Image Analysis Method and System for Root Growth Status of Seedlings after Transplanting

Abstract

This invention discloses a method and system for analyzing root growth status images after seedling transplantation, specifically relating to the field of root growth status image analysis technology. Based on the identifier between a single seedling and the image acquisition device, a temporal acquisition and numbering system for root zone images is constructed. A shallow-buried near-infrared imaging device is used to acquire multi-frame image sequences, and through image rasterization, alignment resampling, and structural enhancement processing, a root map with spatial consistency and separable boundaries is formed. Further, edge migration vector fields and active root tip regions are extracted, and the root tip activity change rate and root incremental density gradient are calculated. These are fused to generate a recovery trend index, and combined with trend labels and structural feature vectors, an activity level determination and response suggestion map is constructed. This enables intelligent analysis and management decisions regarding root recovery status, effectively solving the problems of lagging root status identification, strong subjectivity in judgment, and lack of dynamic quantitative support.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of root growth state image analysis, and more particularly to a seedling transplanting root growth state image analysis method and system. BACKGROUND

[0002] In the application scenario of seedling transplanting, the root system is the basic structure for maintaining water and nutrient absorption of the plant, and the timeliness and stability of its growth recovery directly determine the adaptability and survival rate of the seedling in the new environment. Especially in the early stage of transplanting, the root system is affected by multiple factors such as environmental stress, soil disturbance and physical damage, and the recovery process is extremely slow, and the external morphology is difficult to present early change signals. The aboveground performance may not be consistent with the actual state of the root system, causing the traditional management method relying on visual observation to fail to identify potential root problems in a timely manner, thereby affecting the scientificity and response timeliness of the maintenance decision.

[0003] At present, the identification of the root state after seedling transplanting mainly relies on manual experience or end-point survival statistics, and lacks continuous, quantitative and objective analysis methods. Especially in the early stage of root growth, the root system is still in the shallow layer, the structure is sparse, and the boundary is not clear. The existing image technology cannot extract effective features, and cannot accurately judge the root tip activity, capillary root density and root distribution trend, which makes it impossible to build a dynamic evolution model of the root growth trend. The above problems limit the early detection and early intervention ability of the root abnormal state in the seedling maintenance process, which is not conducive to improving the overall transplanting effect. Therefore, it is urgent to build an intelligent identification method that can non-destructively perceive the early state of the seedling root system in an image manner and realize qualitative and quantitative analysis. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the present application provides a seedling transplanting root growth state image analysis method and system, which constructs a multi-frame image time sequence acquisition system based on shallow buried near-infrared imaging, combines the fusion analysis of root tip active change rate and root increment density gradient, generates a recovery trend index and realizes intelligent judgment of root zone activity level, to solve the problems of root state recognition lag, strong subjectivity of judgment and lack of dynamic quantitative support in the background technology.

[0005] To achieve the above purpose, the present application provides the following technical scheme: a seedling transplanting root growth state image analysis method, comprising:

[0006] Step S1: setting a shallow buried optical detection component in the target root zone after seedling transplanting, and acquiring multiple frames of near-infrared images in a continuous time period by using time-sharing control and establishing an original image frame;

[0007] Step S2: performing structure equalization preprocessing operation on each frame image in the original image frames, constructing preprocessed image frames through reflection distribution smoothing and boundary softening;

[0008] Step S3: taking the preprocessed image frames as input, extracting the migration vector field of the root zone edge and the region increment feature between the continuous image frames, forming the root system growth dynamic parameter sequence based on the time axis, and the region increment feature at least includes the root frame difference graph, the gray increment, the density increment graph and the local density direction change rate graph;

[0009] Step S4: respectively calculating the root tip active change rate AAR and the root system increment density gradient IDG, the root tip active change rate AAR is obtained based on the time differential result of the root tip normal change vector, and the root system increment density gradient is obtained by extracting the density increment and gradient direction of the root system region in the partition grid; inputting the root tip active change rate and the root system increment density gradient into the preset frame-level mapping function, and outputting the comprehensive growth state index Γm reflecting the root system growth activity degree of the corresponding image frame;

[0010] Step S5: constructing a root zone activity level model based on the root zone growth trend state and the comprehensive growth state index distribution graph, and outputting a root zone structure graph containing growth state level, confidence interval confidence level and response suggestion.

[0011] Preferably, when collecting the original image frames, a spatial index and a collection time label are bound to each image frame based on the identification label of the seedling, forming a traceable double-index image data structure, and the atlas is called based on the index in the subsequent processing process; performing frame-level resampling and registration operation on the collected original image frames, constructing inter-frame alignment image sequence through nonlinear time interpolation and local elastic registration algorithm, so as to improve the consistency and accuracy of subsequent dynamic parameter calculation.

[0012] Preferably, in the construction process of the original image frames, the sampling mode of the image frames is difference resampling operation based on the pixel change intensity between frames, the sampling density of the image frames is increased in the period when the root system region grows rapidly or the structure changes significantly, so as to enhance the time coverage ability of the key dynamic change; for the static region with structure change intensity lower than the preset threshold between the continuous frames, the sampling frequency is reduced to avoid redundant image collection, so as to ensure the complete expression of the key growth information, and improve the processing efficiency of the image sequence and the response focusing ability of the subsequent model.

