single-frame positioning error ±0.3 meters) are weighted and fused, and the weight is dynamically adjusted through adaptive extended Kalman filter (AEKF) based on innovation covariance: the initialization process noise covariance Q=diag([0.1,0.1,0.1]), the observation noise covariance R=diag([0.3,0.3,0.3]), the actual covariance C_k is calculated every 10 seconds according to the innovation sequence d_k=z_k-Hx_{k|k-1}, if the difference between C_k and the theoretical covariance S_k=HP_{k|k-1}H^T+R exceeds the threshold, then Q and R are updated according to the Sage-Husa algorithm, and then the fusion weight is adjusted to ensure that the fusion error is minimized. The actual measurement verification (the flight speed of the unmanned aerial vehicle is 0.5m / s, and the slope of the field terrain is ≤15°), the real-time monitoring error of the mobile device position is controlled within ±0.5 meters, and the trajectory deviation is ≤0.3 meters / 100 meters.

[0069] The inertial measurement unit (IMU) is combined with a posture monitoring algorithm to record the inclination changes of the mobile device in space, the sampling interval is set to 0.1 seconds, the quaternion method is used to calculate the Euler angle changes of the mobile device, and the IMU calibration data is fused to compensate for the sensor zero offset. The inclination angle deviation of the mobile device in the three-axis direction is analyzed to be controlled within ±2 degrees, ensuring the stability of the posture data.

[0070] Through a spatial consistency analysis algorithm, combined with the position and posture data during the movement of the mobile device, a spatial calibration method based on triangulation is used to calculate the offset of the mobile device on the movement trajectory. The calibration period is set to every 2 seconds, and the trajectory deviation value is analyzed to be controlled within 0.3 meters to ensure the consistency of the data in the spatial dimension.

[0071] To further improve the business relevance, the integrated pose data is associated with the distribution characteristics of the plant image group. The grid division method is used to divide the monitoring area into 5x5 meter cells, and the plant distribution density in each cell is calculated to provide spatial reference for subsequent precision agriculture management. The above process is automatically processed by the algorithm, from data fusion to posture monitoring to consistency analysis, forming a rigorous logic chain to ensure the spatial data reliability of the mobile device in the plant monitoring scene.

[0072] In the embodiment, step 5 specifically comprises the following steps:

[0073] For the collected integrated pose data, image preprocessing techniques are used to denoise and standardize the original image to obtain the preliminary processed image data. Through the distortion correction algorithm, the image distortion caused by the shaking of the mobile device is corrected for the preliminary processed image data to generate a distortion corrected image. For the corrected preprocessed image, the trajectory correction algorithm is used to analyze the deviation trajectory of the mobile device during movement to determine the deviation adjustment parameter. According to the determined deviation adjustment parameter, the position offset correction is performed on the preprocessed image to generate a position corrected image. The plant position information is extracted from the position corrected image, the image segmentation technique is used to separate the target area, and the preliminary mapping data of plant distribution is obtained. For the preliminary mapping data, the spatial consistency verification method is used to adjust the abnormal points in the position mapping to obtain the final plant position distribution map. According to the final plant position distribution map, a data visualization tool is used to generate an intuitive distribution view to determine the complete presentation of the plant distribution.

[0074] In specific implementation, for the integrated pose data, the distortion correction algorithm is used to correct the image distortion caused by the shaking of the mobile device, and the radial distortion model in the image processing algorithm is used to repair the collected plant images.

[0075] The distortion parameters of the imaging system are obtained by camera calibration method. Specifically, for the multispectral camera, Zhang Zhengyou calibration method is adopted, and a plurality of (not less than 10) images of different poses are collected by using a checkerboard calibration board for calibration. The calibration values of the radial distortion coefficients k1 and k2 are calculated by least square optimization. The typical distortion parameter values are k1=0.15±0.05 and k2=0.03±0.02. The calibration process is completed offline before system deployment, or online calibration is performed in the field by using the built-in checkerboard pattern of the mobile device (triggered every 30 minutes).

[0076] ;

[0077] x, y are the original pixel coordinates, and r is the radial distance of the pixel to the image center point.

[0078] Then, the formula is:

[0079] ;

[0080] ;

[0081] wherein, , is the radial distortion coefficient (typical value: , and is the corrected pixel coordinate.

