Vegetable growth monitoring method and system based on flexible sensor

Abstract

The present invention relates to the field of sensor data processing technology, and in particular to a vegetable growth monitoring method and system based on flexible sensors. The method comprises the following steps: collecting sensor resistance values ​​through flexible sensors to obtain sensor resistance data, and calculating deformation variables based on the sensor resistance data to obtain deformation variable data; performing multi-point growth behavior processing on the deformation variable data to obtain multi-point growth behavior data; performing micromorphological growth trend modeling based on the multi-point growth behavior data to obtain a growth trend model; and performing growth trend prediction on the growth trend model to obtain growth trend prediction data. The present invention achieves high-precision dynamic monitoring of structural changes during vegetable growth by constructing a multi-point deformation variable perception and behavior analysis mechanism based on flexible sensors. Combined with regional behavior modeling and trend prediction methods, it can accurately identify abnormal growth trends and provide real-time, interpretable growth assessments for smart agriculture.

Description

Technical Field

[0001] The present invention relates to the technical field of sensor data processing, and in particular to a vegetable growth monitoring method and system based on a flexible sensor. Background Art

[0002] With the development of modern agricultural technology, especially the continuous advancement of facility agriculture and smart planting systems, real-time perception and refined management of crop growth status have become important for improving crop quality and yield. Vegetables, as a crop type extremely sensitive to environmental changes, require more precise and timely monitoring methods to support their growth. Currently, common vegetable growth monitoring methods rely primarily on image analysis and environmental parameter collection. However, these methods are limited in capturing changes in plant structure and responding to subtle growth behaviors, making it difficult to fully understand and predict growth status. Summary of the Invention

[0003] In order to solve the above technical problems, the present invention proposes a vegetable growth monitoring method and system based on flexible sensors to solve at least one of the above technical problems.

[0004] The present application provides a vegetable growth monitoring method based on a flexible sensor, the method comprising:

[0005] S1. Collecting sensor resistance through a flexible sensor to obtain sensor resistance data, and calculating a deformation variable based on the sensor resistance data to obtain deformation variable data;

[0006] S2. performing multi-point growth behavior processing on the deformation variable data to obtain multi-point growth behavior data;

[0007] S3. Modeling the micromorphological growth situation based on the multi-point growth behavior data to obtain a growth situation model;

[0008] S4. Perform growth trend prediction on the growth trend model to obtain growth trend prediction data.

[0009] The present invention introduces flexible sensors to collect tiny deformations on the surface of vegetables, achieving high-sensitivity and high-fit monitoring of deformation characteristics during vegetable growth, thus avoiding interference with normal plant growth caused by traditional rigid sensors. Combined with multi-point growth behavior data processing, the system can obtain the synchronous growth behavior of vegetables in different parts, achieving a more comprehensive perception of growth status. By performing micromorphological situation modeling and prediction on deformation data, not only is the ability to dynamically grasp growth trends improved, but abnormal growth trends such as diseases and stress can also be identified at an early stage, providing a refined decision-making basis for intelligent agricultural management, which is significantly different from traditional methods that rely on image recognition or single-point environmental collection.

[0010] Optionally, S1 includes:

[0011] The flexible sensor arranged at a preset first position data on the vegetable is used to collect the sensor resistance value to obtain the first sensor resistance value data;

[0012] The flexible sensor is arranged at a preset second position data on the vegetable to collect the sensor resistance value, thereby obtaining the second sensor resistance value data, wherein the first position and the second position are different positions;

[0013] Constructing a resistance space map based on the first sensor resistance data and the second sensor resistance data to obtain sensor resistance data;

[0014] The deformation variable data is calculated based on the sensor resistance data to obtain the deformation variable data.

[0015] This invention deploys multiple flexible sensors at different vegetable growth sites (such as stems and leaves) to simultaneously collect multi-source resistance data, effectively improving the spatial resolution of the overall vegetable growth status. By constructing a resistance spatial map, deformation trends at different plant locations can be visualized at a multidimensional level, enhancing the ability to identify local growth differences and synergistic relationships.

[0016] Optionally, the step of generating the first position data and the second position data includes:

[0017] Get vegetable image data;

[0018] Extracting the petiole extension axis according to the vegetable image data to obtain petiole extension axis data;

[0019] Calculate the main stem area center point based on the petiole extension axis data to obtain the main stem area center point data;

[0020] A tension mapping model is constructed according to the center of gravity data of the main stem area to obtain a tension mapping model;

[0021] Position the extreme tension point according to the tension mapping model to obtain the extreme tension point data;

[0022] The position data corresponding to the maximum value data in the tension extreme difference point data is determined as the first position data, and the position data corresponding to the minimum value data in the tension extreme difference point data is determined as the second position data.

[0023] In the present invention, the extension axis of the vegetable petiole is extracted through image recognition technology, which can effectively capture the main morphological direction of the plant in the current growth state; combined with the calculation of the center of gravity of the main stem area, the force center of the entire plant structure can be further located. In constructing the tension mapping model, the strain trend of the petiole to stem area is converted into a spatial tension distribution map. The extreme tension points identified by this model represent the key areas where the tension effect is the strongest (maximum) or weakest (minimum) in local growth, and often correspond to the maximum expansion, maximum contraction or structural transition position of the plant tissue. Placing flexible sensors in these extreme point areas can significantly improve the perception sensitivity and behavioral response representativeness of micro-deformations, avoid redundant monitoring in low dynamic areas, and thus achieve differentiated layout based on growth tension distribution.

[0024] Optionally, the deformation variable data includes single deformation variable data and linkage deformation variable data, and the deformation variable calculation includes:

[0025] According to the sensor resistance data, a resistance change vector field is constructed to obtain resistance change vector field data;

[0026] Perform strain inversion mapping calculation based on resistance change vector field data to obtain strain inversion data;

[0027] Calculate the monomer deformation variable according to the strain inversion data to obtain the monomer deformation variable data;

[0028] The linkage deformation variable data is processed according to the monomer deformation variable data to obtain the linkage deformation variable data.

[0029] This invention constructs a resistance variation vector field to effectively reveal the direction and magnitude of resistance changes over time at different flexible sensor locations. Strain inversion mapping is then used to achieve high-precision conversion from resistance changes to physical strain. This allows for the acquisition of individual deformation data at each sensor location, reflecting the local growth morphology. Through linked deformation processing, the coordinated deformation relationships between different locations are identified, allowing for the construction of a cross-regional growth linkage model.

[0030] Optionally, S2 includes:

[0031] Extracting linkage behavior features from the shape variable data to obtain linkage behavior feature data;

[0032] Divide the micro-behavior segments according to the linkage behavior feature data to obtain micro-behavior segment data;

[0033] A growth behavior map is constructed based on the microscopic behavior fragment data to obtain multi-point growth behavior data.

[0034] By extracting linkage behavior features from deformation data, this method reveals the dynamic coordination patterns between various sensor regions during plant growth, identifying linkage deformation behaviors such as synchronized expansion and opposing torsion. Dividing these linkage features into microscopic behavioral segments with temporal and structural continuity facilitates the extraction of representative behavioral units from local growth data. By constructing a growth behavior map, a structured representation of the association paths, behavioral patterns, and temporal evolution between multi-point deformation variables is achieved.

[0035] Optionally, the linkage behavior feature extraction includes:

[0036] Performing cross-point deformation coupling processing on the deformation variable data to obtain cross-point deformation coupling data;

[0037] Perform local tension collaborative clustering based on cross-point deformation coupling data to obtain local tension collaborative data;

[0038] The collaborative deformation features of the local tension collaborative data are extracted to obtain the linkage behavior feature data.

[0039] By performing cross-point deformation coupling processing on deformation data, the present invention can identify deformation correlations between different sensing points within the same time period, capturing potential mechanical coupling and growth linkage features. Using a local tension synergistic clustering method, the tension states of the coupled regions are clustered and grouped, automatically identifying structural unit regions with significant tension synergy. Synergistic deformation feature extraction can systematically identify regional linkage behavior characteristics exhibited by vegetables at specific growth stages, such as isotropic expansion and torsional coupling.