[0013] Preferably, the original image frame preprocessing process comprises:

[0014] Frame-by-frame scheduling and pixel normalization are performed on the original image frame set to construct preprocessed images in a unified brightness range;

[0015] Performing reflection feature enhancement and bidirectional sliding average brightness smoothing processing to generate a reflection feature enhanced image;

[0016] An edge enhancement operator based on structural gradient direction guidance is used to extract root zone edge features to generate an initial root zone target contour map;

[0017] According to the spatial structure offset degree ΔRg between adjacent frames, a continuous structure invariant region is dynamically determined as a false recognition signal, and mask removal and cleaning are performed, and a preprocessed image frame is output.

[0018] Preferably, the extraction of the root system growth dynamic parameter sequence comprises:

[0019] Based on the structural enhancement atlas package, frame-level structure images and corresponding initial root zone target contour maps RGm are extracted from the target frame set in time sequence;

[0020] Taking any two adjacent frames in the inter-frame structure alignment image sequence as a processing unit, edge point sets are extracted for the corresponding initial root zone target contour map, and edge corresponding point pair sets are generated by minimum distance addition matching, vector migration values between all point pairs are calculated, and a migration vector field Vm of the root zone edge is constructed; a migration vector field set on the time axis is constructed for all sequence frames;

[0021] Adjacent frames are subtracted to form a frame difference map, the gray level increment of the two within the initial root zone target contour map RGm mask range is calculated, and the original grid structure is partitioned, the pixel increment density in the unit area of each sub-region is calculated to form a density increment map; at the same time, gradient direction field extraction is performed in the density increment map, a local density direction change rate map is constructed, and a root system increment density gradient IDGm is generated.

[0022] Preferably, the comprehensive growth state index acquisition method is:

[0023] According to the root tip active change rate and the root system increment density gradient output frame by frame from the target frame set, a root system growth dynamic parameter sequence is reorganized on the time axis, and a frame-level mapping function Γm=f(AAR,IDGm) is constructed to generate a comprehensive growth state index Γm;

[0024] In the calculation structure of the comprehensive growth state index Γm, the following combined expression is used: ;

[0025] Wherein, the term ln(1+α·AAR) improves the response sensitivity to weak changes through logarithmic transformation, and (1+β·IDG) retains the weighted influence of density growth on the formation of the overall structure; α is used to control the sensitivity to the root tip active change rate AAR fluctuation, and β controls the threshold of the root system increment density gradient.

[0026] Preferably, the α and β coefficients are obtained by:

[0027] At the early stage of transplanting, the root zone image sequence is selected from several representative samples, and the corresponding root tip active change rate AAR and root increment density gradient are extracted for each frame image;

[0028] Grid traversal is performed on the values of a and β under different combinations, the comprehensive growth state index Γm sequence corresponding to each combination is calculated, and the change trend thereof is correlated with the subsequent real survival result at the frame level;

[0029] According to the survival rate data, a pair of labels is constructed, the fluctuation characteristics of the comprehensive growth state index Γm sequence and the growth outcome are mapped into a scoring model, and the optimal parameter pair is determined by minimizing the prediction error;

[0030] Using the optimal parameter pair as the initial setting, combined with the sliding time window statistical features constructed by the sampling data under different ecological conditions or soil disturbance, Bayesian optimization or genetic algorithm fine tuning is performed to obtain a robust adaptive parameter adjustment mechanism;

[0031] The finally fitted parameter pair is embedded into the mapping frame-level mapping function and deployed in the online image analysis system to generate the comprehensive growth state index.

[0032] Preferably, the current growth trend state of the root zone is labeled and classified based on the comprehensive growth state index Γm, and the specific operation is as follows:

[0033] After obtaining the comprehensive growth state index Γm, trend state identification processing is performed according to its distribution in the time sequence; based on the preset trend state determination interval, the value range of the comprehensive growth state index Γm is divided into several level sections, and each level corresponds to a root zone growth trend state;

[0034] In the execution process, the root zone growth trend state to which the comprehensive growth state index Γm belongs is determined according to the change slope and interval landing point of the comprehensive growth state index Γm in the continuous frames, and the root zone growth trend state sequence is recorded in time sequence;

[0035] The generated root zone growth trend state sequence is subjected to time sequence continuity analysis, and if there is frequent switching or local fluctuation instability of the root zone growth trend state, the adjacent frame comprehensive growth state index Γm value and boundary fluctuation are corrected to make the output root zone growth trend state have time consistency and structural rationality.

[0036] Preferably, on the basis of level boundary division, the comprehensive growth state index Γm fluctuation range and slope variance of each trend label within a specified time window are counted to form a credibility index vector; according to the credibility index vector, each frame-level label is given a confidence interval confidence level and labeled in the label sequence as a stability basis for subsequent response suggestion judgment.

[0037] To achieve the object of the present application, a seedling root growth state image analysis system after transplanting is provided, comprising:

[0038] An image acquisition module: a shallow-buried optical detection assembly is arranged in the target root area after the completion of seedling transplanting, near-infrared images are continuously acquired through time-sharing control mode, and a plurality of original image frame sequences are generated;

[0039] An image preprocessing module: taking the original image frame sequence as input, structure equalization processing is performed on each frame of image, including smoothing of reflection distribution and softening of root area edge, so that the separation clarity between the root area and the background is improved, and a high-recognizability preprocessed image frame is constructed;

[0040] A dynamic modeling module: taking the preprocessed image frame as input, the migration vector field of the root area edge between adjacent frames is analyzed, and the area increment feature is extracted, the area increment feature includes a root frame difference graph, a gray increment, a density increment graph and a local density direction change rate graph, and a root growth dynamic parameter sequence is constructed;

[0041] A growth state generation module: taking the area increment feature as input, the root tip active change rate AAR and the root increment density gradient IDG are calculated, the root tip active change rate AAR and the root increment density gradient IDG are input into a preset frame-level mapping function, and a comprehensive growth state index Γ m is output corresponding to each frame of image;

[0042] A grade analysis module: a root area activity grade model is constructed based on the root area growth trend state and the comprehensive growth state index distribution graph, and a root area structure atlas containing the growth state grade, the confidence interval confidence grade and the response suggestion is output.

