Plant salt tolerance grade evaluation method based on three-dimensional point cloud

Abstract

The invention discloses a plant salt tolerance grade evaluation method based on three-dimensional point cloud. The method comprises the following steps: obtaining three-dimensional point cloud data of a plant to be detected; processing the three-dimensional point cloud data of the plant to be detected, and removing noise and abnormal points in the data; inputting the three-dimensional point cloud data into a learning model to obtain three-dimensional point cloud data of each leaf of the plant to be detected; the salt tolerance coefficient of the three-dimensional point cloud data of each leaf is obtained, and the salt tolerance coefficient of the plant is obtained; carrying out salt tolerance evaluation and grading; the fresh weight of each plant is measured, and photosynthetic phenotype parameters of the plants are measured through a photosynthetic phenotype imaging system; and verifying and optimizing. According to the plant salt tolerance grade evaluation method based on the three-dimensional point cloud, the limitation of an existing plant salt tolerance evaluation method is overcome, salt tolerance key characteristics are extracted through the data dimension reduction technology to establish a salt tolerance evaluation membership function, plant salt tolerance quantification and automatic grade division are achieved, and the plant salt tolerance evaluation efficiency is improved. And a scientific basis is provided for screening salt-resistant varieties.

Description

Technical Field

[0001] The invention relates to the technical field of plant screening, and in particular to a plant salt tolerance grade evaluation method based on three-dimensional point cloud. Background Art

[0002] Existing methods for evaluating plant salt tolerance mainly rely on physiological index measurements and subjective judgments in laboratory environments. However, these methods have obvious limitations: first, manual measurement is inefficient, the amount of data is small, and it is difficult to fully reflect the salt tolerance of plants; second, subjective judgments are affected by human factors and lack objectivity and accuracy. Therefore, how to achieve a more efficient, accurate and objective salt tolerance assessment has become an urgent problem to be solved.

[0003] At present, there are some plant salt tolerance assessment tools based on two-dimensional image analysis and machine learning technology on the market. For example, two-dimensional images of plant leaves collected by RGB cameras are used to infer their salt tolerance by analyzing color, texture and morphological characteristics. Although these methods have made progress in improving evaluation efficiency, are easy to operate, low-cost, and suitable for laboratory environments, they have obvious limitations. Two-dimensional images cannot fully reflect the three-dimensional structural information of Chinese cabbage, especially when it comes to complex features such as leaf thickness, spatial morphology, and curling, the accuracy and comprehensiveness of the evaluation results are limited.

[0004] In addition, although technologies such as fluorescence imaging and hyperspectral imaging have been used to a certain extent in the assessment of plant physiological status, they also rely on two-dimensional image data. Fluorescence imaging is evaluated through photosynthesis-related indicators, which makes it difficult to capture the three-dimensional morphological characteristics of Chinese cabbage and is more sensitive to environmental conditions. Although hyperspectral imaging can analyze the chemical composition of plants, its processing is complex and does not pay enough attention to morphological characteristics. Neither of them can effectively characterize the three-dimensional structure of Chinese cabbage, especially when evaluating spatial characteristics such as the degree of leaf curling. In addition, existing technologies usually only focus on a few physiological parameters such as chlorophyll content, ignoring the important role of three-dimensional morphological characteristics of leaves in salt tolerance assessment, which leads to incomplete and inaccurate evaluation results.

[0005] In the evaluation of plant salt tolerance, taking Chinese cabbage as an example, salt stress has a significant impact on its three-dimensional morphology, especially the increased curling of the leaves. The leaves of other plants such as lettuce and fruits also show curling characteristics under salt stress. This change not only affects the efficiency of photosynthesis, but is also closely related to the salt tolerance of the plant. Therefore, studying the changes in the three-dimensional morphology of these plants under salt stress can provide more comprehensive and accurate information for salt tolerance evaluation.

[0006] In order to overcome the limitations of existing methods, it is particularly important to develop an automated salt tolerance assessment method based on three-dimensional point clouds. By obtaining the three-dimensional point cloud of Chinese cabbage, its spatial morphology, leaf thickness, curling degree and other characteristics can be fully reflected, thereby improving the accuracy and comprehensiveness of salt tolerance assessment. In addition, combined with deep learning technology, it is possible to effectively achieve accurate assessment of plant salt tolerance. This method provides a new technical path and idea for plant salt tolerance research. Summary of the invention

[0007] The purpose of the present invention is to provide a plant salt tolerance grade assessment method based on three-dimensional point cloud to overcome the limitations of existing plant salt tolerance assessment methods. This method uses a visible light camera to obtain three-dimensional point cloud data of plants of different varieties, and combines deep learning technology to perform instance segmentation and parameter calculation on each plant leaf. On this basis, by comparing the experimental group with the control group, relative characteristic parameters are extracted to analyze the key features of salt tolerance. Principal component analysis is used to extract key features of salt tolerance and establish a membership function for salt tolerance evaluation, ultimately achieving quantitative plant salt tolerance and automatic grade division, providing a scientific basis for the screening of salt-tolerant varieties.

