Intelligent monitoring system for growth conditions of afforestation and greening seedlings based on deep learning

Abstract

The invention relates to the technical field of forestry intelligent monitoring, and particularly discloses an intelligent monitoring system for the growth condition of afforestation and greening nursery stocks based on deep learning, which is characterized in that multi-modal growth data of the nursery stocks are acquired through a multi-spectral imaging sensor, a three-dimensional laser scanning sensor and an environment monitoring sensor, and a standardized data set is formed through space-time alignment processing; extracting morphological structure and spectral response features through spatial domain and frequency domain parallel analysis, and generating multi-dimensional feature representation through cross-modal fusion; converting the parameters into growth state parameters by utilizing a feature recombination and space mapping technology; identifying an abnormal growth mode through time sequence dynamic analysis; and finally, adaptively adjusting the working parameters of the sensor according to the abnormal type to form closed-loop monitoring. The problem that an existing monitoring system cannot autonomously optimize a monitoring strategy according to the abnormal state is solved, and accurate monitoring and intelligent regulation and control of the nursery stock growth condition are achieved.

Description

Technical Field

[0001] This invention relates to the field of intelligent forestry monitoring technology, specifically to an intelligent monitoring system for the growth status of afforestation and greening seedlings based on deep learning. Background Technology

[0002] Currently, monitoring the growth of afforestation seedlings mainly relies on a combination of manual inspections and single-sensor data collection, which suffers from low monitoring efficiency, limited data dimensions, and poor timeliness. Traditional methods are mostly based on fixed-point and fixed-time data collection, making it difficult to achieve continuous dynamic monitoring of seedling growth over a large area. Furthermore, the lack of multi-source data fusion and analysis capabilities makes it impossible to effectively distinguish the morphological characteristics of seedlings and symbiotic vegetation, resulting in insufficient accuracy in growth status assessment. Existing monitoring systems often rely on fixed thresholds or simple statistical analysis at the data processing level, making it difficult to adapt to the dynamic changes in seedling growth under complex site conditions, and even more difficult to achieve intelligent early warning and adaptive adjustment of monitoring strategies for abnormal growth patterns.

[0003] In existing technologies, seedling growth monitoring systems generally suffer from a disconnect between data processing and anomaly response; that is, they cannot dynamically adjust monitoring strategies based on identified abnormal growth patterns. When the system detects abnormalities such as stunted growth or pests and diseases, it continues to collect data using fixed sensor parameters, resulting in subsequent data that cannot be enhanced to capture abnormal characteristics, making it difficult to support accurate diagnosis and intervention decisions. This separation between the "identification" and "control" stages means that while the monitoring system can detect problems, it cannot autonomously optimize the monitoring process, thus limiting its practical application in precision forestry. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent monitoring system for the growth status of afforestation seedlings based on deep learning, so as to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A deep learning-based intelligent monitoring system for the growth status of afforestation seedlings includes:

[0007] The multi-source data acquisition and preprocessing module collects multimodal growth data of seedlings through multiple smart sensors deployed in the monitoring area, performs spatiotemporal alignment processing on the multimodal growth data, and generates a standardized multi-source data set.

[0008] The multidimensional feature parsing and fusion module is used to perform multi-level feature parsing and cross-modal fusion on standardized multi-source datasets. By decomposing and reconstructing the datasets in different feature dimensions, it generates a unified multidimensional feature representation.

[0009] The growth parameter conversion module converts the unified multidimensional feature representation into seedling growth state parameters through feature recombination and spatial mapping technology;

[0010] The abnormal growth identification module is used to perform dynamic analysis of seedling growth status parameters based on time-series changes. By establishing the correlation between parameter change trends and growth status, it identifies abnormal growth patterns in seedlings.

[0011] The intelligent monitoring and control module identifies abnormal growth patterns in seedlings, analyzes the correlation between parameter change trends and growth status, establishes mapping rules between parameter change patterns and intelligent sensor configurations, and automatically configures the working parameter combination of intelligent sensors according to the type of abnormal pattern, thereby achieving precise monitoring of the growth status of afforestation and greening seedlings.

[0012] As a further aspect of the present invention: the spatiotemporal alignment processing of the multimodal growth data to generate a standardized multi-source data set specifically includes:

[0013] Acquire seedling spectral data, seedling point cloud data, and microenvironment data collected by intelligent sensors to form a multi-source data set;

[0014] A spatiotemporal registration method based on feature point matching is adopted, taking the key nodes of the seedling trunk as spatial reference points to unify the data from different sensors into the same coordinate system, and simultaneously using the clock signal of the acquisition device as a reference for timestamp synchronization.