[0040] Optionally, the micro-behavior segment division includes:

[0041] Calculate the collaborative segment boundary points based on the linkage behavior feature data to obtain the collaborative segment boundary point data;

[0042] Perform morphologically similar segment matching on the linkage behavior feature data based on the collaborative segment boundary point data to obtain morphologically similar segment data;

[0043] Adjacent time series segmentation is performed based on morphological similarity segmentation data to obtain micro-behavior segment data.

[0044] By calculating the coordinated segmentation boundary points of the linked behavioral feature data, this method accurately identifies the key boundary locations where behavioral patterns change during growth, capturing turning points in microscopic growth stages. Subsequently, a morphologically similar segment matching method is used to group behavioral segments with similar structures or tension evolution trends, thereby improving the structural consistency and semantic clarity of the behavioral segments. Combining adjacent temporal segments ensures temporal continuity and logical traceability of microscopic behavioral segments.

[0045] Optionally, S3 includes:

[0046] Divide the structural area according to the multi-point growth behavior data to obtain the structural area data;

[0047] Perform behavior-dominant clustering on the structural region data to obtain behavior-dominant feature data;

[0048] Generate regional behavior embedding vectors based on the behavior-dominant feature data to obtain regional behavior representation data;

[0049] Construct a micro-situation propagation map based on regional behavior representation data to obtain micro-situation propagation map data;

[0050] The micromorphological growth state is judged on the micro-situation propagation map data to obtain the growth state model.

[0051] In the present invention, by dividing the multi-point growth behavior data into structural regions, the growth boundaries of different functional regions can be clearly defined at the spatial structure level, providing a basis for zoning modeling. Subsequently, the typical dominant behavior characteristics in each region are extracted through behavior-dominant clustering to realize the identification of regional functionality and main growth trends. Based on these characteristics, regional-level behavior embedding vectors are generated, which effectively compress multi-dimensional behavior information and retain its semantic relationships, so that the model has structured input capabilities. Constructing a micro-trend propagation map can reveal the growth trend conduction path and coupling mechanism between different regions, and the micro-morphological growth state is judged through the map to generate a complete structural and behavior-driven growth trend model.

[0052] Optionally, S4 includes:

[0053] Perform multi-scale temporal deconstruction of the growth trend model to obtain trend deconstruction data;

[0054] Based on the situation deconstruction data, the micro-situation evolution trend model is modeled to obtain the micro-situation evolution model;

[0055] The micro-trend evolution model is used to identify morphological abnormal trends and obtain growth trend prediction data.

[0056] By performing multi-scale temporal deconstruction of the growth state model, this method can extract key evolutionary features of plant growth states at different time granularities, enabling simultaneous perception of short-term fluctuations and long-term trends. Using the deconstructed data to construct a micro-state evolution model helps accurately characterize the evolutionary trajectory of local structures or regional behaviors in time series. Identifying morphological anomalies based on this evolutionary model can proactively identify potential risks such as growth stagnation, abnormal tension accumulation, or localized degradation.

[0057] Optionally, the present application further provides a flexible sensor-based vegetable growth monitoring system for executing the flexible sensor-based vegetable growth monitoring method described above, wherein the flexible sensor-based vegetable growth monitoring system comprises:

[0058] The flexible sensing perception and deformation variable calculation module is used to collect sensor resistance through the flexible sensor to obtain sensor resistance data, and calculate the deformation variable based on the sensor resistance data to obtain deformation variable data;

[0059] A multi-point growth behavior analysis module is used to perform multi-point growth behavior processing on the deformation variable data to obtain multi-point growth behavior data;

[0060] A micro-situation modeling module is used to perform micro-morphological growth situation modeling based on multi-point growth behavior data to obtain a growth situation model;

[0061] The growth trend prediction module is used to perform growth trend prediction on the growth trend model to obtain growth trend prediction data.

[0062] The present invention aims to employ flexible strain sensors deployed at key vegetable growth sites and, through high-temporal and spatial resolution resistance data collection, enable in-situ monitoring of the subtle deformation responses of plants during their natural growth process. Flexible sensors offer excellent conformability and biocompatibility, overcoming the interference issues caused by traditional rigid sensing devices, and improving the authenticity and continuity of collected data. By constructing a vector field from resistance data and inverting strain, multidimensional deformation information, including single-point micro-changes and multi-point coordinated deformation, can be extracted.

[0063] We conduct multi-point linkage behavior analysis on deformation data to construct a multi-point growth behavior map with a more structured hierarchy and behavioral semantics. This process not only identifies the coordinated growth relationships between different plant regions but also accurately locates the boundaries of behavioral pattern changes, enabling stage-by-stage summarization and behavioral coding of the growth process, significantly enhancing the depth of understanding and modeling granularity of plant growth ontology.

[0064] By dividing multi-point behavioral data into structural regions and extracting dominant behavioral features, combined with regional behavioral vector embedding and the construction of micro-trend propagation maps, we achieve structured modeling of growth trends at the regional level. Compared to traditional methods that build models based on whole-plant averages or single-point trends, this method effectively reveals inter-regional tension flows, trend transmission pathways, and coupling relationships, achieving a technological leap from "single-point detection" to "regional trend modeling."

[0065] Multi-scale temporal deconstruction and micro-trend evolution modeling of growth patterns not only captures the dynamics of growth trends at a fine-grained level but also enables early prediction of abnormal behavior. Combined with mechanisms for identifying morphological anomalies, this method can provide early warning of issues such as growth stagnation, pathological distortions, and regional degradation, providing technical support for greater predictive and regulatory capabilities in smart agriculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Other features, objects and advantages of the present application will become more apparent upon reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0067] Figure 1 A flowchart showing the steps of a vegetable growth monitoring method based on a flexible sensor according to an embodiment is shown;

[0068] Figure 2 A flowchart showing the steps of a flexible sensing perception and deformation calculation method according to an embodiment is shown;

[0069] Figure 3 A flowchart showing the steps of a multi-point growth behavior analysis method according to an embodiment is shown;

[0070] Figure 4 A flow chart showing the steps of a micro-situation modeling method according to an embodiment is shown;

[0071] Figure 5 A flowchart showing the steps of a growth trend prediction method according to one embodiment is shown;

[0072] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0073] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0074] Furthermore, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0075] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0076] See also Figures 1 to 5 , the present application provides a vegetable growth monitoring method based on a flexible sensor, the method comprising:

[0077] S1. Collecting sensor resistance through a flexible sensor to obtain sensor resistance data, and calculating a deformation variable based on the sensor resistance data to obtain deformation variable data;

[0078] In one embodiment, the sensor uses a flexible strain sensor based on the piezoresistive effect. The sensor is composed of a conductive polymer material as a matrix, such as a composite conductive structure formed by doping carbon nanotubes (CNT) in polydimethylsiloxane (PDMS), and is used to achieve sensitive response to micro-deformations of vegetables. During the data acquisition process, the system sets the sensor's resistance sampling frequency to ten times per second (10Hz) to ensure a high time domain resolution, so as to accurately capture the continuous response signal of the plant structure changing over time. To obtain the resistance change output by the sensor, a differential bridge circuit is used to measure the voltage across the two ends of the sensor in real time, and the resistance change value of the sensor at the current moment is obtained through the difference output. This change value is defined as the difference between the current resistance value and the initial static resistance value, which is recorded as "resistance change". In terms of deformation calculation, the system derives strain based on the resistance change, the initial resistance value, and the sensitivity coefficient of the sensor (i.e., the strain gauge coefficient). Specifically, the deformation The calculation formula is as follows: ,in, is the difference between the current resistance and the initial resistance; is the initial resistance; is the calibrated sensitivity coefficient of the sensor, which is 2.0~3.5.