[0008] To achieve the above object, the present invention provides a method for evaluating plant salt tolerance based on three-dimensional point cloud, comprising the following steps:

[0009] S1. Acquire three-dimensional point cloud data of plants to be tested, and divide the plants to be tested of each variety into a control group and a test group;

[0010] S2, processing the three-dimensional point cloud data of the plant to be tested to remove noise and abnormal points in the three-dimensional point cloud data of the plant to be tested;

[0011] S3, inputting the three-dimensional point cloud data of the plant to be tested, after removing noise and abnormal points, into the trained deep learning model to obtain the three-dimensional point cloud data of each leaf of the plant to be tested;

[0012] S4, inputting the three-dimensional point cloud data of each blade into the point cloud completion model to complete the point cloud of the blade with missing information;

[0013] S5, obtaining the salt tolerance coefficient of each leaf three-dimensional point cloud data, averaging the salt tolerance coefficients of all leaves of each plant to be tested, averaging the salt tolerance coefficients of the test group, and comparing them with the salt tolerance coefficients of the control group to obtain the salt tolerance coefficient of the plant;

[0014] S6. Salt tolerance evaluation and classification: The physiological indexes measured on the last day of the test are used as the criteria for establishing salt tolerance classification. Through correlation analysis with the salt tolerance coefficient measured every day, a specific number of days is established as the basis for judging whether the plant is salt tolerant;

[0015] S7. On the last day of measurement, the fresh weight of each plant was measured and its photosynthetic phenotypic parameters were measured using the existing photosynthetic phenotypic imaging system. Finally, the difference results of Hue, fresh weight, near infrared, far infrared, green light, and red light were screened out through difference analysis.

[0016] S8. Verification and optimization: verify the accuracy of the salt tolerance classification through actual data, and optimize the screening method to ensure that the screening method has high generalization ability and practical value.

[0017] Preferably, in step S2, the process of processing the three-dimensional point cloud data of the plant to be tested includes: scale matching, scale matching of the three-dimensional point cloud data of plants of different varieties, ensuring that data of different varieties are compared at the same scale, and setting plants of each variety to a non-crossing state to produce a data structure that conforms to the format of the instance segmentation data set.

[0018] Preferably, in step S3, the trained deep learning model includes: a preprocessing module, a feature extraction module, and a point cloud segmentation module.

[0019] Preferably, in step S4, using PF-Net to complete the point cloud is an effective method because the model can learn the distribution of data and generate high-quality completion results, ensuring that the completed point cloud can reflect the true blade shape and improve the accuracy of subsequent analysis.

[0020] Preferably, the preprocessing module includes normalization of the three-dimensional point cloud data, normalizing the input three-dimensional point cloud data to a standard coordinate range, and also includes data enhancement, which increases the diversity of training data and improves the generalization ability of the model through rotation, translation, scaling and noise injection.

[0021] Preferably, the feature extraction module uses OneFormer3D as the core feature extraction network to segment the three-dimensional point cloud data. The network can capture the long-distance dependencies and global contextual information of the three-dimensional point cloud data through the Transformer self-attention mechanism, thereby improving the model's ability to understand complex plant structures.

[0022] Preferably, the point cloud segmentation module performs instance segmentation on the three-dimensional point cloud data of the plant, segments each leaf from the whole, and generates separate three-dimensional point cloud data for each leaf; instance segmentation includes evaluation of segmentation accuracy and error analysis. After each training, the segmentation results are evaluated by standard evaluation indicators, and the segmentation results are visualized. At the same time, missed and misclassified problems are identified. Based on the error analysis results, the model parameters are adjusted to enhance the robustness of the model. At the same time, the early stopping method is used in the training of the model to avoid overfitting, and cross-validation is used to ensure the generalization ability of the model.

[0023] Preferably, in step S5, the process of obtaining the salt tolerance coefficient of each leaf three-dimensional point cloud data is as follows: multiple parameter calculations are performed on the leaf three-dimensional point cloud data, and the membership of the salt tolerance grade is calculated based on the obtained parameters: the same characteristic parameters of each leaf are extracted from the original control group, and the average value of each parameter is calculated, and the characteristic parameters of each plant in the experimental group are ratioed with the average value of the control group to obtain relative characteristic parameters, and then the dimensions of different parameters are normalized, and 5 principal components are extracted by the principal component analysis method, and the corresponding contribution rate and the normalized principal components are used to calculate the comprehensive salt tolerance coefficient of each leaf, so as to obtain the salt tolerance coefficient of each variety and each experimental group plant.