[0015] Through data standardization, the registered multi-source data is converted into a standard data set with a unified scale and format, thus completing the spatiotemporal alignment process;

[0016] The intelligent sensors include: a multispectral imaging sensor, a three-dimensional laser scanning sensor, and a microenvironment monitoring sensor. The microenvironment monitoring sensor includes: an air temperature and humidity sensor, a soil temperature and humidity sensor, and a photosynthetically active radiation sensor.

[0017] As a further aspect of the present invention: the step of decomposing and reconstructing the data set according to different feature dimensions to generate a unified multidimensional feature representation specifically includes:

[0018] Standardized multi-source datasets are input into the spatial domain analysis channel and the frequency domain analysis channel for parallel processing. Morphological and structural features of seedlings are extracted in the spatial domain analysis channel, and spectral response features of seedlings are extracted in the frequency domain analysis channel.

[0019] The spatial morphological and structural features and the frequency domain spectral response features are fused across modes, and a fused feature matrix is ​​generated through feature cross-enhancement processing.

[0020] The fusion feature matrix is ​​dimensionally normalized and feature optimization is performed to eliminate redundant information, retain key feature components, and form a unified multidimensional feature representation.

[0021] As a further aspect of the present invention: the conversion of the unified multidimensional feature representation into seedling growth state parameters through feature recombination and spatial mapping technology specifically includes:

[0022] The unified multidimensional feature representations are grouped and reorganized according to plant physiological characteristics to form feature subsets corresponding to morphological structure, physiological function and environmental response respectively;

[0023] Based on the physiological characteristics of seedlings at different growth stages, a spatial mapping relationship between feature subsets and growth state parameters is established, and each feature subset is converted into the corresponding growth state parameters through nonlinear transformation.

[0024] Normalization and confidence assessment were performed on each growth state parameter to generate a set of seedling growth state parameters.

[0025] As a further aspect of the present invention: the conversion of each feature subset into corresponding growth state parameters through nonlinear transformation specifically includes:

[0026] A time series of vegetation indices was established based on the physiological characteristics of seedlings at different growth stages, and a characteristic evolution trajectory reflecting growth patterns was constructed.

[0027] A dynamic reference surface is generated based on the feature evolution trajectory. A subset of features is projected onto the dynamic reference surface, and the nonlinear adjustment of the feature values ​​is achieved through surface deformation.

[0028] Based on the distribution of the projected eigenvalues ​​on the dynamic reference surface, the corresponding seedling growth state parameters are calculated.

[0029] As a further aspect of the present invention: the identification of abnormal patterns in seedling growth specifically includes:

[0030] A baseline growth trajectory reflecting normal growth patterns is constructed, and continuously collected seedling growth status parameters are matched and compared with the baseline growth trajectory.

[0031] The growth state parameter sequence is decomposed to extract fluctuation feature components at different time scales;

[0032] By analyzing the correspondence between each fluctuation characteristic component and typical abnormal growth patterns, abnormal patterns that deviate from the normal growth trajectory can be identified.

[0033] As a further aspect of the present invention: the decomposition of the growth state parameter sequence to extract fluctuation feature components at different time scales specifically includes:

[0034] Multiple time observation windows with different period lengths are constructed to synchronously divide the growth state parameter sequence into multiple corresponding time scale subsequences.

[0035] Sliding variance was calculated for each time scale subsequence to obtain characteristic curves reflecting the intensity of fluctuations at each scale.

[0036] By performing extreme point detection and envelope analysis on characteristic curves at various scales, characteristic components representing periodic fluctuation patterns are extracted.

[0037] As a further aspect of the present invention: the extraction of characteristic components representing the periodic fluctuation pattern by performing extreme point detection and envelope analysis on characteristic curves at various scales specifically includes:

[0038] Local maxima and minima are located on characteristic curves at various scales to form a sequence of extreme points;

[0039] Connect all the maxima and minima respectively to construct the upper and lower envelopes;

[0040] By calculating the vertical distance sequence between the upper and lower envelopes, feature components representing the fluctuation amplitude are extracted. At the same time, by analyzing the time interval distribution between extreme points, feature components representing the fluctuation period are extracted.

[0041] As a further aspect of the present invention: the automatic configuration of the working parameter combination of the intelligent sensor according to the type of abnormal mode to achieve accurate monitoring of the growth status of afforestation seedlings specifically includes:

[0042] A predefined configuration strategy library is established based on the classification of abnormal modes, and a correspondence is established between different types of abnormal modes and corresponding sensor parameter configuration schemes.

[0043] Based on the dynamic characteristics of parameter change trends, the optimal configuration scheme is selected from the predefined configuration strategy library, and the parameters are optimized and adjusted according to the current environmental conditions.