[0082] The corrected coordinates are obtained. Error analysis shows that the pixel offset after correction is reduced to within 0.5 pixels, ensuring that the image accuracy is improved by about 30%.

[0083] For the deviation of the mobile device during movement, a trajectory correction algorithm is adopted for adjustment. The Kalman filter algorithm is used to smooth the movement trajectory of the mobile device. Based on the motion model of the mobile device, the movement speed is provided in real time by the positioning unit. The initial value of the trajectory deviation is set according to the system accuracy index (for example, 0.05 meters). Through the filter prediction and update steps, the deviation is reduced to 0.01 meters. Analysis shows that the trajectory smoothness is improved by about 40%, thereby ensuring the accuracy of the plant position distribution.

[0084] Based on the corrected image and trajectory data, a plant position distribution map is generated, the plant position is mapped into a two-dimensional coordinate system by using a gridding algorithm, the grid resolution is set to 0.1 meter x 0.1 meter, through statistical analysis of the plant density in each grid, the error range of the distribution map is controlled within 0.02 meters, and it is ensured that the final distribution map can accurately reflect the actual spatial distribution of plants. The above process is automatically processed by algorithm, forming a whole chain logic from image correction to trajectory adjustment to distribution map generation, ensuring data consistency and accuracy.

[0085] In one embodiment disclosed in the present application, step 6 has the following specific steps:

[0086] Through the corrected plant position distribution map, the detailed feature data of the field terrain is obtained by using the pre-established geographic information database, the preliminary analysis result of the terrain undulation and obstacle distribution is determined; according to the preliminary analysis result of the terrain undulation and obstacle, the path planning algorithm is adopted, combined with the high coverage target, the initial moving route scheme is generated, and the path planning data covering the main area of the field is obtained; for the path planning data, the regional recognition method is adopted, the unmeasured area range of the field is analyzed, and it is judged whether there is an insufficient coverage area, if the unmeasured area exceeds the preset threshold value, the path planning data is adjusted, and the updated moving route is obtained; from the updated moving route, the specific position information of the unmeasured area is obtained, the path is locally adjusted by the optimization strategy, the priority path of supplementary measurement is determined, and the optimized measurement path data is obtained; according to the optimized measurement path data, combined with the characteristic analysis result of the field terrain, the K-means clustering algorithm is adopted to classify and verify the plant position, and the optimized plant distribution information is obtained; through the refined plant distribution information, combined with the path data of supplementary measurement, the measurement results of all areas in the field are integrated, and the final plant distribution data is determined; for the final plant distribution data, the data storage technology is adopted, and it is recorded in the preset database to obtain the complete field plant distribution file.

[0087] In specific implementation, based on the corrected plant position distribution map, a high-coverage movement route is generated through a path planning algorithm. An A* algorithm based on field terrain features is adopted, combined with digital elevation model (DEM) data (generated in real time through unmanned aerial vehicle aerial photography or preloaded from public geographic information systems such as Google Earth), and a field terrain slope threshold of 15 degrees is set to avoid areas with excessively large slope. A path with a total length of 500 meters is planned to ensure coverage of 80% of the key field area. The algorithm calculates the cost function (including distance and slope weight, with weight based on terrain complexity adaptive adjustment: flat terrain when terrain slope standard deviation σ≤5°, distance weight 0.7, slope weight 0.3; complex terrain when σ>5°, distance weight 0.4, slope weight 0.6) of each grid point (grid size 1m x 1m), preferentially selects a flat path, generates an optimal route, and after path planning is completed, the system automatically outputs a coverage report showing that the actual coverage area is 400 square meters, meeting the expected target.

[0088] wherein the cost function is as follows:

[0089] ;

[0090] is the path cost of grid point ,

[0091] is the distance cost (normalized to 0~1) from the current position to grid point ,

[0092] is the slope cost (normalized to 0~1, the greater the slope, the higher the cost) of grid point ,

[0093] Weight coefficient:

[0094] Flat terrain (slope standard ): ;

[0095] Complex terrain ( > ): .