[0043] The technical effects and advantages of the present application are as follows:

[0044] (1) The present application realizes the spatial consistency modeling and time comparability reconstruction of the root area structure dynamic behavior in the early stage of seedling transplanting through the shallow-buried multi-frame near-infrared imaging device and the image grid resampling mechanism, effectively solves the problems of lack of continuous observation, large spatial positioning error and image processing results that cannot be uniformly numbered and traced in the root recovery process of the traditional method. At the same time, with the help of the structure-enhanced atlas and the root area mask, the suppression of the soil interference area and the pseudo-edge signal is completed, so that the subsequent dynamic migration analysis has higher robustness and boundary fidelity; in the root area dynamic change parameter extraction stage, the root tip active change rate and the root increment density gradient are generated in combination through the explicit clustering of the vector field and the root tip active area, so that the dynamic change has structure interpretation and morphological symmetry basis, and a reliable data basis is established for trend fusion analysis.

[0045] (2) The application fuses the normalized root tip active change rate and the root increment density gradient through self-adaptation, constructs a root area structure map containing a growth state level, a confidence interval confidence level and a response suggestion, provides a qualitative judgment basis and distributed response support for a decision system, and improves the decision-making ability and application landing nature of the image analysis system. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 A root system growth state image analysis method flowchart after seedling transplanting of the application. DETAILED DESCRIPTION

[0047] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art.

[0048] At the same time, it should be understood that the sizes of the various parts shown in the drawings are not drawn in accordance with the actual proportional relationship for the sake of description.

[0049] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way limiting of the application or its use.

[0050] Techniques, methods, and devices known to those of ordinary skill in the relevant art can not be discussed in detail herein, but should be considered as part of the specification, where appropriate.

[0051] Reference Figure 1 The root system growth state image analysis method flowchart after seedling transplanting of the application provides a root system growth state image analysis method as shown in Figure 1 The root system growth state image analysis method flowchart after seedling transplanting of the application provides a root system growth state image analysis method as shown in

[0052] Step S1: A shallow-buried optical detection assembly is set in a target root area after seedling transplanting is completed, a plurality of near-infrared images in a continuous time period are collected by using a time-sharing control mode, and an original image frame is established to maintain the spatial distribution consistency and time variation comparability of the root area image;

[0053] Step S2: Structure equalization preprocessing operation is performed on each image in the original image frame, the separability of the root system area is enhanced through reflection distribution smoothing and boundary softening, and a high-recognizability preprocessing image frame is constructed;

[0054] Step S3: Taking the pre-processed image frames as input, extract the root zone edge migration vector field and the region increment feature between consecutive image frames, form the root system growth dynamic parameter sequence based on the time axis, and the region increment feature at least includes root frame difference map, gray increment, density increment map and local density direction change rate map;

[0055] Step S4: Calculate the root tip active change rate AAR and the root system increment density gradient IDG respectively, the root tip active change rate AAR is obtained based on the time differential result of the root tip normal change vector, and the root system increment density gradient is obtained by extracting the density increment and gradient direction of the root system region in the partition grid; input the root tip active change rate and the root system increment density gradient into the preset frame-level mapping function, and output the comprehensive growth state index Γm reflecting the root system growth activity degree of the corresponding image frame;

[0056] Step S5: Based on the root zone growth trend state and the comprehensive growth state index distribution map, a root zone activity level model is constructed, and a root zone structure map containing growth state level, confidence interval confidence level and response suggestion is output, which provides qualitative judgment basis and distributed response support for the decision system.

[0057] Further, the step S1: root zone image time sequence data acquisition mechanism construction, comprising:

[0058] Substep 101: detection component layout and collection area positioning

[0059] The near-infrared imaging detection component is shallowly buried in the soil layer above the root zone after the seedling transplanting is completed, the detection component includes a multispectral imaging head, a time-sharing exposure control unit and a time sequence label encoding module, the imaging area boundary coordinates are set according to the seedling root zone physiological characteristics and the initial planting parameters of the plant, the corresponding root zone space identification label Map is generated, and a one-to-one mapping relationship with the current seedling number is established, so as to ensure the consistency of the spatial distribution of the collected images and the uniqueness of the identification; the identification label Map will be used for index calling in the subsequent image matrix and index mapping;

[0060] Substep 102: multi-frame image acquisition and time label encoding

[0061] After the identification label Map is generated and the area boundary binding is completed, the time-sharing exposure control mechanism of the detection component is started, the near-infrared image sequence containing continuous time sequence is collected according to the preset imaging interval, each image frame is bound with the current collection time stamp Tk, and the space is classified according to the identification label Map, forming an image data set Gseq={Gk|Map, Tk} with double index structure, wherein Gk represents the image frame of the root zone at time k;

[0062] Explanatory note: the image frame Gk enhances the reflection difference between the root system tissue and the surrounding soil background in the near-infrared band, which helps to enhance the root zone boundary and extract the structure in the subsequent processing; the structure will become the basic data input of the entire image analysis process;

[0063] Sub-step 103: image rasterization and time sequence unified resampling:

[0064] In order to improve the spatial alignment accuracy of image analysis, the image frame Gk is divided into blocks and rasterized according to the spatial grid to generate a standardized image matrix GGk={Gij^k}, where Gij^k is the pixel block of the (i, j) grid unit in the k time image; then the standardized image matrix GGk is uniformly resampled on the time sequence axis, and the non-uniform sampling frame is uniformly mapped to the target frame set Tseq={T1, T2, …, Tm} (representing the image data set corresponding to time 1, 2, …, …, m) by using spline interpolation and inter-frame intensity matching strategy, and an aligned image set GGalign={GGm} is generated.