[0044] Generate sensor control commands to dynamically adjust the band combination, acquisition frequency, and resolution parameters of the multispectral sensor in the smart sensor, thereby achieving precise monitoring of the growth status of afforestation seedlings.

[0045] The beneficial effects of this invention are:

[0046] (1) A spatiotemporally aligned multi-source data set was established through the collaborative acquisition of multispectral imaging, three-dimensional laser scanning and environmental monitoring sensors; by adopting technologies such as parallel processing of spatial and frequency domains and cross-modal feature fusion, the morphological structure, physiological function and environmental response characteristics of seedlings can be accurately extracted from complex backgrounds.

[0047] (2) Anomaly pattern recognition based on time-series dynamic analysis and adaptive adjustment of sensor parameters form a complete monitoring closed loop. The system can automatically adjust the band combination and acquisition frequency of the multispectral sensor according to the identified abnormal patterns such as growth stagnation and growth acceleration, realizing the transformation from "passive monitoring" to "active sensing". This intelligent control mechanism improves monitoring efficiency and reduces energy consumption by optimizing the sensor working mode, achieving a balance between accurate monitoring and resource optimization. Attached Figure Description

[0048] The invention will now be further described with reference to the accompanying drawings.

[0049] Figure 1 This is a schematic diagram of the system of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Please see Figure 1 As shown, this invention is an intelligent monitoring system for the growth status of afforestation seedlings based on deep learning, comprising:

[0052] The multi-source data acquisition and preprocessing module collects multimodal growth data of seedlings through multiple smart sensors deployed in the monitoring area, performs spatiotemporal alignment processing on the multimodal growth data, and generates a standardized multi-source data set.

[0053] The multidimensional feature parsing and fusion module is used to perform multi-level feature parsing and cross-modal fusion on standardized multi-source datasets. By decomposing and reconstructing the datasets in different feature dimensions, it generates a unified multidimensional feature representation.

[0054] The growth parameter conversion module converts the unified multidimensional feature representation into seedling growth state parameters through feature recombination and spatial mapping technology;

[0055] The abnormal growth identification module is used to perform dynamic analysis of seedling growth status parameters based on time-series changes. By establishing the correlation between parameter change trends and growth status, it identifies abnormal growth patterns in seedlings.

[0056] The intelligent monitoring and control module identifies abnormal growth patterns in seedlings, analyzes the correlation between parameter change trends and growth status, establishes mapping rules between parameter change patterns and intelligent sensor configurations, and automatically configures the working parameter combination of intelligent sensors according to the type of abnormal pattern, thereby achieving precise monitoring of the growth status of afforestation and greening seedlings.

[0057] In the multi-source data acquisition and preprocessing module, multimodal growth data of seedlings are collected by various intelligent sensors deployed within the monitoring area. A multispectral imaging sensor acquires spectral reflectance data of the seedlings by receiving visible light and near-infrared electromagnetic waves reflected from the seedling canopy. A three-dimensional laser scanning sensor measures the spatial coordinates of various points on the seedling surface by emitting a laser beam and receiving the reflected signal, forming three-dimensional point cloud data of the seedlings. An air temperature and humidity sensor measures the temperature and humidity parameters of the surrounding air using a capacitive sensing element; a soil temperature and humidity sensor measures the moisture content and temperature of the root zone soil using the time-domain reflectometry principle; and a photosynthetically active radiation sensor measures the photosynthetic photon flux density in the 400-700 nanometer wavelength band using a silicon photodiode.

[0058] Spatiotemporal registration was performed on the collected multi-source data. For spatial registration, at least three key nodes with distinct morphological features on the main trunk of the seedling were selected as spatial reference points. A unified coordinate system was established by calculating the spatial coordinates of these reference points in the 3D point cloud data. Pixels in the multispectral images were mapped to this coordinate system through perspective transformation, and the acquisition locations of the microenvironment data were registered to the same coordinate system using spatial interpolation. For time synchronization, the acquisition time of the 3D laser scanning sensor was used as the reference time. By calculating the difference between the acquisition time of each sensor and the reference time, time offset compensation was performed on the multispectral image data and microenvironment data to ensure that all data had a unified time reference.

[0059] The registered multi-source data underwent standardization processing. For spectral data, the original digital quantization values ​​were converted into reflectance values. The conversion process employed an empirical linear calibration method, establishing a linear relationship between the digital quantization values ​​and reflectance by measuring the reflectance of a standard whiteboard. For point cloud data, the spatial coordinate values ​​were uniformly transformed to a local coordinate system with the root-stem junction of the seedling as the origin, and the point cloud density was resampled to maintain a density of 10–15 points per square centimeter. For microenvironment data, the measured values ​​of air temperature and humidity, soil temperature and humidity, and photosynthetically active radiation were normalized to the range of 0–1. The normalization formula was: the measured value minus the historical minimum value, divided by the difference between the historical maximum and minimum values.