[0079] S2. performing multi-point growth behavior processing on the deformation variable data to obtain multi-point growth behavior data;

[0080] In one embodiment, the system deploys flexible strain sensor nodes in multiple key parts of the plant body, including but not limited to the petiole area, the middle section of the main stem, and the typical leaf area. Each sensor node is numbered separately, for example, numbered Node_1 to Node_n, and forms a spatially distributed sensor network to sense the growth and deformation of various parts of the vegetable. During the data processing process, the system uses a sliding time window mechanism to perform segmented analysis on the deformation variable time series data of each sensor node. Specifically, the time window length is set to 60 seconds and the sliding step length is 10 seconds. The system continuously calculates the deformation variable within each time period to extract the microscopic growth behavior characteristics. In terms of behavior recognition rules, the system sets the following judgment criteria: if the deformation variable of a certain sensor node shows a monotonically increasing trend within three consecutive time windows, and the rate of change per unit time is continuously greater than the preset threshold (for example, the deformation variable increase per second is greater than 0.01), then the node is judged to have rapid growth behavior within the time segment; if the deformation variable change curves of two or more different sensor nodes have a strong linear correlation within the same time window, that is, their Pearson correlation coefficient is greater than 0.8, and the number of sample points is not less than five, then the system determines that there is coordinated growth behavior within the time window. For event segments that meet the above behavior conditions, the system performs structured annotation on them and outputs basic information for each growth behavior event, including the node number, the start time and end time of the behavior, the average deformation variable value within the segment, and the behavior type label (for example, "rapid growth", "slow growth" or "stable state"). The specific structure is as follows: node number, which indicates the corresponding sensing point; start time and end time, which indicate the continuous time segment of growth behavior; average deformation value, which is the arithmetic mean of all deformations in the time segment; behavior type label, which is based on the rule recognition result, such as fast, coordinated, slow or stable.

[0081] S3. Modeling the micromorphological growth situation based on the multi-point growth behavior data to obtain a growth situation model;

[0082] In one embodiment, to achieve temporal reasoning and spatial coordination modeling of multi-point flexible growth behaviors of vegetables, the system uses a Bayesian dynamic network to model the behavioral sequence data of sensor nodes. The modeling process uses the morphological behavioral characteristics of each sensor node as state variables, constructs temporal dependencies between nodes, and forms a micromorphological growth trend map with causal directionality and probabilistic structure. Specifically, in this Bayesian dynamic network model, network nodes are growth state variables corresponding to each flexible sensor deployment location, indicating whether the node is in an active growth state within a certain period of time. Directed edges in the network represent the behavioral precedence relationship between different sensor nodes, and the direction of the edge indicates the temporal causal order, for example, "a growth event at a petiole node occurs before a growth event at a main stem node." In terms of model parameter construction, the system extracts the following key indicators for each pair of sensor nodes with behavioral linkage: a node activity index, which measures the frequency of growth behavior events occurring at each node within a specified time window; a higher frequency indicates a more active node; and an inter-node morphological coordination index, which analyzes the temporal co-occurrence of behavioral events at two nodes and calculates the probability of coordination based on the behavioral sequence. For example, in a certain historical sample, if a growth event for node A is immediately followed by an event for node B, and the occurrence rate exceeds 80% of the total sample, the synergy probability is recorded as 0.8. The model's output structure is a set of causal structures, where each directed connection is represented as a triplet: a source node and a target node (representing the growth order); a behavioral linkage probability (representing the probability that the target node will grow after the source node grows); and a behavioral time delay value (representing the average time, in minutes, from the source node's growth to the target node's growth). If the linkage probability between a pair of nodes exceeds 0.8 and the time delay is less than 5 minutes, the system determines that there is a strong coupled growth relationship between the node pairs and marks them as a high synergy region. Furthermore, if a network structure contains closed dependency paths between three or more nodes, for example, node A influences B, B influences C, and C in turn leads A (forming a closed loop of A→B→C→A), the structure can be identified as a periodic structural growth pattern, indicating that the region exhibits periodic and recurring growth rhythmic behavior.

[0083] S4. Perform growth trend prediction on the growth trend model to obtain growth trend prediction data.

[0084] In one embodiment, in order to achieve forward-looking prediction of plant micromorphological growth trends, the system constructs a multivariate time series prediction model based on long short-term memory neural network (LSTM). The model can combine the historical deformation changes of multiple sensor nodes, learn their temporal evolution laws, and predict the growth trend in the future period. In terms of model input, the system collects past continuous At each time point, the distribution of The deformation data of each sensor node is constructed as a multivariate time series input sample. Each input sample contains Under this condition, the deformation variable value corresponding to each node forms a A sequence of dimensional feature vectors. This sequence is used to capture the temporal coordination of growth behavior between different spatial locations. The model training goal is to predict the shape trend of each node over a future period (e.g., the next T minutes), that is, to output a sequence of shape estimates for each node at subsequent time steps. For example, a multivariate time series prediction model processes locally stored historical shape data through one or more stacked LSTM units. This processed data is then fed into a self-attention weighted calculation to generate weighted data. This weighted data is then processed through a fully connected layer to produce a sequence of shape estimates. The system sets the prediction time step to coincide with the historical sampling step. In terms of prediction result judgment rules, if the prediction results of a certain node show an increasing trend in deformation variables within three consecutive time steps, and the increase is stable (that is, the predicted value of each step is higher than the previous step), the system automatically triggers a "continuous growth period" warning, indicating that the part is in a highly active growth state; if the prediction results show that the growth rate of the deformation variable has slowed significantly, that is, the change amplitude within multiple consecutive time steps is lower than the preset threshold (such as a single-step growth value less than 0.005), the system determines it as a "growth stagnation" trend and triggers corresponding prompts to support management decisions. The system supports visual output of prediction results, specifically comparing the predicted deformation variable curve with the historical observation curve, automatically generating a trend chart, and marking the difference sections between the predicted path and the historical path (such as the time period when the predicted curve is higher or lower than the historical average). This visualization function allows users to intuitively judge the future growth potential and risk changes of vegetables.

[0085] Optionally, S1 includes:

[0086] S11, collecting sensor resistance data by using a flexible sensor arranged at a preset first position data on the vegetable to obtain first sensor resistance data;

[0087] In one embodiment, the first position is preferably set at the upper third of the height of the main stem of the vegetable, which has good response sensitivity to longitudinal stretching. The type of flexible sensor used is a piezoresistive sensor with a thickness of no more than 0.3 mm and a working resistance value range of 1 kilo-ohm to 10 kilo-ohm. The data acquisition method is to use an analog-to-digital conversion (ADC) acquisition module to sample electrical signals. The module has 12-bit accuracy and a maximum sampling frequency of 500 Hz. In this embodiment, 50 sampling points are selected per second for recording, and continuous sampling for 5 minutes is set as a complete observation cycle. After collection, the original resistance signal is pre-processed by a sliding average filtering algorithm, and the filter window size is 5 data points. The resistance acquisition results formed are output in time series to form a resistance time series data set, where each set of data contains a sampling time and a corresponding resistance value, that is, each pair of data is represented as "at the The sensor resistance value collected at each moment".

[0088] S12, collecting sensor resistance data by using a flexible sensor arranged at a preset second position data on the vegetable to obtain second sensor resistance data, wherein the first position and the second position are different positions;

[0089] In one embodiment, the second position is preferably set at the base area of ​​the vegetable petiole or unfolded leaf, which is highly sensitive to changes in the unfolding angle and slight bending deformation, and is helpful in identifying the lateral structural behavior of the plant. The specifications of the second sensor are consistent with those of the first sensor. Both are flexible piezoresistive sensors with a thickness of no more than 0.3 mm and a working resistance range of 1 kilo-ohm to 10 kilo-ohm, aiming to ensure the uniformity of the data processing process. In terms of data acquisition, the system adopts a unified clock drive or synchronous channel architecture to ensure that the resistance data of the two sensors can be synchronously collected at the same time point, forming a data comparison relationship with consistent time sequence. In the second set of resistance time series formed after acquisition, each data point consists of the sampling moment and the resistance value of the second sensor at the corresponding moment, so as to perform modeling and analysis with a one-to-one correspondence between the time points of the data collected by the first sensor.