[0065] Explanatory note: this processing process is not a traditional uniform down-sampling, but a differential resampling based on the sampling intensity change rate, so that more intensive frame coverage is obtained at the time of rapid growth or significant change of the root system, thereby retaining key dynamic change information, while avoiding redundant sampling in static and unchanged areas, improving data processing efficiency and model attention.

[0066] Sub-step 104: unified image set storage and atlas generation interface establishment

[0067] The aligned image set GGalign and the corresponding identification label Map and the target frame set Tseq are input into the root zone atlas cache system as a unified data package, and an atlas call index table Index_Table={Map, Tseq, GGm} is constructed for subsequent step S2 phase call execution structure balanced preprocessing.

[0068] The overall technical synergy effect of this step S1 is explained: this step realizes the unified collection and labeled archiving of seedling root zone images in space-time dimensions, and through the overall process of numbered identification-time sequence archiving-raster resampling-function index, it realizes the transformation of image data structure to a standardized expression method of "controllable, extensible and reconfigurable"; provides a highly organized collection platform for the entire root image analysis process.

[0069] Further, the step S2: image structure balanced preprocessing and root target separability enhancement mechanism construction, comprising:

[0070] Sub-step 201: image frame scheduling and structure initialization processing

[0071] With the image set GGalign completed in step S1 as input, the single seedling is scheduled according to the identification label Map, and the calling function is executed frame by frame according to the target frame set Tseq, and the image frame GGm corresponding to each time Tm is extracted as the current processing target; at the same time, the pixel value normalization and domain balance processing are performed on the image frame GGm, and the rasterized preprocessed image GGeqm in the unified brightness range is constructed for subsequent reflection feature processing;

[0072] Substep 202: Local reflection feature enhancement and regional brightness smoothing processing

[0073] The regional reflection fitting enhancement operation is performed on the preprocessed image GGeqm, and in each frame, the overexposure or overdark value of each unit edge is adjusted; the brightness smoothing is performed based on the bidirectional neighborhood sliding average filter to generate a reflection feature enhanced image, ensuring the stability of the root zone texture features under the background fluctuation condition;

[0074] Substep 203: Boundary-oriented edge enhancement and coarse target extraction

[0075] On the basis of the reflection feature enhanced image, the edge enhancement operator (DEGF) based on the structure gradient direction guidance is used to extract the root zone target edge feature Em, and the initial root zone target contour map RGm is generated based on the edge closure degree threshold to construct a frame-level coarse recognition root zone region mask; the initial root zone target contour map is nested into the original image to generate a root zone saliency map, realizing the integrated processing of structure strengthening and boundary imaging;

[0076] It is explained that this step realizes the direction perception and morphological integrity constraint of boundary extraction by gradient direction guidance instead of intensity threshold segmentation, effectively avoiding the breaking and misjudgment phenomenon in the root hair structure under the traditional Canny or Sobel operator; the initial root zone target contour map RGm generated will be directly used as the mask basis for dynamic change parameter extraction in step S3;

[0077] Substep 204: Anti-pseudo signal detection and saliency consistency screening

[0078] The structure consistency verification is performed on the root zone saliency map, and the spatial structure offset degree ΔRg of the root zone mask change in adjacent images is calculated using the interframe root system region morphological difference statistical parameter ΔRg=||RGm−RG{m−1}||;

[0079] If the spatial structure offset degree ΔRg is lower than the dynamic change threshold for a plurality of continuous time points, it is determined that the corresponding region is a pseudo-identification signal, and the mask is removed from the image and updated to the preprocessed image frame;

[0080] Explanatory note: through the inter-frame structure variation threshold dynamic identification mechanism, effectively avoid the false root signal interference caused by water droplets, reflection, local shadow, improve the accuracy and continuity of subsequent dynamic parameter extraction; the ΔRg index will be provided to the S3 stage dynamic migration vector calculation as one of the morphological drift reference parameters;

[0081] Sub-step 205: structure atlas storage and processing state registration

[0082] All pre-processed image frames and corresponding identification labels Map, target frame set Tseq, initial root zone target contour graph RGm, root zone target edge feature Em, spatial structure offset degree ΔRg are packaged to generate structure enhanced atlas package, and are uniformly registered into structure atlas index, forming a pre-processing atlas scheduling interface; the subsequent steps can extract the root zone image and mask information processed by the scheduling interface according to the time sequence, as the direct input of dynamic parameter sequence analysis.

[0083] This step technical synergy effect note: step S2 realizes the complete transition from spatial organization (S1 output) to image separability enhancement, which on the one hand retains the high stability structure of root zone boundary details, and on the other hand eliminates the dynamic interference background signal, finally provides the data with stronger morphological consistency and texture continuity to the S3 stage. Especially in the multi-layer nested process of image brightness normalization + local reflection change rate adjustment + structure guided edge extraction + cross-frame consistency verification, the technical features of each step have a high coupling effect, and are not independent superposition.