[0060] The standardized multi-source data is stored in a unified data format to form a standardized multi-source data set. The data storage adopts a hierarchical structure: the first layer contains spatial coordinate information, the second layer contains spectral feature information, and the third layer contains environmental parameter information. The data in each layer are linked together through a spatiotemporal index.

[0061] The multidimensional feature parsing and fusion module specifically includes:

[0062] Standardized multi-source datasets are input into the spatial domain analysis channel and the frequency domain analysis channel for parallel processing. In the spatial domain analysis channel, 3D point cloud data is processed to extract the morphological and structural features of the seedlings. Specifically, this process includes: triangulating the point cloud data to construct a 3D mesh model of the seedling surface; calculating the surface area, volume, and other geometric parameters of the seedlings based on this mesh model; extracting the spatial orientation information of the seedling trunk and branches by calculating the normal vector distribution characteristics of the point cloud data; and extracting the external contour features of the seedling canopy using the alpha-shape algorithm to obtain parameters such as the maximum projected area and canopy height. In the frequency domain analysis channel, multispectral data is processed to extract the spectral response features of the seedlings. Specifically, this process includes: performing a discrete Fourier transform on the spectral data to convert the spectral sequence from the spatial domain to the frequency domain; calculating the energy distribution characteristics of the spectral curve in the frequency domain and extracting the energy ratio of low-frequency and high-frequency components; analyzing the detailed features of the spectral curve at different scales using wavelet transform to obtain a multi-scale feature representation of the spectral response; and calculating vegetation index features such as the normalized difference vegetation index (NDVI) and photochemical vegetation index based on the spectral data.

[0063] This method performs cross-modal fusion of spatial morphological structural features and frequency domain spectral response features. The specific implementation process includes: standardizing both types of features to ensure they have the same numerical range; constructing a feature cross-enhancement matrix, where rows correspond to spatial features and columns to frequency features; determining the correlation strength of different feature combinations by calculating the mutual information values ​​between spatial and frequency features; weighting the feature combinations based on the mutual information values, using a normalized mutual information method to calculate the weights, i.e., dividing each mutual information value by the sum of all mutual information values; and organizing the weighted feature combinations in matrix form to generate a fused feature matrix. In the feature cross-enhancement process, for each combination of spatial and frequency features, the product and difference terms are calculated, and these cross terms are added to the fused feature matrix as new feature dimensions.

[0064] The fused feature matrix undergoes dimensionality normalization and feature optimization. The specific process includes: calculating the variance of each feature in the fused feature matrix, setting a variance threshold of 30% of the average variance of all features, and discarding features with variances below this threshold as redundant features; employing a feature optimization method based on correlation coefficients, calculating the Pearson correlation coefficient between features, and retaining the feature with the larger variance for feature pairs with an absolute correlation coefficient greater than 0.8, while discarding the feature with the smaller variance; performing principal component analysis on the optimized features, projecting the features into a new orthogonal basis space, and selecting the top k principal components with a cumulative contribution rate of 85% as the final feature representation; and sorting the selected principal component features according to their contribution rate to form a unified multidimensional feature representation.

[0065] The generated multidimensional feature representations undergo quality assessment. The assessment process includes: calculating the signal-to-noise ratio (SNR) of the feature representations, which is the ratio of the mean to the standard deviation of the feature values; checking for outliers in the feature representations using the three-standard-deviation principle, where feature values ​​differing from the mean by more than three standard deviations are considered outliers; replacing detected outliers with the weighted average of neighboring feature values, where the weights are determined by the distance between feature values, with closer neighboring features receiving greater weight. Multidimensional feature representations that pass the quality assessment will be used for subsequent growth state parameter conversion processing.

[0066] In the growth parameter conversion module, the unified multidimensional feature representations are grouped and reorganized according to plant physiological characteristics. The grouping process is based on the correspondence between features and plant physiological functions, dividing the features into three main subsets. The morphological structure feature subset includes feature parameters that directly reflect plant morphology, such as seedling height, crown diameter, number of branches, and leaf area. The physiological function feature subset includes feature parameters that reflect plant physiological activities, such as chlorophyll content, photosynthetic rate, and transpiration efficiency. The environmental response feature subset includes feature parameters that reflect the interaction between plants and the environment, such as temperature adaptability, water use efficiency, and light responsivity. Within each feature subset, the feature parameters are ordered according to their biological significance, and the correlations between the feature parameters are established.