[0024] Preferably, in step S6, the salt tolerance evaluation and grading are specifically as follows: the salt tolerance coefficient of the test group in step S1 is averaged, and compared with the salt tolerance coefficient of the control group to obtain the salt tolerance coefficient of the plant of this variety, and the physiological indicators measured on the last day of the test are used as the standard for establishing the salt tolerance grade, and a correlation analysis is performed with the salt tolerance coefficient measured every day to establish a certain number of days as the basis for judging whether the plant is salt tolerant; a multivariate stepwise regression model is established using the original relative characteristic parameters and the salt tolerance coefficient, and finally the characteristic parameters that determine the salt tolerance curling of the plant are screened out, and then cluster analysis is performed according to the calculated salt tolerance coefficient, and a convolutional neural network classification model based on relative characteristic parameters is established. This model integrates LSTM to better solve the accurate classification of salt tolerance of different plants.

[0025] Preferably, in step S8, the optimization process includes adjusting model parameters, improving data processing flow, and continuously improving the reliability of the evaluation method based on a feedback mechanism.

[0026] The advantages and positive effects of the plant salt tolerance grade assessment method based on three-dimensional point cloud described in the present invention are:

[0027] 1. The automated 3D point cloud data acquisition and deep learning instance segmentation technology of the present invention significantly improves the speed of plant salt tolerance assessment. Compared with traditional manual measurement methods, automated technology can quickly process large amounts of data, reducing tedious manual operations and time costs.

[0028] 2. The introduction of deep learning technology can accurately analyze the morphological characteristics of plants. By dividing the salt tolerance levels based on multi-dimensional relative characteristic parameters, it can effectively avoid errors caused by human subjective judgment, thereby improving the accuracy and reliability of the evaluation results.

[0029] 3. The point cloud data of different varieties of plants are scale-matched and normalized to ensure the comparability of different data. This standardized processing method improves the reliability of salt tolerance assessment and enables data of different varieties to be fairly compared at a unified scale.

[0030] 4. The technical solution of the present invention has strong universality and can be extended to the salt tolerance assessment of various crops or the classification and assessment of other agronomic traits. The extensibility of this technology provides broad application prospects and research value for agricultural research.

[0031] 5. The present invention provides the ability to optimize the model through a feedback mechanism, ensuring that the salt tolerance assessment method has efficient generalization ability and practical value.

[0032] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A schematic diagram of the operation of the present invention;

[0034] Figure 2 This is a schematic diagram of the collection environment of Example 1 of the present invention;

[0035] Figure 3 1 is a diagram of the feature matching process of the three-dimensional point cloud of Example 1 of the present invention, wherein (A) is an original point cloud that is not at the same scale, (B) is a point cloud that is not at the same scale for visualizing different plants, (C) is a point cloud that is at the same scale after scale matching and normalization, and (D) is a point cloud that is at the same scale after visualizing scale matching and normalization;

[0036] Figure 4 A process diagram of calculating parameters for segmentation and post-segmentation registration in Example 1 of the present invention;

[0037] Figure 5 The following are analysis diagrams of the significant difference verification indicators between the Chinese cabbage plant test group and the control group measured on the last day of Example 3 of the present invention, wherein (A) is a comparison of fresh weight and Hue value, and (B) is a comparison of multispectral grayscale value and chlorophyll difference. DETAILED DESCRIPTION

[0038] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0039] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0040] Unless otherwise defined, the reagents and equipment used in the present invention are all obtained from the prior art.

[0041] A method for evaluating plant salt tolerance based on three-dimensional point cloud, the process is as follows Figure 1As shown in the figure, the limitations of existing plant salt tolerance assessment methods are overcome. Taking Chinese cabbage as an example, the three-dimensional morphology including the degree of curling of Chinese cabbage will change after being subjected to salt (sodium chloride) stress. The three-dimensional point cloud data of Chinese cabbage of different varieties is obtained by a visible light camera, and the deep learning technology is combined to perform instance segmentation and parameter calculation on each Chinese cabbage leaf. The principal component analysis is used to extract the key features of salt tolerance and establish the salt tolerance evaluation membership function, which finally realizes the quantification of Chinese cabbage salt tolerance and automatic classification of grades.

