Green plant flowering period synchronous control method and system based on dynamic perception

Abstract

The application discloses a kind of green plant flowering synchronous regulation method and system based on dynamic perception, the method includes: using visible light-near infrared dual-channel imaging equipment scanning obtains green plant canopy multispectral image sequence and environmental light spectrum distribution data, through extracting chlorophyll fluorescence intensity value, anthocyanin reflectivity and so on Feature generation plant physiological state vector, provide accurate data for subsequent regulation;With the aid of flowering convolution prediction model optimized by transfer learning, the flowering time window of each green plant is accurately predicted by fusing environmental light spectrum and historical phenology data;Then, through spatiotemporal superposition analysis, abnormal green plants in flowering period are screened out, and spectral control instructions are generated by calculating compensation spectrum parameters, and finally directional irradiation is carried out through rotatable array type spectral emitter.The application can improve the accuracy of synchronous regulation of green plant flowering.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and system for synchronously controlling the flowering period of green plants based on dynamic perception. Background Technology

[0002] In the field of green plant flowering period regulation technology, existing technologies mostly employ fixed and standardized control strategies. Common practices include setting uniform light duration and spectral combinations based on experience (such as a fixed ratio of red to blue light), or applying indiscriminate light treatment to the entire green area according to traditional phenological patterns. Although some technologies have emerged in recent years that use environmental sensors to monitor basic environmental parameters such as temperature and humidity to regulate plant flowering periods, limitations remain in the detection and precise control of individual plant physiological states. These technologies fail to fully consider the differences in growth stages among different plants, resulting in the inability to achieve synchronous regulation of flowering periods for plants in the same area. Consequently, some plants in the same area still flower earlier than others.

[0003] Therefore, existing technologies lack dynamic monitoring and differentiated regulation methods for different physiological states of green plants, resulting in insufficient precision in the synchronous regulation of green plant flowering periods. Summary of the Invention

[0004] This invention provides a method and system for synchronous regulation of green plant flowering periods based on dynamic perception, which can improve the accuracy of synchronous regulation of green plant flowering periods.

[0005] An embodiment of the present invention provides a method for synchronously controlling the flowering period of green plants based on dynamic sensing, comprising the following steps:

[0006] The visible light-near infrared dual-channel imaging device is controlled to scan the green plants in the target green plant area to obtain the canopy multispectral image sequence of the green plants at different spatial locations and the ambient light spectral distribution data of the target green plant area;

[0007] Based on the canopy image, the chlorophyll fluorescence intensity value and anthocyanin reflectance characteristics of each plant are extracted to generate a plant physiological state vector containing spatial location markers.

[0008] The plant physiological state vector is input into the flowering period convolution prediction model optimized by transfer learning, and combined with the ambient light spectral distribution and historical phenological data, the predicted flowering time window for each plant is obtained.

[0009] Spatiotemporal overlay analysis is performed on the predicted flowering time windows of each of the green plants to calculate the average flowering period of the population. Based on the predicted flowering time windows of each of the green plants and the average flowering period of the population, green plants with abnormal flowering periods are screened into the control target set.

[0010] For the green plants at various spatial locations in the control target set, calculate the dominant wavelength and photon flux density of the required compensation spectrum, and generate spectral control instructions for the green plants at the corresponding spatial locations.

[0011] Based on the spectral modulation command, a rotatable array-type spectral emitter is controlled to provide directional illumination to the green plants at the corresponding spatial locations.

[0012] As an improvement to the above scheme, the step of extracting the chlorophyll fluorescence intensity value and anthocyanin reflectance features of each plant based on the canopy image to generate a plant physiological state vector containing spatial location markers includes the following sub-steps:

[0013] The acquired canopy multispectral image sequence was subjected to spectral reflectance feature separation processing to extract the fluorescence response spectrum of the red channel and the reflectance feature spectrum of the near-infrared channel, respectively.