[0067] Based on the physiological characteristics of seedlings at different growth stages, a spatial mapping relationship between feature subsets and growth state parameters was established. For each growth stage—seedling, growth period, and maturity—corresponding spatial mapping relationships were established. In the seedling stage, the focus was on establishing the mapping relationship between morphological and structural features and basic growth parameters; in the growth period, the mapping relationship between physiological functional features and metabolic activity parameters was established; and in the maturity period, the mapping relationship between environmental response features and adaptive parameters was established. The establishment of spatial mapping relationships employed a multidimensional scaling analysis method, determining the optimal mapping path by calculating the Euclidean distance between feature subsets and growth state parameters.

[0068] Nonlinear transformation converts each feature subset into corresponding growth state parameters. The nonlinear transformation process includes the following specific steps: A vegetation index time series is constructed based on the physiological characteristics of seedlings at different growth stages. This series consists of vegetation index observations over 30 consecutive growth days. Based on the vegetation index time series, a feature evolution trajectory reflecting growth patterns is constructed using spline interpolation. This trajectory characterizes the changes in seedlings throughout their complete growth cycle. A dynamic reference surface is generated based on the feature evolution trajectory. The surface is generated using a non-uniform rational B-spline method, and dynamic updates are achieved by adjusting control points. The feature subsets are projected onto the dynamic reference surface. The projection process is achieved by calculating the distance between the feature value and the nearest point on the surface. Nonlinear adjustment of the feature values ​​is achieved through surface deformation. The deformation is calculated based on the degree of difference between the feature value and the reference value. Based on the distribution of the projected feature values ​​on the dynamic reference surface, the corresponding seedling growth state parameters, including growth rate, health status, and environmental adaptability, are calculated.

[0069] All growth state parameters were normalized. The normalization process employed a minimum-maximum normalization method, converting each parameter to a range of 0 to 100. For growth rate parameters, corresponding maximum and minimum values ​​were set according to the seedling species; for health status parameters, 0 represented mortality and 100 represented optimal health; for environmental adaptability parameters, 0 represented complete maladaptation and 100 represented complete adaptation. The normalization formula was: parameter value minus the minimum value, divided by the difference between the maximum and minimum values, and then multiplied by 100.

[0070] The normalized growth state parameters were assessed for confidence. The confidence assessment was based on the following factors: the completeness of the feature data, the strength of the association between features and parameters, and the stability of parameter calculation. A weighted average method was used to calculate the confidence score, with a weight of 0.4 for data completeness, 0.35 for association strength, and 0.25 for calculation stability. A confidence threshold of 70% was set; parameters below this threshold required recalculation. The recalculation used a backup mapping relationship, which was an auxiliary calculation path established from historical data. Parameters that passed the confidence assessment formed a set of seedling growth state parameters, which will be used for subsequent identification of abnormal growth patterns.

[0071] In the abnormal growth identification module, a baseline growth trajectory reflecting normal growth patterns is constructed. This trajectory is built using historical normal growth data, selecting observation data from three consecutive growing seasons of the same tree species under the same site conditions as training samples. The baseline growth trajectory includes the normal variation range of various growth status parameters of the seedling throughout its complete growth cycle. The percentile method is used to determine the normal fluctuation range of the parameters, where the normal range is defined as the numerical range between the 10th and 90th percentiles. When continuously collected seedling growth status parameters are matched and compared with the baseline growth trajectory, a dynamic time warping algorithm is used to calculate the similarity between the current growth trajectory and the baseline growth trajectory. The similarity threshold is set at 85%; growth trajectories below this threshold are considered abnormal trajectories.

[0072] The growth state parameter sequence is decomposed to extract fluctuation characteristic components at different time scales. Specifically, this involves constructing three time observation windows with different period lengths: 30 days, 60 days, and 90 days, simultaneously dividing the growth state parameter sequence into corresponding three time-scale subsequences. Moving variance is calculated for each time-scale subsequence, with the moving window size set to 1 / 10 of the corresponding time observation window length, i.e., 3 days, 6 days, and 9 days respectively. The moving variance is calculated by determining the variance of the parameters within each moving window to obtain characteristic curves reflecting the fluctuation intensity at each scale. By performing extreme point detection and envelope analysis on the characteristic curves at each scale, characteristic components representing the periodic fluctuation pattern are extracted.