[0084] Further, the step S3 includes the following contents:

[0085] Sub-step 301: frame image deconstruction and root zone structure synchronous alignment processing

[0086] Through the calling interface, the structure enhanced atlas package corresponding to the target seedling is obtained, and the frame-level structure image and the corresponding initial root zone target contour graph RGm are extracted from the target frame set according to the time sequence; in order to ensure the spatial consistency of each frame image in the subsequent morphological comparison, pixel-level registration processing is performed, an image alignment algorithm based on local rigid constraint and global elastic fitting is adopted, and a frame structure alignment image sequence {Gĝ_m} is generated, ensuring that the root zone structure maintains the consistency of position and scale on the time axis;

[0087] Sub-step 302: root zone edge migration vector field extraction and time sequence coding construction

[0088] Using any two adjacent frames (Gĝ_{m}, Gĝ_{m+1}) in the inter-frame structure-aligned image sequence as processing units, for the corresponding initial root region target contour maps RGm and RG{m+1}, edge point sets {P_i^m} and {P_j^{m+1}} are extracted, and the set of corresponding edge point pairs C_ij is generated by minimum distance addition matching; the vector migration value Vij=P_j^{m+1}−P_i^m between all point pairs is calculated to construct the migration vector field Vm of the root region edge; and the set of migration vector fields Vseq={Vm} on the time axis is constructed for the entire sequence of frames.

[0089] Explanatory notes: The vector migration field not only includes positional differences, but also implies dynamic information such as root tip growth direction and edge expansion trend; the migration vector field set Vseq will serve as the core supporting parameter for calculating the root tip activity change rate AAR, and has high interpretability and physical mapping capability;

[0090] Sub-step 303: Calculation of regional incremental features

[0091] The regional incremental features include at least the root frame difference map, gray-level increment, density increase map, and local density direction change rate map. The frame difference map ΔGGm is formed by subtracting image frame Gĝ_{m+1} from image frame Gĝ_{m}. The gray-level increment ΔI_ij between the two within the initial root region target contour map RGm mask range is calculated. Following the original grid structure partitioning, the pixel increment density per unit area is calculated for each sub-region to form the density increase map ρ_m. Simultaneously, gradient direction field extraction is performed on the density increase map ρ_m to construct the local density direction change rate map φ_m. Combining ρ_m and φ_m generates the root system incremental density gradient IDGm, which is used for subsequent comprehensive trend determination.

[0092] Explanatory note: The region incremental feature is a necessary input for step S4. Its calculation depends on the pixel-level changes and spatial distribution patterns between frames in the current step. Therefore, it needs to be executed under the combined effect of the grid structure and the edge mask to ensure the consistency of region division.

[0093] Furthermore, step S4 includes:

[0094] Sub-step 401: Determination of active root tip region and generation of initial value of root tip activity change rate

[0095] Based on the clustering characteristics of vector amplitude and direction in the migration vector field Vm, a spatial density weighting method is used to perform clustering analysis on edge point groups to identify regions in continuous frames where the vector exceeds the threshold and the direction is stable, and these regions are marked as root tip active candidate regions. For each root tip active candidate region, the mean vector amplitude and the rate of change of normal angle are calculated to form the root tip active change rate AAR, and its position index Tk in the time axis is recorded.

[0096] Sub-step 402: comprehensive growth state indicator generation and structure interface construction

[0097] According to the root tip active change rate AAR and the root increment density gradient IDGm output frame by frame according to the target frame set Tseq, a root system growth dynamic parameter sequence is reorganized on a time axis, a frame-level mapping function Γm=f(AAR, IDGm) is constructed, a comprehensive growth state indicator Γm is generated, a structured root system update index atlas U_map={Γ1, Γ2,..., Γm} is obtained by summarizing, and an atlas index is embedded for subsequent step S5 calling;

[0098] The frame-level mapping function refers to taking the root tip active change rate AAR and the root increment density gradient IDGm corresponding to each frame image as input, calculating the comprehensive growth state indicator Γm for reflecting the active degree of root system growth in the frame through a preset function model f; the frame-level mapping function not only reflects the dynamic change of the root tip activity, but also comprehensively considers the spatiotemporal distribution characteristics of the root system structure density, so that each frame has growth state expression capability that can be quantitatively analyzed, facilitating subsequent trend calculation use;

[0099] The structured root system update index atlas refers to an ordered set U_map composed of a series of Γm calculated by the frame-level mapping function, which is sequentially arranged according to the time axis of the target frame set Tseq, and records the comprehensive growth state indicators of the root zone in each frame within the entire target observation period; the atlas not only retains the time sequence index information of the original image frame, but also embeds the dynamic characteristic parameters of the root zone, so that it becomes a high-level structured data representation form supporting trend evolution modeling, regional level determination and subsequent response analysis.

[0100] Further supplement, in order to realize the migration analysis ability under the conditions of various environments and species, the present application adopts the following combined expression in the calculation structure of the comprehensive growth state indicator Γm:

[0101] The term ln(1+α·AAR) improves the response sensitivity to weak changes through logarithmic transformation, and is especially suitable for the case where the initial fluctuation of the seedling root tip growth signal is not obvious, while (1+β·IDG) retains the weighted influence of density growth on the overall structure formation; the product of the two constitutes a nonlinear coupling mechanism, which makes the model have higher discrimination and trend amplification ability when dealing with strong and weak root system regions;

[0102] The coefficients α and β are not fixed constants, but can be adaptively set according to the application object, the root zone imaging quality and the seedling type; α is usually used to regulate the sensitivity to AAR fluctuation, and β controls the threshold of structure response, and both of them can be subjected to sensitivity regression in a specified sliding time window to improve the stability and generalization ability of the indicator calculation.