[0042] Example 1

[0043] A method for evaluating plant salt tolerance based on three-dimensional point cloud comprises the following steps:

[0044] Step 1: Obtain the 3D point cloud data of the plants to be tested. Each variety of plants to be tested is divided into a control group and a test group. Use a high-precision visible light camera to take a 360-degree surround shot of the Chinese cabbage (such as Figure 2 As shown in the figure, place the Chinese cabbage on a rotatable turntable and fix the Chinese cabbage in the center of the turntable to ensure that its position is stable and does not move during the rotation. Set equal time intervals to take pictures, take a picture of a plant in one minute, take 30 pictures a week, store the pictures, and save the collected images to guide the establishment of 3D point cloud data. Name and save them according to different varieties every day. Set a test group and a control group for each Chinese cabbage variety. The plants in the test group are watered with 200mmol / L sodium chloride solution every day. 53 Chinese cabbage varieties were collected, and 21 samples, 11 test groups, and 10 control groups were set for each variety. The 3D point cloud data is organized and classified according to information such as variety, test group, control group, and shooting date to ensure that the naming and storage structure of the data file can support subsequent processing and analysis requirements.

[0045] Step 2: Process the three-dimensional point cloud data of the plant to be tested to remove noise and abnormal points in the three-dimensional point cloud data of the plant to be tested. Remove invalid points introduced by noise or acquisition errors to ensure the purity and accuracy of the data. Use Gaussian filtering or median filtering algorithms to smooth the point cloud data and remove noise points. These noise points may be caused by ambient light interference, equipment errors or surface reflections. Filtering can effectively reduce random errors in point cloud data and improve data quality. Point cloud data is processed by filtering and denoising to remove isolated points and abnormal points to improve the accuracy of subsequent processing. Use an optimized density clustering algorithm to detect and remove isolated points and abnormal points in the three-dimensional point cloud data. These points are usually caused by acquisition errors and affect the integrity of the data.

[0046] Scale matching: Scale matching is performed on the point cloud data of different varieties of Chinese cabbage to ensure that the data of different varieties are compared at the same scale. Scale matching uses a method based on feature points or ICP (Iterative Closest Point) algorithm to align the coordinates of each point cloud model to ensure that the center point and main axis of each point cloud model are consistent, so that data of different varieties can be compared in the same coordinate system. The size of different varieties of Chinese cabbage is normalized, and all point cloud data are adjusted to the same scale range using the target scale factor. This can avoid the size impact caused by variety differences and ensure the fairness of data comparison. The coordinate system of the point cloud data is uniformly adjusted to the same reference coordinate system to ensure consistency between different batches of data, such as Figure 3 shown.

[0047] Each variety of Chinese cabbage is set to a non-intersecting state to create a data structure that conforms to the instance segmentation dataset format. Non-intersecting means that the point cloud data of each variety of Chinese cabbage is laid out in a non-intersecting state to ensure that the point cloud data of different varieties do not overlap in the dataset structure. This is not only convenient for subsequent analysis, but also avoids data confusion. In the dataset production, the data of different days are regarded as a scene, and the leaf data of different varieties of Chinese cabbage are regarded as sub-regions in the scene. Each sub-region in the scene contains point cloud data of a specific time point and a specific variety. This structured dataset design is well compatible with the input requirements of deep learning models.

[0048] Step 3: Input the three-dimensional point cloud data with noise and abnormal points of the plant to be tested into the trained deep learning model to obtain the three-dimensional point cloud data of each leaf of the plant to be tested; the trained deep learning model includes: a preprocessing module, a feature extraction module, and a point cloud segmentation module.

[0049] The preprocessing module includes normalization of 3D point cloud data, normalizing the input 3D point cloud data to a standard coordinate range, and also includes data enhancement, which increases the diversity of training data and improves the generalization ability of the model through rotation, translation, scaling and noise injection.

[0050] The feature extraction module uses OneForme3D as the core feature extraction network to segment three-dimensional point cloud data. Through the Transformer self-attention mechanism, this network can capture the long-distance dependencies and global contextual information of three-dimensional point cloud data, and enhance the model's ability to understand complex plant structures.

[0051] The point cloud segmentation module performs instance segmentation on the three-dimensional point cloud data of the plant, separates each leaf from the whole, and generates separate three-dimensional point cloud data for each leaf; instance segmentation includes the evaluation of segmentation accuracy and error analysis. After each training, the segmentation results are evaluated using standard evaluation indicators and visualized. At the same time, missed and misclassified problems are identified. Based on the error analysis results, the model parameters are adjusted to enhance the robustness of the model. At the same time, the early stopping method is used in the model training to avoid overfitting, and cross-validation is used to ensure the generalization ability of the model.