[0096] The unmeasured field area is analyzed by a region identification method. The image segmentation technique is combined with the K-means clustering algorithm to divide the un-covered field area into several independent blocks (the number of blocks is determined adaptively according to the total area and distribution of the un-covered area) by the clustering algorithm (such as K-means). The area of each block is calculated. Assuming that the total unmeasured area is 100 square meters, the largest block area is 30 square meters. The system automatically identifies the priority according to the distance between the block center point coordinates (for example, the center point of the largest block is X: 50 meters, Y: 60 meters) and the planned path. The priority of the block closest to the path is the highest. A supplementary measurement plan is generated.

[0097] An optimized measurement path is determined. The system uses a genetic algorithm to optimize the supplementary path based on the priority of the unmeasured area. The path length limit is set to 100 meters, and the iteration number is 500 times. The total length of the supplementary path is calculated to be 85 meters, covering 90% of the unmeasured area. The path coverage rate (measured area / total field area) is updated to 95%. The plant density data of each grid point passed by the path (for example, the density of a certain grid point is 5 plants per square meter) is recorded.

[0098] All measurement data is integrated to generate plant distribution data. The system predicts the density of the unmeasured small area by spatial interpolation method (Kriging interpolation). Assuming that the prediction error is controlled within 5%, the final output field plant distribution map shows that the total plant number is 2000, the average density is 4 plants per square meter, and the data accuracy reaches 98%. This provides a reliable basis for subsequent agricultural management. The above steps are automatically completed through algorithm and data analysis, forming a complete logical chain from path planning to data generation, ensuring the efficiency and accuracy of the technology implementation.

[0099] In this embodiment, the specific steps of step 7 are as follows:

[0100] The measurement path is scanned by a repeated region detection method to identify the distribution of the repeated region and the covered part, and a preliminary region division result is obtained. According to the preliminary region division result, the data collection frequency is adjusted for the repeated region to reduce the collection density of the repeated region, and an optimized collection frequency configuration is obtained. The mobile device is dynamically adjusted through the optimized collection frequency configuration combined with a feedback mechanism to determine the collection angle of the mobile device in different regions. According to the adjusted collection angle, directional data collection is performed for plant features to obtain high-precision data related to plant features. The obtained high-precision data is used to calibrate the distribution information combined with a preset threshold to determine whether the data meets the accuracy requirement. If not, the collection angle is adjusted again and the data is updated. The distribution information of plant features is constructed through the calibrated high-precision data to obtain the final distribution mapping result. The data is classified by using a K-means clustering algorithm for the distribution mapping result to determine the distribution rule of different plant features in space.

[0101] In the optimization measurement path, the covered area is identified by repeating the region detection algorithm, and the high-resolution camera mounted on the acquisition mobile device is used to capture the plant coverage area at a frequency of 30 frames per second, generating image data with a resolution of 1920x1080. Then, a lightweight image similarity calculation algorithm such as perceptual hashing (pHash) or ORB feature matching is used to calculate the structural similarity SSIM between adjacent frame images. The SSIM threshold is based on the average brightness of the image, and the RGB to grayscale formula is used:

[0102] ;

[0103] where R, G, B are the pixel red, green, and blue three-channel grayscale values (0~255), and Y is the converted grayscale brightness value (0~255).

[0104] Dynamic adjustment: when the average brightness <100 lx, the threshold is 0.8, and when the brightness ≥100 lx, the threshold is 0.85. If the SSIM exceeds the threshold, it is determined as a repeated area, and the geographic coordinates (e.g. latitude and longitude data such as 116.3, 39.9) of the area are automatically marked, and the acquisition priority of the area is reduced to 50% of the original, to reduce redundant data acquisition.

[0105] Adjust the acquisition frequency for the repeated area, dynamically calculate the acquisition frequency according to the overlap rate, for example, the initial frequency is 5 times per second, if the SSIM value exceeds the threshold, then through the formula:

[0106] ;

[0107] where,

[0108] is the optimized acquisition frequency (frames / second),

[0109] is the initial acquisition frequency (frames / second),

[0110] is the image structural similarity (0~1, exceeding the threshold is determined as a repeated area),

[0111] is used to limit the minimum acquisition frequency to avoid data loss.

[0112] At the same time, the adjustment data is stored in the database for subsequent analysis, ensuring that the resource allocation efficiency is improved by about 30%.