[0103] In practical applications, to obtain α, β coefficients suitable for specific seedling species and planting conditions, the present application performs modeling training and fitting through the following parameter optimization process:

[0104] In the early stage of transplanting, root zone image sequences are selected from several representative samples, and the corresponding root tip active change rate AAR and root increment density gradient are extracted for each frame of image;

[0105] Grid traversal is performed on α values and β values under different combinations, the comprehensive growth state index Γm sequence corresponding to each combination is calculated, and the change trend thereof is evaluated in correlation with the subsequent true survival result at the frame level;

[0106] According to the survival rate data, a pair of labels is constructed, the fluctuation characteristics of the comprehensive growth state index Γm sequence and the growth outcome are mapped into a scoring model, and the optimal parameter pair (α*, β*) is determined by minimizing the prediction error;

[0107] Explanatory note, the scoring model refers to a mathematical function or statistical learning framework constructed based on historical annotation data, which is used to measure the consistency or prediction ability between the comprehensive growth state index Γm generated by different parameter combinations and the actual physiological performance (such as survival state, root zone recovery level, etc.); The scoring model takes the Γm features (such as mean, variation rate, trend slope, etc.) extracted from the sample frame sequence as input variables, and takes the actually annotated growth performance as output root zone growth trend state, and obtains the fitting function through regression or classification algorithm (such as logistic regression, support vector machine or gradient boosting tree); The evaluation index (such as prediction accuracy, mean square error or F1 score) output by the scoring function is used to quantitatively score the Γm calculation effect under different parameter settings, and the parameter combination with the optimal evaluation index is finally selected as the model optimal solution, so as to realize the mapping closed loop between the root index generation mechanism and the physiological performance, and ensure that the calculation result has biological relevance and actual guiding value;

[0108] Using the optimal parameter pair as the initial setting, combined with the sliding time window statistical features constructed from the sampling data under different ecological conditions or soil disturbance, Bayesian optimization or genetic algorithm fine tuning is performed to obtain a robust adaptive parameter adjustment mechanism;

[0109] The finally fitted parameter pair is embedded into the mapping frame-level mapping function and deployed in the online image analysis system to generate the comprehensive growth state index.

[0110] Step S4 summarizes the technical effects: This step revolves around the changes in image time sequence frame structure. From spatial registration, edge matching, dynamic vector extraction, active area modeling, incremental density direction calculation to the final generation of structured update index atlas U_map, it completes the technical leap from structural image to dynamic behavior mapping. Unlike traditional frame difference analysis based only on gray scale changes, this scheme introduces edge migration and spatial directionality analysis mechanism, making the root system change analysis have stronger interpretability and morphological stability, especially suitable for complex structure and discontinuous change of early transplanting root zone scene.

[0111] Further, the step S5: based on the comprehensive growth state index Γm, the current growth trend state of the root zone is labeled and classified, and the specific operation is as follows:

[0112] After obtaining the comprehensive growth state index Γm, the trend state is identified according to its distribution in the time sequence; this processing is based on the preset trend state determination interval, which divides the value range of the comprehensive growth state index Γm into several level sections, and each level corresponds to a root zone growth trend state;

[0113] In the execution process, according to the change slope and interval landing point of the comprehensive growth state index Γm in the continuous frame, the root zone growth trend state to which it belongs is determined, and the root zone growth trend state sequence is recorded in time sequence;

[0114] The generated root zone growth trend state sequence is analyzed for time sequence continuity. If there is frequent switching or local fluctuation instability of the root zone growth trend state, further correction is made according to the comprehensive growth state index Γm value and boundary fluctuation of adjacent frames, so that the output root zone growth trend state has time consistency and structural rationality. The above-mentioned root zone growth trend state sequence is used to represent the growth trend type of the root zone in the current monitoring period, and provides a basis for subsequent growth level division and maintenance response.

[0115] Further, the step S5: root activity level model construction and structure atlas output

[0116] In order to realize the level division of root zone growth trend state and the output of regional response suggestion, this step is based on the comprehensive growth state index sequence Γm and its trend label results generated in step S4, combined with time evolution stability analysis, to construct a composite root zone structure atlas containing level determination, reliable interval extraction and response suggestion generation; Specifically includes:

[0117] Sub-step 501: trend level generation and level boundary demarcation

[0118] According to the full sequence distribution of the comprehensive growth state index Γm, an interval division-based grade model is constructed; the value of the comprehensive growth state index Γm is divided into multiple continuous grade sections according to the statistical percentile method, each grade corresponds to a growth state grade mark, and is bound to the corresponding trend evolution direction (enhancement, attenuation or stability); in order to ensure the judgment stability of the boundary interval, the boundary is corrected in combination with the change slope of the comprehensive growth state index sequence Γm at the trend label switching position, and the final boundary interval of each grade is obtained;

[0119] Substep 502: Confidence interval extraction and confidence weighting

[0120] On the basis of grade boundary division, the fluctuation range and slope variance of the comprehensive growth state index Γm of each trend label in a specified time window are counted to form a confidence index vector; the confidence interval confidence level of each frame-level label is given according to the confidence index vector, and is marked in the label sequence as a stability basis for subsequent response suggestion judgment;

[0121] Substep 503: Structure atlas space mapping and response suggestion generation

[0122] In combination with the spatial grid division structure of the original image frame, the trend label with marked grade and confidence is mapped to the sub-region corresponding to the root area image, and the average value of the comprehensive growth state index sequence Γm, the density change amplitude and other auxiliary features in the region are fused to comprehensively judge the growth activity grade; according to a preset response rule library, corresponding response suggestion labels such as "observation", "priority intervention" or "no need to process" are generated for each sub-region, and are packaged into a root area structure atlas containing four fields of position coordinates, grade value, confidence and response suggestion.