[0052] like Figure 4 As shown in the figure, instance segmentation uses a unified model based on the OneFormer3D framework to perform instance segmentation on the point cloud data of Chinese cabbage, segmenting each leaf from the whole and generating separate point cloud data for each leaf. OneFormer3D uses a unified model to handle instance segmentation, semantic segmentation, and panoptic segmentation tasks. The decoder design of OneFormer3D based on Transformer enables the model to dynamically adapt to different segmentation tasks. The mask classifier of OneFormer3D can accurately segment each leaf of Chinese cabbage from the whole and generate separate point cloud data, ensuring that the details of each leaf are effectively captured and analyzed, providing a high-quality data foundation for subsequent parameter calculation and salt tolerance evaluation.

[0053] The OneFormer3D model performs instance segmentation on the point cloud data of Chinese cabbage. The model successfully segments each leaf from the whole. The segmentation results show that the point cloud data of each leaf is accurately extracted, and the edges and details of the leaves are effectively captured.

[0054] Step 4: Use the completion network PF-Net to complete the point cloud of the missing leaves after cutting, which effectively improves the structural integrity of the missing area. This step uses the Pred-GT (the distance from the predicted point to the nearest real point) and GT-Pred (the distance from the real point to the nearest predicted point) distance for evaluation, ranging from 0.5-1.0, which shows that the completed point cloud has high accuracy in shape details and provides a reliable data basis for subsequent salt tolerance evaluation. Using PF-Net for point cloud completion is an effective method because the model can learn the distribution of data and generate high-quality completion results. Ensure that the completed point cloud can reflect the true shape of the leaf and improve the accuracy of subsequent analysis.

[0055] Step 5, obtain the salt tolerance coefficient of each leaf three-dimensional point cloud data, average the salt tolerance coefficients of all leaves of each plant to be tested, average the salt tolerance coefficients of the test group, compare them with the salt tolerance coefficients of the control group, and obtain the salt tolerance coefficient of the plant. Perform multiple parameter calculations on the leaf three-dimensional point cloud data, and calculate the membership of the salt tolerance level based on the obtained parameters: extract the same characteristic parameters of each leaf from the original control group, and calculate the average value of each parameter, calculate the ratio of the characteristic parameters of each plant in the experimental group to the average value of the control group, and obtain the relative characteristic parameters, and then normalize the dimensions of different parameters, extract 5 principal components by principal component analysis method, and calculate the comprehensive salt tolerance coefficient of each leaf according to the corresponding contribution rate and the normalized principal component, and obtain the salt tolerance coefficient of each plant in each test group of each variety, and calculate the corresponding average salt tolerance coefficient of each Chinese cabbage as the salt tolerance coefficient of the plant. Among them, the comprehensive salt tolerance coefficient reflects the comprehensive performance of the leaves in various parameters and is highly correlated with salt tolerance.

[0056] Parameter calculation: Multiple parameter calculations were performed on the segmented leaf point cloud data, including length, width, height and their ratios (length / height, length / width, length / (length+width+height)), three-surface projection areas and their ratios, convex hull area and point cloud volume, etc. In addition, 32 parameters including the perimeter, scale, and roughness of the point cloud were calculated. These parameters fully reflect the geometric morphology and spatial characteristics of the leaf point cloud and are recorded in the corresponding storage table, providing rich data information for subsequent analysis. All calculated parameters are structured and recorded in a table according to information such as leaf number, variety number, and number of days. Each row in the table represents all parameters of a leaf. Then, each parameter value of each Chinese cabbage in the experimental group was divided by the control group (i.e., the average value of all leaves in all samples) to obtain the relative characteristic parameters of each parameter, providing a data basis for subsequent salt tolerance analysis.

[0057] Step 6, salt tolerance evaluation and grading, using the physiological indicators measured on the last day of the experiment as the standard for establishing salt tolerance grades, and conducting correlation analysis with the salt tolerance coefficient measured every day to establish a specific number of days as the basis for judging whether the plant is salt-tolerant; salt tolerance evaluation and grading are specifically as follows: averaging the salt tolerance coefficient of the experimental group in step 1, comparing it with the salt tolerance coefficient of the control group, and obtaining the salt tolerance coefficient of the plant of this variety, using the physiological indicators measured on the last day of the experiment as the standard for establishing salt tolerance grades, and conducting correlation analysis with the salt tolerance coefficient measured every day to establish a specific number of days as the basis for judging whether the plant is salt-tolerant; using the original relative characteristic parameters and salt tolerance coefficient to establish a multivariate stepwise regression model, and finally screening out the characteristic parameters that determine the salt tolerance curling of the plant, and then performing cluster analysis according to the calculated salt tolerance coefficient, and establishing a convolutional neural network classification model based on relative characteristic parameters, which integrates LSTM to better solve the accurate classification of salt tolerance grades of different plants.