[0090] S13, constructing a resistance space map based on the first sensor resistance data and the second sensor resistance data to obtain sensor resistance data;

[0091] In one embodiment, the resistance space graph is defined as a two-dimensional feature space graph, where the horizontal axis represents the resistance value of the first sensor at each time point, and the vertical axis represents the resistance value of the second sensor at the corresponding time point. At each sampling moment, the system constructs a two-dimensional spatial coordinate point from the synchronized resistance pairs of the first and second sensors, denoted as a "resistance point." All resistance points are arranged in chronological order to form a trajectory curve that evolves in two-dimensional space. This trajectory curve can be used to reflect the strain coordination relationship between different monitored parts of the vegetable. For example, if the resistance trajectory distribution is primarily concentrated near the diagonal, it indicates that the deformation processes of the two monitored parts are highly synchronized; if the trajectory shows significant deviation or branching, it indicates asynchronous growth or regional deformation differences between the two monitoring points. A three-dimensional resistance evolution graph is constructed by introducing the time dimension based on the two-dimensional resistance space graph. Each spatial point is composed of a triple, representing the resistance value of the first sensor, the resistance value of the second sensor, and the corresponding sampling time.

[0092] S14. Calculate the deformation variable of the sensor resistance data to obtain deformation variable data.

[0093] In one embodiment, the unit strain of flexible sensors at two locations is calculated using resistance changes. The calculation formula is as follows: The deformation at a given point in time is equal to the difference between the resistance at that moment and the initial resistance of the sensor, divided by the product of the initial resistance and the strain gauge sensitivity coefficient. The initial resistance is the stable output value of the flexible sensor when no strain is applied and should be obtained through static sampling during the system initialization phase. The strain gauge sensitivity coefficient is a calibration parameter of the sensor and is set to 2.0 in this embodiment. Through the above method, deformation sequences for the sensors at the first and second locations are obtained, respectively. To determine whether there is asynchronous growth or strain response differences between different parts of the plant, the system compares the difference between the two deformation sequences and calculates the absolute difference in deformation at each moment. This difference is used to assess the synchronization and growth deviation trends of different regions of the plant. The output data structure includes the sampling time at each time point, the deformation at the first location, the deformation at the second location, and their corresponding deformation difference.

[0094] Optionally, the step of generating the first position data and the second position data includes:

[0095] Get vegetable image data;

[0096] In one embodiment, the image acquisition device uses a red, green, and blue three-channel camera (RGB camera) with a resolution of no less than 1920×1080 pixels; or an imaging device with depth perception capabilities, such as the Intel RealSense series depth cameras, to accurately capture the three-dimensional morphology of vegetables. Image acquisition conditions include uniform ambient lighting, avoiding strong shadows or reflections to maintain overall image brightness consistency. The camera's shooting angle is set at a 45-degree downward angle relative to the ground, ensuring sufficient capture of the main stem while maintaining a full view of the petiole, thereby acquiring the multi-region image features required for structural identification.

[0097] Extracting the petiole extension axis according to the vegetable image data to obtain petiole extension axis data;

[0098] In one embodiment, during the image preprocessing stage, an image binarization method is used to convert the original image into a black and white image, wherein the threshold segmentation strategy may select Otsu's Method to achieve an adaptive binary segmentation effect by maximizing the inter-class variance; subsequently, the Canny edge detection algorithm is applied to extract the edge structure in the image to identify the outer contour of the vegetable leaves. A morphological closing operation is performed on the edge image. The closing operation includes a dilation operation followed by an erosion operation, which can effectively fill small holes and connect broken edge areas. After the edge is closed, a contour recognition operation is performed to extract the contours of closed or semi-closed edge areas from the image. This process analyzes the connectivity of pixels and marks edge paths with closed properties to distinguish multiple leaf areas. After the contour extraction is completed, the central axis extraction process is performed on each leaf area. The specific method involves pixel thinning of each petiole region using a skeletonization algorithm. The Zhang-Suen thinning algorithm is recommended. This algorithm thins the image contour into a single-pixel-wide skeleton line while preserving the topological structure. Starting from the base of the petiole near the main stem, the skeleton path is extended outward to fit the centerline of the petiole and record it as the petiole extension axis. The petiole extension axis data is output as a chain coordinate sequence, and each axis path can be represented as an ordered set of two-dimensional coordinate points.

[0099] Calculate the main stem area center point based on the petiole extension axis data to obtain the main stem area center point data;

[0100] In one embodiment, the main stem region is defined as the geometric aggregation area of ​​the area where the starting points of the central axis of all petioles are located. That is, the starting coordinate point of each petiole corresponds to the junction where it connects to the main stem, so the collection of all starting points constitutes the structural center area of ​​the main stem. The starting point coordinates of each petiole extension axis are extracted and expressed as a plane coordinate pair. The arithmetic mean of all starting point coordinates in the horizontal and vertical directions are calculated and used as the horizontal and vertical coordinate components of the center of gravity of the main stem region. ,in is the horizontal and vertical coordinate components of the center of gravity of the main stem area, is the coordinate quantity data, is the coordinate order term, For the The horizontal axis components, For the The system can set a spatial consistency filtering mechanism for the main stem aggregation area. Specifically, the maximum aggregation radius of the petiole starting point is defined, that is, the radius from all starting points to the center of gravity. If the radius is less than a preset spatial threshold (e.g., 5 pixels), the plant can be considered to have a single main stem structure. Otherwise, further evaluation is required to determine whether there are multiple main stems or heterogeneous plant structures. The output main stem area centroid data is expressed in two-dimensional coordinate form, i.e., centroid coordinates.

[0101] A tension mapping model is constructed according to the center of gravity data of the main stem area to obtain a tension mapping model;

[0102] In one embodiment, based on the structural characteristics of the petiole of the plant, the tensile tension it bears during its growth is inferred. The tensile tension on the petiole is related to its length and the angle at which it deviates from the center of gravity of the main stem. For each petiole, the estimated tension value is calculated based on the spatial relationship between its corresponding central axis data and the center of gravity of the main stem. Let the central axis of the petiole be numbered Axis line, corresponding tension estimate Calculate according to the following relationship: ,in For the Estimated tension in the petiole, is the adjustment coefficient, which is set according to physiological mechanics (it can be set to 1 empirically). is the pixel length of the petiole midline, For the The central axis of the petiole, is the angle between the petiole axis and the main stem vector The cosine value indicates the degree to which the petiole direction deviates from the center of gravity of the main stem. The closer the angle is to the radial direction of the center of gravity, the greater the tension. For the The angle between the petiole axis and the main stem vector. The output of the tension mapping model is a data structure consisting of a set of petiole axes and corresponding tension estimates, represented by the tension mapping model , is the tension mapping model, For the The central axis of the petiole, For the Estimated tension in the petiole, Index number for the petiole, is the total petiole number.

[0103] Position the extreme tension point according to the tension mapping model to obtain the extreme tension point data;

[0104] In one embodiment, a maximum point represents the location where tension on a particular petiole axis reaches a local peak, typically reflecting the region experiencing the strongest tensile stress. In practice, for each petiole midline, the location corresponding to the maximum tension value in its tension distribution is selected as the maximum point. A minimum point is a local tension valley, typically occurring near the center of gravity of the main stem or the distal edge of the petiole, representing a low-response region in the tension conduction path. Such points are useful for identifying tension starting points or buffer segments. The tension values ​​of each petiole in the constructed tension mapping model are serialized as a continuous function along the axis. A local extremum search strategy is employed to analyze each tension function and identify its local maximum and minimum points. This search method, based on a sliding window neighborhood comparison strategy or a standard extremum detection algorithm (such as local neighborhood comparison), is used to identify locations on the function curve where local increases transition to decreases, or vice versa. The extracted tension extreme difference point data includes the maximum and minimum points on all petioles, each output in the form of their corresponding image coordinates.