[0073] In the specific implementation of extreme point detection and envelope analysis, local maxima and minima are first located on the characteristic curves at various scales. Extreme point identification employs a finite difference method, precisely locating the extreme point by calculating the first and second differences of the characteristic curves, requiring the curvature value at the extreme point to be greater than a set threshold of 0.05. All identified extreme points are arranged in chronological order to form an extreme point sequence. Then, all maxima and minima are connected to construct upper and lower envelopes. Cubic spline interpolation is used to construct the envelopes, ensuring they are smooth and pass through all extreme points. By calculating the vertical distance sequence between the upper and lower envelopes, feature components representing fluctuation amplitude are extracted. Simultaneously, by analyzing the time interval distribution between extreme points, feature components representing fluctuation period are extracted. The fluctuation amplitude is calculated using the average of the vertical distance sequence, and the fluctuation period is calculated using the mode of the time intervals between extreme points.

[0074] By analyzing the correspondence between various fluctuation characteristic components and typical abnormal growth patterns, abnormal patterns deviating from the normal growth trajectory are identified. Typical abnormal growth patterns include growth stagnation, accelerated growth, and periodic fluctuation anomalies. For the growth stagnation pattern, the main identifying feature is that the fluctuation amplitude at all time scales is less than 50% of the normal value; for the accelerated growth pattern, the identifying feature is a significant increase in short-term fluctuation amplitude, with an increase exceeding 150% of the normal value; for the periodic fluctuation anomaly pattern, the identifying feature is that the main fluctuation period differs from the normal period by more than 20%. Determining an abnormal pattern requires that the characteristic components at at least two time scales simultaneously meet the abnormal conditions to improve the identification accuracy.

[0075] The identified anomaly patterns were verified and classified. The verification process employed a multi-source data cross-validation method, comparing the identification results with concurrently collected multispectral and point cloud data. Anomaly patterns were categorized into three severity levels: mild, moderate, and severe. The grading criteria were based on three indicators: anomaly duration, anomaly magnitude, and impact range, using a weighted scoring method for comprehensive evaluation. The weights for the three indicators were 0.3, 0.4, and 0.3, respectively. A total score below 40 indicated mild anomaly, 40-70 indicated moderate anomaly, and above 70 indicated severe anomaly. The verified and classified anomaly patterns will be used for subsequent adjustments to monitoring strategies.

[0076] In the intelligent monitoring and control module, a predefined configuration strategy library is first established based on the classification of abnormal modes. Abnormal modes are categorized into three main types according to growth characteristics: stagnant growth, accelerated growth, and periodic fluctuation anomalies. The predefined configuration strategy library contains sensor parameter configuration schemes corresponding to each anomaly type. The configuration scheme for stagnant growth anomalies focuses on enhancing the acquisition capabilities of the multispectral sensor in the red and near-infrared bands, setting the acquisition frequency to 3 times per day and the spatial resolution to 0.5 meters. The configuration scheme for accelerated growth anomalies mainly increases the acquisition density of the 3D laser scanning sensor, setting the point cloud density to 2000 points per square meter and the acquisition frequency to twice per day. The configuration scheme for periodic fluctuation anomalies adjusts the parameters of both the multispectral sensor and the 3D laser scanning sensor simultaneously, increasing the band combination of the multispectral sensor to 10 specific bands and expanding the scanning angle of the 3D laser scanning sensor to 120 degrees.

[0077] Based on the dynamic characteristics of parameter change trends, the optimal configuration scheme is selected from a predefined configuration strategy library. The dynamic characteristics of parameter change trends are characterized by calculating the rate of change and acceleration of change of growth state parameters over the past 7 days. The rate of change is calculated using linear regression, and the acceleration of change is calculated using quadratic differentiation. When selecting the optimal configuration scheme, the matching degree between the current anomaly pattern and various anomaly types in the predefined configuration strategy library is first calculated. The matching degree is calculated based on the similarity of anomaly features, using Euclidean distance as a metric; the smaller the distance value, the higher the matching degree. The two configuration schemes with the highest matching degrees are selected as candidate schemes, and then the parameters are optimized and adjusted according to the current environmental conditions. Environmental conditions include three factors: light intensity, temperature, and soil moisture. The optimization adjustment uses a weighted average method, with light intensity having a weight of 0.4, temperature having a weight of 0.3, and soil moisture having a weight of 0.3.

[0078] The system generates sensor control commands to dynamically adjust the operating parameters of the intelligent sensor. The control commands are generated using a command encoding method, converting the adjustment parameters into 16-bit binary codes. The first 4 bits represent the sensor type, the middle 6 bits represent the parameter type, and the last 6 bits represent the parameter value. Band combination adjustment for the multispectral sensor is achieved through the band selection bit in the control command, supporting a maximum of 16 band combinations. The acquisition frequency adjustment ranges from once per day to six times per day, and the resolution parameter adjustment ranges from 0.1 meters to 1.0 meter. Point cloud density adjustment for the 3D laser scanning sensor is achieved by controlling the laser emission frequency, with an adjustment range from 1000 points per square meter to 5000 points per square meter. Scanning angle adjustment is achieved by controlling the scanning motor speed, with an adjustment range from 60 degrees to 150 degrees.