[0123] To achieve the above purpose, the present application provides the following technical scheme: a seedling transplanting root system growth state image analysis system, comprising:

[0124] Image acquisition module: a shallow-buried optical detection component is arranged in the target root area after seedling transplanting is completed, and near-infrared images are continuously acquired through time-sharing control to generate a sequence of multiple original image frames; the acquisition process ensures the consistency of the images in spatial distribution and the comparability of the images on the time axis; the output result is a sequence of original image frames arranged in chronological order, with time labels and spatial indexes attached, to provide a basic data source for subsequent image processing and dynamic change recognition;

[0125] Image preprocessing module: taking the sequence of original image frames as input, structure balancing is performed on each image, including smoothing of the reflection distribution and softening of the root area edge, to improve the separation clarity between the root system area and the background, and to construct a preprocessed image frame with high distinguishability;

[0126] The dynamic modeling module takes the preprocessed image frames as input, analyzes the migration vector field of the root zone edge between adjacent frames, and extracts the region increment feature, including the root frame difference graph, the gray increment, the density increment graph, and the local density direction change rate graph, to construct a root growth dynamic parameter sequence;

[0127] The growth state generation module takes the region increment feature as input, calculates the root tip active change rate AAR and the root increment density gradient IDG, inputs the root tip active change rate AAR and the root increment density gradient IDG into a preset frame-level mapping function, and outputs the comprehensive growth state index Γ corresponding to each frame of image m ;

[0128] The grade analysis module constructs a root zone activity grade model based on the root zone growth trend state and the comprehensive growth state index distribution graph, outputs a root zone structure graph containing the growth state grade, the confidence interval confidence grade, and the response suggestion, and provides qualitative judgment basis and distributed response support for the decision system.

[0129] Finally, the above-mentioned is only the preferred embodiment of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for image analysis of the growth state of the root system of a seedling after transplanting, characterized by, Comprise: Step S1: set a shallow buried optical detection component in the target root zone after the seedling transplanting is completed, adopt a time-sharing control mode to collect multiple frames of near-infrared images in a continuous time period and establish original image frames; Step S2: perform a structure equalization preprocessing operation on each frame of image in the original image frames, and construct a preprocessed image frame through reflection distribution smoothing and boundary softening; Step S3: take the preprocessed image frame as input, extract the migration vector field of the root zone edge between the continuous image frames and the regional increment feature, form a root system growth dynamic parameter sequence based on the time axis, and the regional increment feature at least includes a root frame difference graph, a gray increment, a density increment graph and a local density direction change rate graph; Step S4: calculate the root tip active change rate AAR and the root system increment density gradient IDG respectively, the root tip active change rate AAR is obtained based on the time differential result of the root tip normal change vector, and the root system increment density gradient is obtained by extracting the density increment and gradient direction of the root system region in the partition grid; input the root tip active change rate and the root system increment density gradient into a preset frame-level mapping function, and output a comprehensive growth state index Γm reflecting the growth activity degree of the corresponding image frame root system; Step S5: construct a root zone activity level model based on the root zone growth trend state and the comprehensive growth state index distribution graph, and output a root zone structure atlas containing a growth state level, a confidence interval confidence level and a response suggestion.

2. The method of claim 1, wherein the method is characterized by: When collecting the original image frames, a spatial index and a collection time label are bound to each image frame based on the identification label of the seedling, a traceable double-index image data structure is formed, and atlas calling is performed based on the index in the subsequent processing process; frame-level resampling and registration operations are performed on the collected original image frames, a frame alignment image sequence is constructed through a nonlinear time interpolation and a local elastic registration algorithm, so that the consistency and accuracy of subsequent dynamic parameter calculation are improved.

3. The method of claim 2, wherein the method is characterized by: In the construction process of the original image frames, the sampling mode of the image frames is a difference resampling operation based on the pixel change intensity between frames, the sampling density of the image frames is increased in the time period when the root system region grows rapidly or the structure changes significantly, so as to enhance the time coverage ability of the key dynamic change; for the static region with a structure change intensity lower than a preset threshold between continuous frames, the sampling frequency is reduced to avoid redundant image collection, so that the key growth information is expressed completely, and the processing efficiency of the image sequence and the response focusing ability of the subsequent model are improved.

4. The method of claim 3, wherein the method is characterized by, The original image frame preprocessing process comprises: frame-by-frame scheduling and pixel normalization are performed on the original image frame set to construct a preprocessed image under a unified brightness range; reflection feature enhancement and bidirectional sliding average brightness smoothing processing are performed to generate a reflection feature enhanced image; a root zone edge feature is extracted based on a structure gradient direction guided edge enhancement operator to generate an initial root zone target contour graph; a continuous structure invariant region is dynamically distinguished as a false recognition signal according to a spatial structure offset degree ΔRg between adjacent frames, and mask removal and cleaning are performed to output a preprocessed image frame.