[0113] To verify the technical effect of the present application, a comparative test is organized. The test is carried out under controllable test conditions (for example, in the experimental field of Henan Agricultural Academy, field area 500 square meters, wheat jointing stage, average wind speed 2-4 m / s, illumination 500-1000 lx). The DJI Jingling 4 multi-spectral unmanned aerial vehicle is used as the control group, and the platform of the present application is used as the experimental group. The phenotype data of wheat at the jointing stage is collected. The results show that the effective data acquisition coverage of the present application is increased from 82.3% to 96.7%, the image availability is increased from 71.5% to 89.2%, and the single plant positioning error is reduced from 0.42 meters to 0.08 meters.

[0114] The feedback mechanism is combined to dynamically adjust the collection angle of the mobile device. Specifically, the attitude data of the mobile device is obtained in real time by the built-in gyroscope and acceleration sensor, the deviation value of the current collection angle and the target angle is calculated (for example, the target angle is 45 degrees, and the current angle is 50 degrees, the deviation is 5 degrees), the PID control algorithm (the control parameters can be adjusted by the system identification method, such as the Ziegler-Nichols method) is used for angle correction, the rotation speed of the mobile device holder is adjusted to 2 degrees per second, and the collection angle error is controlled within ±1 degree.

[0115] Based on the above optimized data collection, the deep learning model (such as ResNet-50) is used to analyze the specific features of plants (such as leaf color, texture) with high precision, extract feature distribution information, generate distribution heat map with resolution of 0.1 meters, and compare and analyze feature change rate (such as leaf color change rate of 20%) based on historical data, so as to realize high-precision distribution information acquisition of specific features of plants. The whole process forms a closed-loop logic from area detection to feature analysis, and the data precision is improved by about 25%.

[0116] The present application reconstructs the phenotype data acquisition problem in dynamic field environment by constructing an intelligent perception system based on physical information closed-loop feedback. The three intercoupled technical problems of environment interference suppression, space consistency maintenance and adaptive path decision are decoupled into quantifiable physical models, and the precise connection of algorithm chain forms a self-optimizing technical closed loop.

[0117] At the data perception layer, the application breaks through the spectral limitation of traditional RGB imaging, directly collects data aiming at the biophysical nature of the reflectance spectrum of plant leaves through the physical band selection characteristics of the multispectral camera. The biological indicator of "plant health status" is converted into the quantifiable "multispectral reflectance difference value" (such as NDVI), ensuring the relevance of data and breeding objectives from the information source. At the same time, through image alignment and interference detection, instead of passively receiving data containing noise, an spatiotemporal reference frame is actively established, modeling random interference such as blurring caused by wind as abnormal displacement of inter-frame pixels, so as to separate "deterministic signals" from "random noise".

[0118] At the data fusion layer, visual data and pose data are processed in a unified spatiotemporal coordinate system. Through sensor fusion algorithms such as Kalman filtering, instead of regarding mobile device jitter as pure interference, it is modeled as an observable and predictable system state variable. Distortion correction is no longer a post-image processing technique, but a feedforward-feedback correction based on a physical motion model, which converts the irregular motion of the mobile device from an error source into a correctable system input, fundamentally ensuring the spatial consistency accuracy of the plant position distribution map.

[0119] At the decision execution layer, the transition from static path planning to dynamic cognitive mapping is realized. The corrected plant distribution map and the field terrain features together constitute a dynamically updated environment cognition map. The path planning algorithm runs on this map, not only pursuing the completeness of geometric coverage, but also forming a perception-planning-action-verification closed loop through region recognition and repeated detection. The feedback mechanism dynamically adjusts the collection angle and frequency according to real-time acquisition data, making the measurement process itself a continuously optimized learning system. The final output of high-precision distribution information is the result of continuous interaction and self-calibration of the system and dynamic environment, rather than a static snapshot of a single measurement.

[0120] Although the preferred embodiments of the application have been described, those skilled in the art, once they know the basic creative concept, can make additional changes and modifications to these embodiments. Therefore, the appended claims are intended to include the preferred embodiments and all changes and modifications falling within the scope of the application.

[0121] The above only describes the preferred embodiments of the application and is not intended to limit the application. It should be noted that any modifications, equivalent replacements and improvements made within the spirit and principles of the application should be included in the protection scope of the application.