[0079] The adjusted sensor performance was evaluated and optimized. The evaluation process took place within 24 hours of the sensor parameter adjustment, assessing the effectiveness of the adjustment by analyzing the quality and completeness of the newly acquired data. Data quality was evaluated using three metrics: signal-to-noise ratio (SNR), contrast, and sharpness. The SNR threshold was set at 20 dB, the contrast threshold at 0.7, and the sharpness threshold at 0.8. If any metric fell below the threshold, the parameter optimization process was initiated. The optimization process employed gradient descent, using the data quality metrics as the optimization target and the sensor operating parameters as the optimization variables, to find the optimal parameter combination through iterative calculation. Each parameter adjustment was controlled within 10% of the current value to ensure a smooth transition in the sensor's operating state. After evaluation and optimization, the system entered a stable monitoring state, continuously monitoring seedling growth and recording monitoring data.

[0080] The working principle of this invention is as follows: A multi-source data acquisition and preprocessing module utilizes multispectral imaging sensors, 3D laser scanning sensors, and environmental monitoring sensors deployed in the monitoring area to collect multimodal growth data of seedlings. This data is then spatiotemporally aligned and standardized to generate a standardized multi-source data set. Subsequently, a multi-dimensional feature analysis and fusion module performs parallel spatial and frequency domain processing on the standardized data, generating a unified multi-dimensional feature representation through cross-modal fusion and feature optimization. A growth parameter conversion module, based on plant physiological characteristics, groups and reorganizes the multi-dimensional feature representations, converting them into seedling growth state parameters with clear physical meaning through nonlinear transformation and spatial mapping techniques. A growth anomaly identification module constructs a baseline growth trajectory, performs time-series analysis and fluctuation feature extraction on growth parameters, and identifies abnormal patterns deviating from normal growth patterns. Finally, an intelligent monitoring and control module dynamically configures the working parameter combinations of intelligent sensors based on the type of abnormal pattern, through a predefined configuration strategy library and parameter optimization adjustments, achieving precise monitoring and early warning of specific growth conditions, forming a complete closed-loop monitoring system from data acquisition to intelligent control.

[0081] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A deep learning-based intelligent monitoring system for the growth status of afforestation seedlings, characterized in that, include: The multi-source data acquisition and preprocessing module collects multimodal growth data of seedlings through multiple smart sensors deployed in the monitoring area, performs spatiotemporal alignment processing on the multimodal growth data, and generates a standardized multi-source data set. The multidimensional feature parsing and fusion module is used to perform multi-level feature parsing and cross-modal fusion on standardized multi-source datasets. By decomposing and reconstructing the datasets in different feature dimensions, it generates a unified multidimensional feature representation. The growth parameter conversion module converts the unified multidimensional feature representation into seedling growth state parameters through feature recombination and spatial mapping technology; The abnormal growth identification module is used to perform dynamic analysis of seedling growth status parameters based on time-series changes. By establishing the correlation between parameter change trends and growth status, it identifies abnormal growth patterns in seedlings. The intelligent monitoring and control module identifies abnormal growth patterns in seedlings, analyzes the correlation between parameter change trends and growth status, establishes mapping rules between parameter change patterns and intelligent sensor configurations, and automatically configures the working parameter combination of intelligent sensors according to the type of abnormal pattern, thereby achieving precise monitoring of the growth status of afforestation and greening seedlings.

2. The intelligent monitoring system for the growth status of afforestation seedlings based on deep learning according to claim 1, characterized in that, The process of performing spatiotemporal alignment on the multimodal growth data to generate a standardized multi-source data set specifically includes: Acquire seedling spectral data, seedling point cloud data, and microenvironment data collected by intelligent sensors to form a multi-source data set; A spatiotemporal registration method based on feature point matching is adopted, taking the key nodes of the seedling trunk as spatial reference points to unify the data from different sensors into the same coordinate system, and simultaneously using the clock signal of the acquisition device as a reference for timestamp synchronization. Through data standardization, the registered multi-source data is converted into a standard data set with a unified scale and format, thus completing the spatiotemporal alignment process; The intelligent sensors include: a multispectral imaging sensor, a three-dimensional laser scanning sensor, and a microenvironment monitoring sensor. The microenvironment monitoring sensor includes: an air temperature and humidity sensor, a soil temperature and humidity sensor, and a photosynthetically active radiation sensor.