[0105] The position data corresponding to the maximum value data in the tension extreme difference point data is determined as the first position data, and the position data corresponding to the minimum value data in the tension extreme difference point data is determined as the second position data.

[0106] In one embodiment, the first position data refers to the image coordinate position corresponding to the tension maximum point in the tension extreme difference point data. Such positions are usually located at the edge of the leaf, the distal end of the petiole, or the maximum force area close to the stem, and have a large structural deformation variable, so they are suitable for deploying flexible sensors for sensing rapid growth or tensile strain. The second position data refers to the image coordinate position corresponding to the tension minimum point in the tension extreme difference point data. Such points are usually close to the center of gravity of the main stem or the inner low-tension area of ​​the petiole, have a relatively stable structural state, and are suitable for deploying sensor devices for recording reference states or monitoring small deformations. If the Euclidean distance between the maximum point and the minimum point is greater than the set minimum threshold (for example, 10 pixels), the point pair is considered to constitute a valid tension layout pair.

[0107] Optionally, the deformation variable data includes single deformation variable data and linkage deformation variable data, and the deformation variable calculation includes:

[0108] According to the sensor resistance data, a resistance change vector field is constructed to obtain resistance change vector field data;

[0109] In one embodiment, the resistance change of each sensor node is defined. Sensors at the time The measured resistance is , the resistance at the initial installation moment is , then the sensing point is at time The resistance change is defined as the difference between the two, which is: ; The system binds the resistance change of each sensor node with its actual installation coordinates in two-dimensional space to form a resistance change record point with position information. That is, each data point can be expressed as a triple, including the lateral position, longitudinal position of the sensor and the resistance change corresponding to the position. In terms of spatial modeling, the system divides the entire monitoring area into two-dimensional spatial grids based on the coordinates of the sensor nodes. For each grid unit, the system extracts the resistance change data from the sensor points within its coverage area and performs spatial interpolation processing. When the sensor points in a certain grid area are sparsely distributed, the Kriging interpolation method or the bivariate spline interpolation method can be used to smoothly fill in the resistance change trend of the area. The system outputs the resistance change vector field result, which is expressed as a three-dimensional mapping function based on position and time. This function takes any two-dimensional spatial coordinates and time As input, the output is the resistance change value of the corresponding position at that time point. This function reflects the micro-deformation intensity and distribution trend of plant tissue in the entire spatial area at a certain moment.

[0110] Perform strain inversion mapping calculation based on resistance change vector field data to obtain strain inversion data;

[0111] In one embodiment, the system uses a calibrated resistance-strain mapping relationship to convert resistance change information into a strain quantity that represents the deformation intensity of plant growth. Specifically, the system sets the following mapping formula: ,in For spatial location in The dependent variable at time, For this location The resistance change at that moment, is the initial resistance distribution (which can be fitted by the initial frame); The calibration sensitivity coefficient of the sensor is determined through experimental calibration, and its value is generally taken from 2.0 to 3.5. The above calculation results form a three-dimensional spatial mapping function, which is recorded as the scalar strain field in the two-dimensional spatial region at each time point. This strain function characterizes the strain intensity of different areas on the surface of the vegetable over time. If the flexible strain sensor used has a two-dimensional directional sensitive structure (such as a strain gauge or a matrix sensor array with a two-axis cross arrangement), the system can deduce the local strain tensor field at that position, such as ,in represents the shear strain, estimated by multi-directional resistance difference, is the horizontal main response variable, First back The shear strain, First back The shear strain, is the vertical principal strain variable. Specifically, in the two-dimensional strain tensor mode, the system can calculate each spatial position The strain tensor on the horizontal axis is the principal strain variable (along Axis direction); vertical principal strain (along Axis direction); shear strain ( deformation component in the coupling direction).

[0112] Calculate the monomer deformation variable according to the strain inversion data to obtain the monomer deformation variable data;

[0113] In one embodiment, the so-called single region refers to a microscopic sensing unit centered on the location of a single flexible strain sensor, which can be an independent sensing point or a fixed spatial neighborhood centered on it. For each single region, the system and the current moment The local strain value under , calculate its linear deformation along the main observation direction. Set the reference length of the region in the initial state to , then the area is at time The length change (i.e., monomer shape variable) under the condition of The local strain, which describes the deformation at time The overall extension or compression of a local structure in a certain direction. If the flexible strain sensor used supports two-dimensional directional strain measurement (such as a tensor strain type structure), the system can use the eigenvalue analysis method of the local strain tensor to extract the principal strain direction angle and principal extension value of the region at the current moment.

[0114] The linkage deformation variable data is processed according to the monomer deformation variable data to obtain the linkage deformation variable data.

[0115] In one embodiment, if multiple monomers are in the time window The internal variables have the same trend (growth / contraction) and the Pearson correlation coefficient , defined as linkage; set sliding time window , step length ; Extract all deformation variable sequence; calculate the dynamic correlation matrix between any two nodes; perform hierarchical clustering on high-correlation nodes to form linkage groups; the system extracts the following feature indicators for each linkage group, including the average linkage amplitude, linkage response delay (maximum response difference duration), and linkage direction consistency, so as to integrate the linkage groups and features together to obtain linkage deformation variable data.

[0116] Optionally, S2 includes:

[0117] S21, extracting linkage behavior features from the shape variable data to obtain linkage behavior feature data;

[0118] In one embodiment, the input data includes a time-series deformation variable sequence corresponding to each flexible sensor, that is, a continuous strain change curve with time as the horizontal axis and deformation variable as the vertical axis. The system uses a sliding window mechanism for data processing. The specific settings are as follows: the time window length is 60 seconds, and the window sliding step is 10 seconds. In each sliding window, the system extracts the deformation variable curves of all sensor nodes in this period as analysis objects. The system constructs a collaborative index between nodes. The Pearson correlation coefficient of the deformation variable curve of each pair of sensor nodes is calculated to measure the similarity of the deformation variable trends of the two. If the correlation coefficient is greater than 0.85, it is considered that the pair of nodes has significant synchronous growth behavior within the period. At the same time, the system also calculates the time difference between the strain peaks of each pair of nodes. If the difference is less than 5 seconds, it is judged that the two nodes have response consistency.

[0119] S22, dividing the micro-behavior segments according to the linkage behavior feature data to obtain micro-behavior segment data;

[0120] In one embodiment, a microbehavioral segment refers to a basic unit characterized by coordinated changes in the deformation response patterns of multiple sensor nodes within a specific time interval. Based on the linkage behavior feature vectors of each sensor node, the system constructs a continuous curve showing the linkage feature changes over time, forming a linkage feature time series. The system performs rate of change analysis on this time series and calculates the magnitude of the continuous time change of each feature indicator. If any feature indicator (such as the number of synchronized nodes, maximum correlation, or average response delay) undergoes a significant sudden change within a certain time period, and its first-order difference value exceeds a preset rate of change threshold, that time point is considered a potential behavioral state boundary. The system incorporates a temporal smoothing mechanism to ignore all transient segments lasting less than 15 seconds. After segment boundaries are identified, the system uses a clustering algorithm to merge the linkage behavior features of different time periods. For example, a density-based spatial clustering algorithm (such as DBSCAN) or a density peak clustering algorithm is used to group time periods with similar behavioral patterns into the same behavioral cluster. Each clustering result constitutes a microbehavioral segment.

[0121] S23. Construct a growth behavior map for the microscopic behavior segment data to obtain multi-point growth behavior data.