[0014] Based on the fluorescence response spectrum, the chlorophyll fluorescence intensity value is calculated using a dynamic threshold segmentation algorithm. At the same time, based on the reflectance feature spectrum, the anthocyanin reflectance feature is analyzed by a convolutional neural network.

[0015] Using the spatial coordinates of each plant as the primary index key, the corresponding chlorophyll fluorescence intensity value and anthocyanin reflectance feature are fused to generate a plant physiological state vector containing spatial location markers.

[0016] As an improvement to the above scheme, the step of inputting the plant physiological state vector into a flower period convolutional prediction model optimized by transfer learning, and combining it with the ambient light spectral distribution and historical phenological data to obtain the predicted flowering time window for each plant, includes the following sub-steps:

[0017] A dual-branch fusion neural network is constructed, wherein the main branch of the dual-branch fusion neural network is used to process the plant physiological state vector, and the auxiliary branch of the dual-branch fusion neural network is used to analyze the temporal fluctuation characteristics of the ambient light spectrum.

[0018] Retrieve flowering phenology records of the same species and latitude as the plant from the historical phenology database, and use a transfer learning algorithm to correct the convolution kernel weight parameters of the main branch to obtain the corrected main branch.

[0019] The real-time ambient light spectral distribution is input into the auxiliary branch to generate a spectral influence feature tensor.

[0020] The plant physiological state vector is input into the corrected main branch, and the spectral influence feature tensor is fused through a cross-modal attention mechanism to obtain the spatiotemporal feature fusion matrix.

[0021] Perform time series prediction decoding on the spatiotemporal feature fusion matrix to generate a probability-weighted prediction flowering time window.

[0022] As an improvement to the above scheme, the step of performing spatiotemporal overlay analysis on the predicted flowering time windows of each of the green plants, calculating the average flowering period of the population, and selecting green plants with abnormal flowering periods to the control target set based on the predicted flowering time windows of each of the green plants and the average flowering period of the population, includes the following sub-steps:

[0023] Construct a spatiotemporal relationship map between the predicted flowering time window of each green plant and its spatial location coordinates;

[0024] A spatiotemporal clustering algorithm based on kernel density estimation is used to calculate the spatiotemporal relationship map to obtain the average flowering period center point and the standard deviation of the flowering period distribution of the population.

[0025] Calculate the spatiotemporal offset between the earliest flowering time of each plant and the center point of the average flowering period of the group, and divide the offset by the standard deviation of the flowering period distribution to generate the flowering lag coefficient.

[0026] Determine whether the absolute value of the flowering lag coefficient exceeds the preset flowering period specific threshold of the green plant. If it does, add the spatial location coordinates of the green plant to the control target set.

[0027] As an improvement to the above scheme, the step of calculating the dominant wavelength and photon flux density of the required compensation spectrum for the green plants at each spatial location in the control target set, and generating spectral control instructions for the green plants at the corresponding spatial locations, includes the following sub-steps:

[0028] The spectral modulation mode of the green plant at the corresponding spatial location is selected according to the sign direction of the flowering lag coefficient. If it is positive, the spectral modulation mode adopts the photo-promoting main wavelength, and if it is negative, the spectral modulation mode adopts the photo-inhibiting main wavelength.

[0029] Based on the chlorophyll fluorescence intensity value of the green plant at the corresponding spatial location, the photosynthetic enzyme activity index is obtained by matching in the photophysiological response rule base.

[0030] The absolute value of the flowering lag coefficient and the photosynthetic enzyme activity index are input into the photomorphodynamic equation to calculate the photon flux compensation coefficient of the plant at the corresponding spatial location.

[0031] By combining the species' light saturation point parameters, the target photon flux density of the green plants at the corresponding spatial locations is calculated using a piecewise function;

[0032] By integrating spatial location coordinates, the dominant wavelength, the target photon flux density, and the pre-dawn irradiation period, a spectral control command containing wavelength-intensity-temporal parameters is generated for the green plant at the corresponding spatial location.