[0058] Step 7. On the last day of measurement, the fresh weight of each Chinese cabbage plant was measured and its photosynthetic phenotypic parameters were measured using an existing photosynthetic phenotypic imaging system. The photosynthetic phenotypic parameters included chlorophyll and spectral data. Finally, the difference results of Hue, fresh weight, near infrared, far infrared, green light, and red light were screened out through difference analysis for subsequent analysis.

[0059] Step 8: Verification and optimization: verify the accuracy of the salt tolerance classification through actual data, and optimize the screening method to ensure that the screening method has high generalization ability and practical value. The optimization process includes adjusting model parameters, improving data processing procedures, and continuously improving the reliability of the evaluation method based on the feedback mechanism.

[0060] Example 2

[0061] Instance segmentation technology based on deep learning.

[0062] The grand unified model based on the OneFormer3D framework is used to perform instance segmentation on the point cloud data of Chinese cabbage, segment each leaf from the whole, and generate separate point cloud data for each leaf.

[0063] In the data input part of the main function of the unified model of the OneFormer3D framework, the corresponding categories are set respectively. The first model selects the first 32 varieties of Chinese cabbage and sets 32 different colors for model visualization.

[0064] Model configuration: In the model, set the backbone network SpConvUNet to 4 layers, the decoder QueryDecoder to 2 layers, reduce the number of strength queries to 100, set the voxel size to 0.1, set the number of model channels to 64, the number of strength categories and the number of semantic categories to 32, first select the data collected in the first five days, set it to 5 scenes, select the first four scenes as the training set, and the fifth scene as the test set. The training set sampling points are 5000, and the test part uses MultiScaleFlipAug for evaluation. In training, due to memory limitations, set the Batch Size to 1, the number of working threads to 1, the optimizer to use the Adam optimizer, the learning rate to 0.0001, the weight decay to 0.05, set the model to be saved once after each epoch, and set the epoch to 100.

[0065] OneFormer3D improves processing efficiency and accuracy by unifying the model to handle instance segmentation, semantic segmentation, and panoramic segmentation tasks. This multi-tasking capability ensures accurate segmentation and integrity analysis of leaves when processing 3D point cloud data of Chinese cabbage. OneFormer3D's Transformer-based decoder design enables the model to dynamically adapt to different segmentation tasks, improving the reliability and consistency of segmentation results. By applying the OneFormer3D framework to instance segmentation of Chinese cabbage point cloud data, the accuracy and efficiency of leaf segmentation can be significantly improved, providing high-quality 3D point cloud data support for subsequent salt tolerance analysis. It not only optimizes the data processing process, but also improves the performance of the overall model, providing strong technical support for Chinese cabbage variety improvement and screening.

[0066] OneFormer3D's mask classifier accurately segments each leaf of Chinese cabbage from the whole, generating separate 3D point cloud data. This precise mask classification capability ensures that the details of each leaf are effectively captured and analyzed, providing a high-quality data foundation for parameter calculation and salt tolerance assessment. OneFormer3D's dynamic prediction capability enables more accurate detection and correction of segmentation errors during the evaluation process. If the segmentation result is found to have a large error, the model performance can be optimized by adjusting the model parameters or retraining to ensure that the final segmentation result has high accuracy and consistency.

[0067] Evaluate the accuracy of instance segmentation through cross-validation or other evaluation methods to ensure that the segmentation results of the model are highly accurate and consistent. If the segmentation error is large, adjust the model parameters or retrain.

[0068] The OneFormer3D model was used to perform instance segmentation on the point cloud data of Chinese cabbage. The model successfully segmented each leaf from the whole. The segmentation results show that the point cloud data of each leaf is accurately extracted, and the edges and details of the leaves are effectively captured. After instance segmentation processing, the point cloud data of the leaves is successfully segmented into multiple independent point cloud blocks. These independent point cloud blocks have good geometric shapes and structures, and can provide a reliable data basis for subsequent analysis.

[0069] Example 3

[0070] Feature extraction technology based on data dimensionality reduction.

[0071] The segmented blade point cloud data is used to calculate multiple parameters, including point cloud convex hull, point cloud surface area and volume, point cloud length, point cloud width, point cloud height, point cloud aspect ratio, point cloud aspect ratio, point cloud projection, point cloud perimeter, point cloud scale, point cloud roughness and other 32 parameters, which are recorded in the corresponding storage table. All calculation parameters are recorded in an Excel table in a structured manner according to the blade number, variety number, day number and other information. Each row in the table represents all the parameters of a blade, which is convenient for subsequent statistical analysis and processing.