[0122] In one embodiment, each node in the constructed graph structure corresponds to a specific sensor monitoring point. Each node contains multiple attributes, including the types of micro-behavioral segments in which the monitoring point has participated, the cumulative number of segments in which it has participated, and the average magnitude of its deformation change within these segments. These attributes are used to describe the typical response characteristics of the monitoring point during vegetable growth. The presence of an edge between any pair of nodes in the graph indicates that the two nodes have exhibited coordinated behavior in at least one behavioral segment. Each edge is also assigned multiple attributes to quantify this synergistic relationship, including the number of linkage participations (i.e., the number of segments in which the pair of nodes co-appears); the total duration of the linkage; and the average degree of deformation synchronization across all shared segments. The specific process of graph construction involves the system traversing each micro-behavioral segment and extracting the set of all sensor nodes involved in that segment. All pairs of nodes in this set are then paired, generating an edge between each pair of nodes. The synergy strength of the edge is weighted by multiplying the segment duration by the correlation coefficient between the pair of nodes within the segment. This weight is used to measure the actual degree of synergy between the node pairs in each time segment. By integrating the inter-node linkage information in all fragments, the system generates a multi-point growth behavior graph containing a complete set of nodes and edges.

[0123] Optionally, the linkage behavior feature extraction includes:

[0124] Performing cross-point deformation coupling processing on the deformation variable data to obtain cross-point deformation coupling data;

[0125] In one embodiment, the deformation data from multiple flexible sensors are systematically coupled and analyzed to construct a dynamic synergistic relationship between different sensing points to reveal the structural linkage characteristics of vegetable plants during their natural growth process. The coupling process mainly combines the two dimensions of temporal correlation and spatial proximity to calculate the degree of deformation synergy between each pair of sensor points. , and define its coupling as ,in For the moment The coupling degree under is the time / space coupling adjustment coefficient, set to 0.7, is the Pearson correlation coefficient of the time series of the shape variable, Sensing point The time series of the shape variables, Sensing point The time series of the shape variables, is the base constant of natural logarithms, Sensing point The space coordinate vector of Sensing point The space coordinate vector of is the spatial perception scale factor, which is 0.2 multiplied by the maximum geometric scale of the sensor point set. The first term is the Pearson correlation of the deformation variable curve; the second term is the spatial distance attenuation factor. All the calculated point-to-point coupling degrees will form a set of cross-point coupling data corresponding to the timestamp, and only the coupling strength exceeding the threshold (such as =0.85) for the point pair.

[0126] Perform local tension collaborative clustering based on cross-point deformation coupling data to obtain local tension collaborative data;

[0127] In one embodiment, in order to discover local areas with cooperative deformation relationships in the plant structure, the above-mentioned cross-point deformation coupling data is clustered to form multiple tension coordination units. Based on the graph structure and combined with the coupling relationship between the sensing points, local growth tissues with cooperative spatial position and deformation trend are identified. Constructing a non-directed graph ,in is the sensing point, each node corresponds to a flexible sensor at a spatial position; For coupling pairs , the edge weight is , is the coupling degree at the corresponding moment. Using a graph-based clustering algorithm, such as the Louvain community discovery algorithm or the spectral clustering algorithm, the above graph The clustering objective is to maximize the sum of edge weights within each cluster (i.e., strong intra-cluster coupling) and minimize the strength of edge connections between clusters (i.e., high inter-cluster independence). After clustering, each cluster is considered a local tension coordination unit. The results of all coordination units are aggregated and output as a tension coordination clustering data set.

[0128] The collaborative deformation features of the local tension collaborative data are extracted to obtain the linkage behavior feature data.

[0129] In one embodiment, for each collaborative unit , extract its temporal deformation evolution characteristics, including the tension amplitude change rate , represents the rate of change of the overall shape of the cluster, is the tension amplitude change rate of the collaborative unit, is the average deformation of the collaborative unit, is the differential operator, is the current observation time; synchronization calculation ,in is the synchronization degree of the collaborative unit, is the standard deviation of the shape variable, For all sensor points in the unit at time The maximum value of the shape variable, For all sensor points in the unit at time The minimum value of the deformation variable, the closer the value is to 1, the more synchronized it is; if the difference in the principal strain direction of the nodes in the cluster is less than 20°, it is recorded as consistent in direction; if the time window summary feature is set as a summary period of 60s, the average synchronization degree, direction consistency rate, and tension change mean within the period are calculated; based on the above characteristic index results, the corresponding behavior type label of each tension coordination unit can be automatically assigned according to the preset logical rules. For example, if the synchronization degree Higher than 0.85 and tension amplitude change rate Above the set threshold , it is marked as "co-growth"; if the synchronization If it is lower than 0.4, it is judged to be in the "differentiation growth" state; the above is integrated into linkage behavior feature data.

[0130] Optionally, the micro-behavior segment division includes:

[0131] Calculate the collaborative segment boundary points based on the linkage behavior feature data to obtain the collaborative segment boundary point data;

[0132] In one embodiment, the input data is a set of coordinated behavior features extracted in the previous stage. Each coordinated behavior cluster contains feature information at multiple consecutive time points, such as coordinated response amplitude, coordinated correlation, and coordinated direction consistency. For each coordinated behavior cluster, the system analyzes the temporal changes in feature values ​​within fixed time windows (e.g., 60 seconds). By calculating the feature difference between adjacent time windows, the cluster's rate of change index at the current moment is obtained. If this rate of change index exceeds a set dynamic threshold (the threshold is set as the sum of the mean standard deviation of the three features), a sudden change in the behavioral state is considered to have occurred at that moment, marking it as a potential behavioral switching point. The dynamic threshold can be determined by a weighted combination of the mean standard deviation of each feature within the cluster. The system merges the potential switching points identified in multiple behavioral clusters and uses a voting mechanism to identify candidate points with high temporal overlap, which are used as the global coordinated behavior switching moments. To avoid false boundaries caused by data jitter, the system implements a redundant boundary filtering mechanism: if the time interval between any two boundary points is less than 15 seconds, it is considered to have dense boundary interference. Only representative boundary points with intermediate time are retained, and the remaining points are deemed redundant and deleted.

[0133] Perform morphologically similar segment matching on the linkage behavior feature data based on the collaborative segment boundary point data to obtain morphologically similar segment data;

[0134] In one embodiment, the system uses the time interval formed by adjacent boundary points as the analysis unit, for example, each continuous time period It is regarded as a segment to be analyzed. In each segment, the statistical representation of its linkage behavior characteristics is extracted to construct the feature vector of the linkage behavior cluster. This vector includes but is not limited to the following indicators, including statistics such as average linkage amplitude, average correlation, directional consistency variance, and the number of active clusters. For the feature vectors of all segments, the system uses similarity measurement methods such as cosine similarity or dynamic time warping (DTW) to compare whether the behavioral morphology of adjacent time segments is consistent. If the similarity index of two segments exceeds the preset threshold (for example, 0.85), they are judged to maintain consistency in behavioral morphology and are marked as morphologically similar segments, which can be merged and classified; if the similarity drops significantly, it is regarded as a morphological transition.

[0135] Adjacent time series segmentation is performed based on morphological similarity segmentation data to obtain micro-behavior segment data.

[0136] In one embodiment, the system determines the temporal continuity of all morphologically segmented data. If two or more adjacent behavioral segments meet the following conditions, such as having the same behavioral type label and the time interval between adjacent segments being less than a set maximum segment gap (e.g., 30 seconds), the system merges these segments into a single, continuous, unified micro-behavior segment, indicating structural continuity and morphological consistency in terms of temporal sequence. If the duration of an independent segment is less than a set threshold (e.g., 20 seconds), it is considered an interference segment and can be selectively deleted or merged into an adjacent main segment of the same type. After all segmentation is completed, the system performs a chronological reordering operation.