5. The method of claim 4, wherein the method is characterized by: The extraction of the root system growth dynamic parameter sequence comprises: Based on the structure-enhanced atlas package, frame-level structure images and corresponding initial root zone target contour maps RGm are extracted from the target frame set in time sequence; Taking any two adjacent frames in the inter-frame structure aligned image sequence as a processing unit, edge point sets are extracted for the corresponding initial root zone target contour map, and edge corresponding point pair sets are generated by minimum distance addition matching. Vector migration values between all point pairs are calculated to construct the migration vector field Vm of the root zone edge. Migration vector field sets on the time axis are constructed for all sequence frames; Frame difference images are formed by subtracting adjacent frames. The gray level increment of both within the initial root zone target contour map RGm mask range is calculated, and the pixel increment density per unit area of each sub-region is calculated according to the original grid structure partitioning to form a density increment map. Meanwhile, the gradient direction field is extracted in the density increment map to construct a local density direction change rate map, and a root increment density gradient IDG is generated.

6. The image analysis method of the root growth state of a seedling after transplanting according to claim 1, characterized by, The comprehensive growth state index acquisition method is: According to the root tip active change rate and the root increment density gradient output frame by frame from the target frame set, a root growth dynamic parameter sequence is reorganized on the time axis, and a frame-level mapping function Γm=f(AAR,IDG) is constructed to generate a comprehensive growth state index Γm. In the calculation structure of the integrated growth status indicator Γm the following combined expression is employed: ; Wherein, the ln(1+α·AAR) term improves the response sensitivity to weak changes through logarithmic transformation, and (1+β·IDG) retains the weighted influence of density growth on the formation of the overall structure; α is used to control the sensitivity to the root tip active change rate AAR fluctuation, and β controls the threshold of the root increment density gradient.

7. The method of claim 6, wherein the method is characterized by: The α and β coefficients are obtained as follows: In the early stage of transplanting, root zone image sequences are selected from several representative samples, and the corresponding root tip active change rate AAR and root increment density gradient are extracted for each frame image. Grid traversal is performed on the α and β values under different combinations, the comprehensive growth state index Γm sequence corresponding to each combination is calculated, and the change trend is evaluated on the frame level. The correlation between the change trend and the subsequent true survival result is evaluated. According to the survival rate data, a pair of labels is constructed, the fluctuation characteristics of the comprehensive growth state index Γm sequence and the growth outcome are mapped into a scoring model, and the optimal parameter pair is determined by minimizing the prediction error. Using the optimal parameter pair as the initial setting, combined with the sliding time window statistical features constructed from the sampling data under different ecological conditions or soil disturbance, Bayesian optimization or genetic algorithm fine tuning is performed to obtain a robust adaptive parameter adjustment mechanism. The parameters obtained by the final fitting are embedded into the mapping frame-level mapping function and deployed in the online image analysis system to generate the comprehensive growth state index.

8. The method of claim 6, wherein the method is characterized by: Based on the comprehensive growth state index Γm, the current growth trend state of the root zone is classified as follows: After obtaining the comprehensive growth state index Γm, the trend state is identified according to its distribution in the time sequence. Based on the preset trend state determination interval, the value range of the comprehensive growth state index Γm is divided into several level sections, and each level corresponds to a root zone growth trend state. In the execution process, the growth trend state of the root zone to which the comprehensive growth state index Γm belongs is determined according to the change slope of the comprehensive growth state index Γm in the continuous frames and the interval landing point, and the growth trend state sequence of the root zone in time sequence is recorded; The time sequence continuity of the generated root zone growth trend state sequence is analyzed, and if there is a situation that the root zone growth trend state frequently switches or locally fluctuates unstably, the comprehensive growth state index Γm value and the boundary fluctuation of adjacent frames are corrected, so that the output root zone growth trend state has time consistency and structural rationality.

9. The method of claim 8, wherein the method is characterized by, On the basis of the grade boundary division, the fluctuation range and slope variance of the comprehensive growth state index Γm of each trend label in the specified time window are counted to form a credibility index vector; according to the credibility index vector, each frame level label is given a confidence interval confidence level, and is marked in the label sequence as a stability basis for subsequent response suggestion judgment.

10. A root growth state image analysis system after transplanting seedlings, characterized by, Comprise: An image acquisition module: a shallow optical detection component is arranged in the target root zone after the seedling transplanting is completed, and near-infrared images are continuously acquired through time-sharing control to generate a sequence of multiple original image frames; An image preprocessing module: taking the sequence of original image frames as input, performing structure balancing processing on each image, including smoothing of reflection distribution and softening of root zone edge, to improve the separation clarity between root area and background, and constructing a high-recognizability preprocessed image frame; A dynamic modeling module: taking the preprocessed image frame as input, analyzing the migration vector field of the root zone edge between adjacent frames, and extracting the area increment feature, the area increment feature including root frame difference graph, gray increment, density increment graph and local density direction change rate graph, and constructing a root growth dynamic parameter sequence; A growth state generation module: taking the area increment feature as input, calculating the root tip active change rate AAR and the root increment density gradient IDG, and inputting the root tip active change rate AAR and the root increment density gradient IDG into a preset frame level mapping function to output the comprehensive growth state index Γm corresponding to each image; A grade analysis module: a root zone activity grade model is constructed based on the root zone growth trend state and the comprehensive growth state index distribution graph, and a root zone structure atlas containing growth state grade, confidence interval confidence level and response suggestion is output.

Images