[0072] Calculate leaf parameters in batches and store them in a structured Excel spreadsheet to improve data processing efficiency. Use principal component analysis to extract key features from multidimensional relative characteristic parameters, calculate the comprehensive salt tolerance coefficient, and reduce redundant information. By calculating the salt tolerance coefficient of the leaves and finding the average value of each plant in the test group, the salt tolerance level of the Chinese cabbage is automatically divided, and the optimal evaluation time point is determined. Analyze the correlation between the physiological indicators measured on the last day of Chinese cabbage and the salt tolerance coefficient, determine the optimal evaluation time point, and accurately judge the salt tolerance level. The salt tolerance evaluation results are displayed by measuring the photosynthetic phenotypic parameters of the Chinese cabbage plants through a photosynthetic phenotypic imaging system, and the salt tolerance differences of different varieties are intuitively displayed in the form of charts, such as Figure 5 shown.

[0073] Figure 5 (A) shows the difference in fresh weight and Hue of Chinese cabbage plants between the experimental group and the control group on the last day. Fresh weight refers to the actual weight of Chinese cabbage when it is picked, which can reflect the plant's ability to adapt to salt during its growth. Figure 5 (A) shows that the fresh weight of Chinese cabbage in the experimental group is significantly different from that in the control group. Usually, the fresh weight of the experimental group is significantly lower than that of the control group, which indicates that the growth of the experimental group is restricted under salt stress. The Hue value is a metric in the color model that reflects the color information of the leaves. Different Hue values ​​can represent the color changes of leaves under different physiological conditions. Figure 5(A) compares the difference in Hue value between the experimental group and the control group; significant changes in Hue value reflect the impact of salt stress on leaf color. Usually salt stress will cause changes in leaf color, such as yellowing or fading. The difference in Hue value between the experimental group and the control group can be used to evaluate the significance of this effect. Figure 5 (B) shows the comparison of multi-spectral grayscale value and chlorophyll difference. Figure 5 (B) shows the difference in grayscale values ​​between the experimental group and the control group in the multispectral image. Multispectral grayscale values ​​evaluate the health status of plants by analyzing the reflectance intensity of plants under different spectra. Different grayscale values ​​can indicate the photosynthesis efficiency and physiological status of leaves. Figure 5 The significant difference shown in (B) indicates that the difference in multispectral grayscale values ​​between the experimental group and the control group can reflect the effect of salt stress on the spectral characteristics of plants; chlorophyll content is a key indicator for evaluating plant health and photosynthesis efficiency. Figure 5 (B) compares the difference in chlorophyll content between the experimental group and the control group. The significant difference indicates that the chlorophyll content of the experimental group decreased due to salt stress, which further illustrates the effect of salt on the physiological state of plants.

[0074] Therefore, the present invention adopts the above-mentioned plant salt tolerance grade assessment method based on three-dimensional point cloud to overcome the limitations of existing plant salt tolerance assessment methods. This method obtains three-dimensional point cloud data of plants of different varieties through a visible light camera, and combines deep learning technology to perform instance segmentation and parameter calculation on each plant leaf. On this basis, relative characteristic parameters are extracted by comparing the experimental group with the control group, and the key characteristics of salt tolerance are analyzed. The key characteristics of salt tolerance are extracted by principal component analysis to establish a salt tolerance evaluation membership function, and finally the quantification of plant salt tolerance and automatic classification of grades are realized, providing a scientific basis for the screening of salt-tolerant varieties.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for evaluating plant salt tolerance based on three-dimensional point cloud, characterized in that: The steps include: S1. Acquire three-dimensional point cloud data of plants to be tested, and divide the plants to be tested of each variety into a control group and a test group; S2, processing the three-dimensional point cloud data of the plant to be tested to remove noise and abnormal points in the three-dimensional point cloud data of the plant to be tested; S3, inputting the three-dimensional point cloud data of the plant to be tested, after removing noise and abnormal points, into the trained deep learning model to obtain the three-dimensional point cloud data of each leaf of the plant to be tested; S4, inputting the three-dimensional point cloud of each blade into the point cloud completion model to complete the point cloud of the blade with missing information; S5, obtaining the salt tolerance coefficient of each leaf three-dimensional point cloud data, averaging the salt tolerance coefficients of all leaves of each plant to be tested, averaging the salt tolerance coefficients of the test group, and comparing them with the salt tolerance coefficients of the control group to obtain the salt tolerance coefficient of the plant; S6. Salt tolerance evaluation and classification: The physiological indexes measured on the last day of the test are used as the criteria for establishing salt tolerance classification. Through correlation analysis with the salt tolerance coefficient measured every day, a specific number of days is established as the basis for judging whether the plant is salt tolerant; S7. On the last day of measurement, the fresh weight of each plant was measured and its photosynthetic phenotypic parameters were measured using the existing photosynthetic phenotypic imaging system. Finally, the difference results of Hue, fresh weight, near infrared, far infrared, green light, and red light were screened out through difference analysis. S8. Verification and optimization: verify the accuracy of the salt tolerance classification through actual data, and optimize the screening method to ensure that the screening method has high generalization ability and practical value.