3. The intelligent monitoring system for the growth status of afforestation seedlings based on deep learning according to claim 1, characterized in that, The process of decomposing and reconstructing the data set across different feature dimensions to generate a unified multidimensional feature representation specifically includes: Standardized multi-source datasets are input into the spatial domain analysis channel and the frequency domain analysis channel for parallel processing. Morphological and structural features of seedlings are extracted in the spatial domain analysis channel, and spectral response features of seedlings are extracted in the frequency domain analysis channel. The spatial morphological and structural features and the frequency domain spectral response features are fused across modes, and a fused feature matrix is ​​generated through feature cross-enhancement processing. The fusion feature matrix is ​​dimensionally normalized and feature optimization is performed to eliminate redundant information, retain key feature components, and form a unified multidimensional feature representation.

4. The intelligent monitoring system for the growth status of afforestation seedlings based on deep learning according to claim 1, characterized in that, The process of converting a unified multidimensional feature representation into seedling growth state parameters through feature recombination and spatial mapping techniques specifically includes: The unified multidimensional feature representations are grouped and reorganized according to plant physiological characteristics to form feature subsets corresponding to morphological structure, physiological function and environmental response respectively; Based on the physiological characteristics of seedlings at different growth stages, a spatial mapping relationship between feature subsets and growth state parameters is established, and each feature subset is converted into the corresponding growth state parameters through nonlinear transformation. Normalization and confidence assessment were performed on each growth state parameter to generate a set of seedling growth state parameters.

5. The intelligent monitoring system for the growth status of afforestation seedlings based on deep learning according to claim 4, characterized in that, The process of converting each feature subset into corresponding growth state parameters through nonlinear transformation specifically includes: A time series of vegetation indices was established based on the physiological characteristics of seedlings at different growth stages, and a characteristic evolution trajectory reflecting growth patterns was constructed. A dynamic reference surface is generated based on the feature evolution trajectory. A subset of features is projected onto the dynamic reference surface, and the nonlinear adjustment of the feature values ​​is achieved through surface deformation. Based on the distribution of the projected eigenvalues ​​on the dynamic reference surface, the corresponding seedling growth state parameters are calculated.

6. The intelligent monitoring system for the growth status of afforestation seedlings based on deep learning according to claim 1, characterized in that, The identification of abnormal patterns in seedling growth specifically includes: A baseline growth trajectory reflecting normal growth patterns is constructed, and continuously collected seedling growth status parameters are matched and compared with the baseline growth trajectory. The growth state parameter sequence is decomposed to extract fluctuation feature components at different time scales; By analyzing the correspondence between each fluctuation characteristic component and typical abnormal growth patterns, abnormal patterns that deviate from the normal growth trajectory can be identified.

7. The intelligent monitoring system for the growth status of afforestation seedlings based on deep learning according to claim 6, characterized in that, The decomposition of the growth state parameter sequence to extract fluctuation feature components at different time scales specifically includes: Multiple time observation windows with different period lengths are constructed to synchronously divide the growth state parameter sequence into multiple corresponding time scale subsequences. Sliding variance was calculated for each time scale subsequence to obtain characteristic curves reflecting the intensity of fluctuations at each scale. By performing extreme point detection and envelope analysis on characteristic curves at various scales, characteristic components representing periodic fluctuation patterns are extracted.

8. The intelligent monitoring system for the growth status of afforestation seedlings based on deep learning according to claim 7, characterized in that, The process of extracting characteristic components representing periodic fluctuations by performing extreme point detection and envelope analysis on characteristic curves at various scales specifically includes: Local maxima and minima are located on characteristic curves at various scales to form a sequence of extreme points; Connect all the maxima and minima respectively to construct the upper and lower envelopes; By calculating the vertical distance sequence between the upper and lower envelopes, feature components representing the fluctuation amplitude are extracted. At the same time, by analyzing the time interval distribution between extreme points, feature components representing the fluctuation period are extracted.

9. The intelligent monitoring system for the growth status of afforestation seedlings based on deep learning according to claim 1, characterized in that, The automatic configuration of intelligent sensor operating parameter combinations based on the type of abnormal mode enables precise monitoring of the growth status of afforestation seedlings, specifically including: A predefined configuration strategy library is established based on the classification of abnormal modes, and a correspondence is established between different types of abnormal modes and corresponding sensor parameter configuration schemes. Based on the dynamic characteristics of parameter change trends, the optimal configuration scheme is selected from the predefined configuration strategy library, and the parameters are optimized and adjusted according to the current environmental conditions. Generate sensor control commands to dynamically adjust the band combination, acquisition frequency, and resolution parameters of the multispectral sensor in the smart sensor, thereby achieving precise monitoring of the growth status of afforestation seedlings.

Images