Claims

1. An intelligent breeding phenotyping collection and mobile measurement platform, characterized in that, The system comprises an image acquisition unit, a positioning and attitude sensing unit, and a central processing unit; The central processing unit is configured to perform the following operations: Step 1: acquiring a plant image sequence in a dynamic field environment by a multispectral camera, obtaining a target image set after preprocessing and screening by using a waveband selection algorithm based on the reflectance spectrum characteristics of plant leaves, and calculating NDVI values based on the PROSAIL model according to the reflectivity of plant leaves at 450 nm, 550 nm, 650 nm, 750 nm, and 850 nm, and setting a threshold value of 0.3 to automatically screen the effective vegetation area; Step 2: for the target image set, performing time and space alignment processing on the continuous frame images by using an image alignment algorithm, identifying the blur area caused by wind by using an interference detection algorithm, and determining the aligned and corrected image; the interference detection algorithm specifically comprises: calculating the gradient amplitude of the local area of the image by using a gradient analysis-based method, judging the blur area according to the gradient threshold value, and dynamically adjusting the gradient threshold value based on real-time wind speed data: when the wind speed is less than 2 m / s, the threshold value is 10, and when the wind speed is greater than or equal to 2 m / s, the threshold value is 8; Step 3: if the local contrast of the plant contour in the aligned and corrected image is lower than a preset threshold value, performing local processing on the low-contrast area by using an image enhancement algorithm, removing the influence of environmental noise by using a noise extraction algorithm, and obtaining a contrast-enhanced image; the image enhancement algorithm specifically comprises: using a local histogram equalization algorithm (CLAHE), and adaptively setting the window size to 8×8 pixels according to the typical size of plant leaves, and dynamically adjusting the clipping value of the contrast enhancement based on the local contrast distribution of the image; the noise extraction algorithm specifically comprises: performing multi-scale decomposition on the image by using wavelet transform, and filtering out environmental noise by using a wavelet soft threshold method; Step 4: based on the contrast-enhanced image, integrating position data and visual data by using a sensor fusion algorithm, monitoring the position of the mobile device in real time, recording the tilt change of the mobile device in space by using an attitude monitoring, and determining the integrated pose data; the sensor fusion algorithm specifically comprises: using an extended Kalman filter (EKF)-based fusion algorithm, and weightedly fusing the position data and the visual data, and dynamically adjusting the weight by using an adaptive extended Kalman filter (AEKF) based on the innovation covariance; Step 5: for the integrated pose data, correcting the image distortion caused by the shaking of the mobile device by using a distortion rectification algorithm, adjusting the deviation of the mobile device during movement by using a trajectory correction algorithm, and obtaining a corrected plant position distribution map; the distortion rectification algorithm specifically comprises: obtaining radial distortion coefficients k1 and k2 by camera calibration, and correcting the image based on a radial distortion model; the trajectory correction algorithm specifically comprises: smoothing the movement trajectory of the mobile device by using a Kalman filter, and reducing the trajectory deviation based on the motion model prediction and update steps. Step 6: Based on the corrected plant position distribution map, a path planning algorithm is used to generate a mobile route with high coverage based on the field terrain characteristics, and a region recognition method is used to analyze the unmeasured field region to determine the optimized measurement path and final plant distribution data; the path planning algorithm is as follows: using digital elevation model (DEM) data, the path is planned by A* algorithm, and the cost function includes distance cost and slope cost, wherein when the flat terrain and the slope standard deviation σ≤5°, the distance weight is 0.7 and the slope weight is 0.3; when the complex terrain and σ>5°, the distance weight is 0.4 and the slope weight is 0.6; Step 7: If there is a repeated area in the optimized measurement path, a repeated area detection algorithm is used to identify the covered area and adjust the acquisition frequency, and a feedback mechanism is combined to dynamically adjust the acquisition angle of the mobile device to obtain high-precision distribution information for specific characteristics of the plant; the repeated area detection algorithm is specifically as follows: the structural similarity (SSIM) of adjacent frame images is calculated, and when the SSIM exceeds a threshold value, it is determined that it is a repeated area; the acquisition frequency is dynamically adjusted according to the overlap rate, and the formula is . wherein; to optimize the post-acquisition frequency, frames / second; For initial acquisition frequency, frames per second; For image structure similarity, the value is 0~1, and exceeding the threshold value is determined as a repeated area; to limit the minimum acquisition frequency to avoid data loss; The feedback mechanism dynamically adjusts the collection angle, which is specifically: through the built-in gyroscope and acceleration sensor, the mobile device posture data is obtained in real time, the deviation of the current collection angle and the target angle is calculated, and the PID control algorithm is used for angle correction.