[0137] Optionally, S3 includes:

[0138] S31, dividing the structural region according to the multi-point growth behavior data to obtain structural region data;

[0139] In one embodiment, the spatial distribution of the aforementioned behavioral segments is integrated to identify a set of spatial regional units, each representing a structural region with a significant concentration of growth behaviors during monitoring. This demarcation serves as the foundational spatial hierarchical unit for subsequent growth state modeling and regional propagation analysis. The system extracts the sensory location corresponding to each behavioral segment (e.g., the spatial centroid of its node or location within the cluster) and calibrates it as the coordinates of the centroid point in a two-dimensional plane. The set of centroid points for all behavioral segments is then input into a density clustering algorithm (e.g., DBSCAN) to identify groups of regions with significant spatial clustering characteristics. Clustering parameters are set, such as the minimum number of samples within a cluster, which is set to 3, indicating that a region is considered a valid cluster only if it contains at least three behavioral segments; and the distance threshold is set to 10 pixels, indicating that two behavioral segments are considered densely connected if the distance between their centroids is less than this threshold. The system identifies each valid cluster as a spatial structural region, indicating that multiple growth events with significant spatial proximity, temporal correlation, and behavioral characteristics have occurred within that spatial range.

[0140] S32, performing behavior-dominant clustering processing on the structural region data to obtain behavior-dominant feature data;

[0141] In one embodiment, for each structural region, the system extracts a feature set of all behavioral segments contained within it, forming a behavioral feature set for that region. This feature set includes metrics such as the average deformation value, behavior frequency, amplitude intensity, and label type for each segment, forming a multidimensional sequence of behavior vectors. For continuous metrics in this behavioral feature set (such as deformation, amplitude, and frequency), the system uses a standard z-score normalization method. This method subtracts the mean of each dimension from each feature value and divides it by the standard deviation, making all features comparable within the same dimension. The normalized feature set is then input into a clustering algorithm for analysis. The system can use KMeans or Gaussian Mixture Model (GMM) algorithms to cluster the behavioral segments. The clustering results are used to identify multiple typical behavioral pattern centers within the region. The system then focuses on extracting representative feature vectors corresponding to these cluster centers (based on a pre-set parameter label library, including coordinated growth, vigorous expansion, and slow peristalsis). Within the clustering results, the system identifies the cluster with the highest frequency (i.e., the behavior cluster containing the most segments) and defines the behavior type corresponding to this cluster as the dominant behavior label for the region. For example, when most of the fragments in a certain structural region are clustered into the “co-growth” class, its region label is set to “co-growth”.

[0142] S33, generating a regional-level behavior embedding vector based on the behavior-dominant feature data to obtain regional behavior representation data;

[0143] In one embodiment, the dominant behavior feature vector is concatenated with other statistical information, and a dimensionality reduction method (such as PCA or autoencoder neural network) is used to compress the embedding vector to a fixed dimension (such as 16 dimensions); all region embeddings are normalized to ensure comparability.

[0144] S34, constructing a micro-situation propagation map based on the regional behavior representation data to obtain micro-situation propagation map data;

[0145] In one embodiment, each node in the graph corresponds to a regional behavior embedding vector generated in the previous step, that is, the behavior feature vector of each structural region The system determines whether to establish an edge between two region nodes based on the following two conditions: a spatial proximity edge, if the spatial distance between the geometric centroids of the two structural regions is less than a preset distance threshold (e.g. , in pixels or centimeters), it is considered that there is a direct spatial association and an edge is added between the two nodes; behavioral similarity edge, if the cosine similarity between the embedding vectors of the two regions exceeds the preset behavioral similarity threshold (for example , set to 0.85), then the behavior patterns are considered to be highly consistent, and an edge is added between the two. The above two types of edges can overlap, and the system supports enhancing the edge weight expression when both space and behavior meet the conditions. For any pair of regional nodes and If the edge conditions are met, the system calculates the edge weight , which is used to represent the propagation potential or influence intensity of growth behavior between regions. The edge weight is composed of a weighted combination of spatial similarity and behavioral similarity: ,in is the node linkage weight value, is the distance penalty weight coefficient, which is 0.4. is the spatial or topological distance between nodes, is the embedding similarity weight coefficient, which is set to 0.6. is the embedding feature similarity function, that is and The similarity calculation is performed to obtain the result. The micro-situation propagation graph generated by the system is represented as a graph structure , where the node set Consists of all region embedding vectors; edge set Contains edges that meet the connection conditions. Each edge contains a triplet of information: the starting node number, the target node number, and the corresponding edge weight.

[0146] S35. Perform micromorphological growth state discrimination on the micro-situation propagation map data to obtain a growth state model.

[0147] In one embodiment, the system uses the Louvain community partitioning algorithm to perform unsupervised clustering on the micro-situation propagation graph. This method, based on the principle of maximizing the modularity of the graph, automatically identifies locally highly connected substructure regions (i.e., communities or subgraphs) within the graph. For each identified community structure, the system statistically analyzes the distribution characteristics of the behavioral embedding vectors of the nodes it contains, including embedding concentration (inter-vector variance or clustering tightness) and dominant behavior type consistency (coverage of the largest proportion of behavior types). If the embedding distribution within a community is concentrated and the behavior type consistency is high (e.g., the proportion of similar behaviors exceeds 80%), the community is determined to be a stable growth domain. If the embedding vector distribution is more dispersed and the behavior labels are diverse and mixed, the community is determined to be a polymorphic competition domain, indicating the presence of potential unstable growth or incomplete differentiation trends within the region.

[0148] In one embodiment, if the system already has annotated data on regional behavior trends over a certain historical period (e.g., manually recorded labels such as "rapid expansion" and "growth stagnation"), a graph convolutional neural network (GCN) model is constructed (using historical data to train the neural network, including convolutional layers, pooling layers, fully connected layers, and weight calculation to build a deep model) to perform node-level state inference. Model training uses a micro-situation propagation graph as input and regional behavior embeddings as node features to train a graph convolution classification model. The model outputs a corresponding growth trend label for each regional node. Label categories include, but are not limited to, stable coordination states, where multiple regions exhibit similar behavior with low volatility; asynchronous differentiation states, where some regions advance or lag behind the main behavioral rhythm; rapid transition states, where sudden changes in shape or rapid reversals in direction; and pauses and stagnation states, where almost invisible changes in variables lead to growth stagnation. This graph neural network classification method can enhance inference accuracy through propagation structures even with a small number of training samples, making it suitable for deployment in long-term monitoring systems to provide trend warnings and strategic intervention prompts.

[0149] Optionally, S4 includes:

[0150] S41. Perform multi-scale situation time series deconstruction on the growth situation model to obtain situation deconstruction data;

[0151] In one embodiment, the system sets multiple typical time windows as different analysis scales, for example, 5 minutes, 30 minutes, and 2 hours. Each time scale represents a level of analytical granularity, used to capture rapidly changing behaviors, medium-term structural evolution, and long-term trends. For each spatial region, the system constructs corresponding situation label sequences at different time scales. If a dominant segment label exists within the time period (for example, a certain type of behavior accounts for more than 70%), this label is directly assigned. If the various behavior labels frequently alternate and there is no clear dominant trend, the system determines a state of "situational disorder." If an embedded feature expression method (such as linkage vector or graph embedding) is used, the nearest standard situation type is determined by the minimum cluster center distance criterion. This sequence reflects the evolution of the dominant growth situation in the region within a given time scale, such as the temporal evolution of labels such as "co-expansion," "morphological differentiation," and "structural imbalance." The system further extracts the following key behavioral characteristic indicators from the above label sequences, including label switching frequency, which measures the volatility of the situation state at a specific time scale and reflects the stability or mutation tendency of regional behavior; dominant situation duration, which refers to the length of time that the region continuously maintains the same situation and reflects the ability to maintain the main behavior; label switching path, which records the typical change pattern of the situation on the time axis, such as "coordination" → "expansion" → "imbalance", etc.; regional state synchronization, which evaluates whether multiple spatial regions have consistent state switching within the same time period, reflecting the linkage or coordination of the overall spatial system.