2. The method for evaluating plant salt tolerance based on three-dimensional point cloud according to claim 1, characterized in that: In step S2, the process of processing the three-dimensional point cloud data of the plant to be tested includes: scale matching, scale matching of the three-dimensional point cloud data of plants of different varieties, ensuring that the data of different varieties are compared at the same scale, and setting the plants of each variety to a non-crossing state to produce a data structure that conforms to the format of the instance segmentation data set.

3. The method for evaluating plant salt tolerance based on three-dimensional point cloud according to claim 1, characterized in that: In step S3, the trained deep learning model includes: a preprocessing module, a feature extraction module, and a point cloud segmentation module.

4. The method for evaluating plant salt tolerance based on three-dimensional point cloud according to claim 3, characterized in that: The preprocessing module includes normalization of 3D point cloud data, normalizing the input 3D point cloud data to a standard coordinate range, and also includes data enhancement, which increases the diversity of training data and improves the generalization ability of the model through rotation, translation, scaling and noise injection.

5. The method for evaluating plant salt tolerance based on three-dimensional point cloud according to claim 3, characterized in that: The feature extraction module uses OneFormer3D as the core feature extraction network to perform instance segmentation on 3D point cloud data. Through the Transformer self-attention mechanism, this network can capture the long-distance dependencies and global contextual information of 3D point cloud data, thereby improving the model's ability to understand complex plant structures.

6. The method for evaluating plant salt tolerance based on three-dimensional point cloud according to claim 3, characterized in that: The point cloud segmentation module performs instance segmentation on the three-dimensional point cloud data of the plant, separates each leaf from the whole, and generates separate three-dimensional point cloud data for each leaf; instance segmentation includes the evaluation of segmentation accuracy and error analysis. After each training, the segmentation results are evaluated using standard evaluation indicators and visualized. At the same time, missed and misclassified problems are identified. Based on the error analysis results, the model parameters are adjusted to enhance the robustness of the model. At the same time, the early stopping method is used in the model training to avoid overfitting, and cross-validation is used to ensure the generalization ability of the model.

7. The method for evaluating plant salt tolerance based on three-dimensional point cloud according to claim 1, characterized in that: In step S5, the process of obtaining the salt tolerance coefficient of each leaf three-dimensional point cloud data is as follows: multiple parameters are calculated for the leaf three-dimensional point cloud data, and the membership of the salt tolerance grade is calculated based on the obtained parameters: the same characteristic parameters of each leaf are extracted from the original control group, and the average value of each parameter is calculated, and the characteristic parameters of each plant in the experimental group are ratioed with the average value of the control group to obtain relative characteristic parameters, and then the dimensions of different parameters are normalized, and 5 principal components are extracted by the principal component analysis method, and the corresponding contribution rate and the normalized principal components are used to calculate the comprehensive salt tolerance coefficient of each leaf, and the salt tolerance coefficient of each variety and each experimental group of plants is obtained.

8. The method for evaluating plant salt tolerance based on three-dimensional point cloud according to claim 1, characterized in that: In step S6, the salt tolerance evaluation and grading are specifically as follows: the salt tolerance coefficient of the test group in step S1 is averaged, and compared with the salt tolerance coefficient of the control group to obtain the salt tolerance coefficient of the plant of this variety, and the physiological indicators measured on the last day of the test are used as the standard for establishing the salt tolerance grade, and a correlation analysis is performed with the salt tolerance coefficient measured every day to establish a certain number of days as the basis for judging whether the plant is salt-tolerant; a multivariate stepwise regression model is established using the original relative characteristic parameters and the salt tolerance coefficient, and finally the characteristic parameters that determine the salt tolerance curling of the plant are screened out, and then cluster analysis is performed according to the calculated salt tolerance coefficient, and a convolutional neural network classification model based on relative characteristic parameters is established. This model integrates LSTM to better solve the accurate classification of salt tolerance of different plants.

9. The method for evaluating plant salt tolerance based on three-dimensional point cloud according to claim 1, characterized in that: In step S8, the optimization process includes adjusting model parameters, improving data processing flow, and continuously improving the reliability of the evaluation method based on a feedback mechanism.

Images