2. The intelligent breeding phenotyping platform for mobile measurements according to claim 1, characterized in that, Step 1 is specifically as follows: A multi-spectral camera is used to obtain a plant image sequence in a dynamic field environment, and original image data is extracted according to the characteristics of plant leaves to obtain a target image set after preprocessing and screening; An image preprocessing technique is used to denoise the original image sequence, and environmental interference factors are preliminarily filtered to obtain a denoised preprocessing result image; According to the influence of the dynamic environment in the preprocessing result image, an image segmentation technique is applied to separate the plant leaf area and the background area, determine the clear boundary of the leaf area, and obtain a segmented segmentation result image; For the leaf reflection characteristic data in the segmentation result image, if abnormal fluctuations are detected in the reflection data, the mean value smoothing technique is used to correct the abnormal area to obtain a third image set after correction; Through the plant leaf characteristics in the third image set, key band data corresponding to the band selection algorithm is extracted, and it is judged whether the key band data meets the preset threshold range to obtain a fourth image set after screening; According to the image acquisition process data in the fourth image set, the influence degree of the dynamic field environment on the image quality is analyzed, and the final image set output by this sequence is taken as the input of step 2; For the leaf reflection analysis result in the final image set, the original data collected by the multi-spectral camera is combined to obtain the reflection characteristic distribution of the plant leaves under different bands, and the health status information of the plant leaves is judged.

3. The intelligent breeding phenotyping platform for mobile measurements according to claim 1, wherein, Step 2 is specifically as follows: For the initial image, an image alignment algorithm is used to process consecutive frames, and on the basis of time alignment and spatial alignment, inter-frame consistency correction is completed to obtain an aligned image group; Through the aligned image group, a disturbance detection algorithm is applied to analyze the wind influence in the dynamic environment, identify the specific position of the blurred area, and determine the disturbed image range; According to the disturbed image range, a region segmentation technique is used to separate the blurred area and the clear area to obtain a segmented image subset; For the segmented image subset, a pixel interpolation technique is used to repair the blurred area to obtain a repaired image combination; According to the repaired image combination, a contrast adjustment method is applied to optimize the image details and highlight the plant target area to determine an optimized image data set; Through the optimized image data set, if it is still detected that the local area is blurred, a preset threshold is used for screening to remove the image part that does not meet the standard, and a final available image set is obtained; For the final available image set, data normalization processing is used to unify the image format, and a standardized image set suitable for subsequent analysis is obtained.

4. The intelligent breeding phenotyping platform for mobile measurements and data collection of claim 1, wherein, Step 3 is specifically as follows: Through preliminary analysis of the image set, the pixel distribution of the plant contour region is detected, and it is judged whether the local contrast is lower than the preset threshold, to obtain a preliminary contrast evaluation result; If the preliminary contrast evaluation result shows that the local contrast is lower than the preset threshold, an image enhancement algorithm is used for local adjustment of the low-contrast region to generate an enhanced plant contour image; For the enhanced plant contour image, a noise extraction algorithm is used to identify and separate environmental noise to obtain a denoised intermediate image data; According to the denoised intermediate image data, it is detected whether there is residual noise interference in the plant image, and if there is residual noise, further removal is performed through iterative filtering processing to obtain a noise-free plant image; After obtaining the noise-free plant image, the edge details of the plant contour are sharpened to determine the final clear image; Through verification of the pixel distribution of the final clear image, the contour integrity of the plant image is judged to obtain a plant image group that meets the expectation.