[0152] S42. Modeling the micro-situation evolution trend based on the situation deconstruction data to obtain a micro-situation evolution model;

[0153] In one embodiment, the system uses the historical situation evolution sequence data of each spatial region to carry out modeling analysis to support trend inference of future situations. The regional situation label set is standardized and a state space set is constructed. For example, different situation behaviors such as "cooperation", "differentiation", "stagnation" and "jump" are marked as a finite state set. Based on the label transfer path of adjacent moments in the historical situation sequence, the transition frequency between different states is counted and normalized to form a state transition probability matrix. Each element in the matrix represents the conditional probability of the system transferring from the current state to the target state. Given the probability distribution vector of the current state, the system recursively infers the state probability distribution of future moments by matrix multiplication, and then infers the most likely situation trend at the next moment or after several moments to obtain a micro-situation evolution model.

[0154] S43. Identify the abnormal morphological trends of the micro-trend evolution model to obtain growth trend prediction data.

[0155] In one embodiment, the system uses predicted multi-step future trends as input and designs multiple anomaly discrimination rules to identify potential atypical behavioral transitions or areas of dramatic structural imbalance, providing early warning and trend-based anomaly alerts. The system detects whether there are dramatic transition path patterns in the region's future state label sequence, paying particular attention to transition sequences such as transitions from a "coordinated stability" state to an "asynchronous jump" state, further developing into "structural imbalance" or "dramatic contraction." If the system predicts that the cumulative probability of such a path occurring within three consecutive time steps exceeds a set threshold (e.g., 60%), the region is identified as a high-risk transition area and marked as exhibiting an anomalous structural evolution trend. The system measures the uncertainty of the state probability distribution at each predicted time step, using the information entropy metric to measure its volatility. Higher entropy values ​​indicate greater dispersion and uncertainty in the region's future state predictions. The probability value of each predicted state is used as input, logarithmically transformed, and weighted summed to form an overall uncertainty score. If a region exhibits a significant upward trend in entropy over multiple consecutive time steps, the system identifies it as experiencing a "trend disturbance" risk—meaning the direction of future state change is highly uncertain, making it a potentially unstable region. The system leverages the spatial adjacency structure between regions to collaboratively analyze the situational forecasts for adjacent regions. If the system identifies multiple geographically adjacent regions as being on the verge of transition within the same time window (i.e., the probability of each state transition is above a set threshold, such as 50%), the system identifies these spatial regions as experiencing a trend of interconnected risk transmission.

[0156] Optionally, the present application further provides a flexible sensor-based vegetable growth monitoring system for executing the flexible sensor-based vegetable growth monitoring method described above, wherein the flexible sensor-based vegetable growth monitoring system comprises:

[0157] The flexible sensing perception and deformation variable calculation module is used to collect sensor resistance through the flexible sensor to obtain sensor resistance data, and calculate the deformation variable based on the sensor resistance data to obtain deformation variable data;

[0158] A multi-point growth behavior analysis module is used to perform multi-point growth behavior processing on the deformation variable data to obtain multi-point growth behavior data;

[0159] A micro-situation modeling module is used to perform micro-morphological growth situation modeling based on multi-point growth behavior data to obtain a growth situation model;

[0160] The growth trend prediction module is used to perform growth trend prediction on the growth trend model to obtain growth trend prediction data.

[0161] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, the scope of the present invention is limited by the attached application documents rather than the above description, and it is intended that all changes that fall within the meaning and scope of equivalent elements of the application documents are included in the present invention.

[0162] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A vegetable growth monitoring method based on flexible sensors, characterized in that: The method comprises: S1. Collecting sensor resistance using a flexible sensor at a preset first position on the vegetable to obtain first sensor resistance data; collecting sensor resistance using a flexible sensor at a preset second position on the vegetable to obtain second sensor resistance data, wherein the first position and the second position are different positions; constructing a resistance space graph based on the first sensor resistance data and the second sensor resistance data to obtain sensor resistance data; and calculating a deformation variable based on the sensor resistance data to obtain deformation variable data. S2. performing multi-point growth behavior processing on the deformation variable data to obtain multi-point growth behavior data; S3. Modeling the micromorphological growth situation based on the multi-point growth behavior data to obtain a growth situation model; S4. performing growth trend prediction on the growth trend model to obtain growth trend prediction data; The step of generating the first position data and the second position data includes: Get vegetable image data; Extracting the petiole extension axis according to the vegetable image data to obtain petiole extension axis data; Calculate the main stem area center point based on the petiole extension axis data to obtain the main stem area center point data; A tension mapping model is constructed according to the center of gravity data of the main stem area to obtain a tension mapping model; Position the extreme tension point according to the tension mapping model to obtain the extreme tension point data; The position data corresponding to the maximum value data in the tension extreme difference point data is determined as the first position data, and the position data corresponding to the minimum value data in the tension extreme difference point data is determined as the second position data.

2. The method according to claim 1, characterized in that The deformation data includes single deformation data and linkage deformation data. The deformation calculation includes: According to the sensor resistance data, a resistance change vector field is constructed to obtain resistance change vector field data; Performing strain inversion mapping calculation based on the resistance change vector field data to obtain strain inversion data, wherein the strain inversion mapping calculation is a calculation process of converting the resistance change information into the strain amount at each spatial position on the target area based on the resistance change vector field data using a preset resistance-strain mapping function; Calculate the monomer deformation variable according to the strain inversion data to obtain the monomer deformation variable data; The linkage deformation variable data is processed according to the monomer deformation variable data to obtain the linkage deformation variable data.

3. The method according to claim 1, characterized in that S2 include: Extracting linkage behavior features from the shape variable data to obtain linkage behavior feature data; Divide the micro-behavior segments according to the linkage behavior feature data to obtain micro-behavior segment data; A growth behavior map is constructed based on the microscopic behavior fragment data to obtain multi-point growth behavior data.

4. The method according to claim 3, characterized in that The linkage behavior feature extraction includes: Performing cross-point deformation coupling processing on the deformation variable data to obtain cross-point deformation coupling data; Perform local tension collaborative clustering based on cross-point deformation coupling data to obtain local tension collaborative data; The collaborative deformation features of the local tension collaborative data are extracted to obtain the linkage behavior feature data.

5. The method according to claim 3, characterized in that The micro-behavior segmentation includes: Calculate the collaborative segment boundary points based on the linkage behavior feature data to obtain the collaborative segment boundary point data; Perform morphologically similar segment matching on the linkage behavior feature data based on the collaborative segment boundary point data to obtain morphologically similar segment data; Adjacent time series segmentation is performed based on morphological similarity segmentation data to obtain micro-behavior segment data.

6. The method according to claim 1, characterized in that S3 includes: Divide the structural area according to the multi-point growth behavior data to obtain the structural area data; Perform behavior-dominant clustering on the structural region data to obtain behavior-dominant feature data; Generate regional behavior embedding vectors based on the behavior-dominant feature data to obtain regional behavior representation data; Construct a micro-situation propagation map based on regional behavior representation data to obtain micro-situation propagation map data; The micromorphological growth state is judged on the micro-situation propagation map data to obtain the growth state model.

7. The method according to claim 1, characterized in that S4 include: Perform multi-scale temporal deconstruction of the growth trend model to obtain trend deconstruction data; Based on the situation deconstruction data, the micro-situation evolution trend model is modeled to obtain the micro-situation evolution model; The micro-trend evolution model is used to identify morphological abnormal trends and obtain growth trend prediction data.

8. A vegetable growth monitoring system based on flexible sensors, characterized in that: For executing the vegetable growth monitoring method based on flexible sensors according to claim 1, the vegetable growth monitoring system based on flexible sensors comprises: The flexible sensing perception and deformation variable calculation module is used to collect sensor resistance through the flexible sensor to obtain sensor resistance data, and calculate the deformation variable based on the sensor resistance data to obtain deformation variable data; A multi-point growth behavior analysis module is used to perform multi-point growth behavior processing on the deformation variable data to obtain multi-point growth behavior data; A micro-situation modeling module is used to perform micro-morphological growth situation modeling based on multi-point growth behavior data to obtain a growth situation model; The growth trend prediction module is used to perform growth trend prediction on the growth trend model to obtain growth trend prediction data.

Images