5. The intelligent breeding phenotyping platform for mobile measurements and data collection of claim 1, wherein, Step 4 is specifically as follows: Through the combination of the plant image and the position information, a fusion method is used to track the position of the mobile device in real time to obtain preliminary positioning data of the mobile device in the current environment; According to the preliminary positioning data, the position of the mobile device is corrected in combination with the visual information to obtain a more accurate spatial position distribution; For the spatial position distribution, posture monitoring data of the mobile device during movement is obtained to record the specific situation of the inclination change and determine the dynamic adjustment information of the mobile device in space; The key points of the inclination change are extracted from the dynamic adjustment information, an analysis model established in advance is used to evaluate the spatial consistency, and the stability performance of the mobile device during movement is judged; If the stability performance is lower than the preset threshold, the position information and the visual information are processed again through a data fusion algorithm to obtain an optimized mobile device position state; Through the optimized mobile device position state, the spatial consistency is continuously tracked in combination with the posture monitoring data during movement to determine the final mobile device dynamic distribution result; After obtaining the final mobile device dynamic distribution result, the coverage range of the plant image is analyzed to obtain spatial correspondence data of the mobile device and the target region.

6. The intelligent breeding phenotyping platform for mobile measurements and data collection of claim 1, wherein, Step 5 is specifically as follows: For the collected integrated pose data, image preprocessing techniques are used to denoise and standardize the original image to obtain preliminary processed image data; Through a distortion correction algorithm, the image distortion caused by the shaking of the mobile device is corrected for the preliminary processed image data to generate a distortion corrected image; For the corrected preprocessed image, a trajectory correction algorithm is used to analyze the deviation trajectory of the mobile device during movement to determine a deviation adjustment parameter; According to the determined deviation adjustment parameter, the pre-processed image is corrected in position offset to generate a position corrected image; Plant position information is extracted from the position corrected image, and an image segmentation technique is used to separate the target area to obtain preliminary mapping data of plant distribution; For the preliminary mapping data, an abnormal point in the position mapping is adjusted through a spatial consistency verification method to obtain a final plant position distribution map; According to the final plant position distribution map, a data visualization tool is used to generate an intuitive distribution view to determine the complete presentation of plant distribution.

7. The intelligent breeding phenotyping platform for mobile measurements and data collection of claim 1, wherein, Step 6 includes the following specific steps: Through the corrected plant position distribution map, the pre-established geographic information database is used to obtain detailed feature data of the field terrain to determine preliminary analysis results of the terrain undulations and obstacle distribution; According to the preliminary analysis results of the terrain undulations and obstacles, a path planning algorithm is used in combination with a high-coverage target to generate an initial movement route scheme to obtain path planning data covering the main areas of the field; For the path planning data, a region recognition method is used to analyze the unmeasured area range of the field to determine whether there is an insufficient coverage area. If the unmeasured area exceeds a preset threshold, the path planning data is adjusted to obtain updated movement routes; From the updated movement routes, specific position information of the unmeasured area is obtained, and a local adjustment is made to the path through an optimization strategy to determine the priority path for supplementary measurement to obtain optimized measurement path data; According to the optimized measurement path data, in combination with the feature analysis results of the field terrain, a K-means clustering algorithm is used to classify and verify the plant positions to obtain optimized plant distribution information; Through the refined plant distribution information, in combination with the path data for supplementary measurement, the measurement results of all areas of the field are integrated to determine the final plant distribution data; For the final plant distribution data, a data storage technology is used to record it in a preset database to obtain a complete field plant distribution file.

8. The intelligent breeding phenotyping platform for mobile measurements and data collection of claim 1, wherein, Step 7 includes the following specific steps: Through a repeated area detection method, the measurement path is scanned to identify the distribution status of the repeated area and the covered part to obtain preliminary area division results; According to the preliminary area division results, the data collection frequency is adjusted for the repeated area to reduce the collection density of the repeated area to obtain an optimized collection frequency configuration; Through the optimized collection frequency configuration, in combination with a feedback mechanism, the mobile device is dynamically adjusted to determine the collection angle of the mobile device in different areas; According to the adjusted collection angle, directional data collection is performed for plant features to obtain high-precision data related to plant features; Using the obtained high-precision data, in combination with a preset threshold, the distribution information is calibrated to determine whether the data meets the precision requirements. If not, the collection angle is re-adjusted and the data is updated; Through the calibrated high-precision data, distribution information of plant features is constructed to obtain final distribution mapping results; For the distribution mapping results, a K-means clustering algorithm is used to classify the data to determine the distribution rules of different plant features in space.

Images