[0033] As an improvement to the above solution, the step of controlling a rotatable array-type spectral emitter to provide directional illumination to the greenery at the corresponding spatial location based on the spectral modulation command includes the following sub-steps:

[0034] The spatial position coordinates in the spectral control command are analyzed to calculate the illumination azimuth angle of the rotatable array spectral emitter.

[0035] Configure the spectral waveform output mode of the rotatable array spectral emitter according to the dominant wavelength parameter;

[0036] During the pre-dawn illumination period each day, a rotatable array-type spectral emitter is controlled to directionally illuminate the green plants at the corresponding spatial locations according to the illumination azimuth angle, with the directional illumination parameters of the spectrum configured according to the target photon flux density.

[0037] Another embodiment of the present invention provides a dynamic sensing-based synchronous control system for the flowering period of green plants, comprising:

[0038] The scanning module is used to control the visible light-near infrared dual-channel imaging device to scan the green plants in the target green plant area, and obtain the canopy multispectral image sequence of the green plants at different spatial locations and the ambient light spectral distribution data of the target green plant area;

[0039] The feature extraction module is used to extract the chlorophyll fluorescence intensity value and anthocyanin reflectance features of each plant based on the canopy image, and generate a plant physiological state vector containing spatial location markers.

[0040] The prediction module is used to input the plant physiological state vector into the flowering period convolution prediction model optimized by transfer learning, and combine it with the ambient light spectral distribution and historical phenological data to obtain the predicted flowering time window for each plant.

[0041] The calculation module is used to perform spatiotemporal overlay analysis on the predicted flowering time windows of each of the green plants, calculate the average flowering period of the group, and filter out green plants with abnormal flowering periods to the control target set based on the predicted flowering time windows of each of the green plants and the average flowering period of the group.

[0042] The instruction generation module is used to calculate the dominant wavelength and photon flux density of the required compensation spectrum for the green plants at various spatial locations in the control target set, and generate spectral control instructions for the green plants at the corresponding spatial locations.

[0043] The control module is used to control a rotatable array-type spectral emitter to provide directional illumination to the green plants at the corresponding spatial locations based on the spectral control command.

[0044] As an improvement to the above scheme, the feature extraction module is specifically used for:

[0045] The acquired canopy multispectral image sequence was subjected to spectral reflectance feature separation processing to extract the fluorescence response spectrum of the red channel and the reflectance feature spectrum of the near-infrared channel, respectively.

[0046] Based on the fluorescence response spectrum, the chlorophyll fluorescence intensity value is calculated using a dynamic threshold segmentation algorithm. At the same time, based on the reflectance feature spectrum, the anthocyanin reflectance feature is analyzed by a convolutional neural network.

[0047] Using the spatial coordinates of each plant as the primary index key, the corresponding chlorophyll fluorescence intensity value and anthocyanin reflectance feature are fused to generate a plant physiological state vector containing spatial location markers.

[0048] As an improvement to the above scheme, the prediction module is specifically used for:

[0049] A dual-branch fusion neural network is constructed, wherein the main branch of the dual-branch fusion neural network is used to process the plant physiological state vector, and the auxiliary branch of the dual-branch fusion neural network is used to analyze the temporal fluctuation characteristics of the ambient light spectrum.

[0050] The flowering phenology records of the same species and latitude as the plant were retrieved from the historical phenology database. The convolution kernel weight parameters of the main branch were corrected using a transfer learning algorithm to obtain the corrected main branch.

[0051] The real-time ambient light spectral distribution is input into the auxiliary branch to generate a spectral influence feature tensor.

[0052] The plant physiological state vector is input into the corrected main branch, and the spectral influence feature tensor is fused through a cross-modal attention mechanism to obtain the spatiotemporal feature fusion matrix.

[0053] Perform time series prediction decoding on the spatiotemporal feature fusion matrix to generate a probability-weighted prediction flowering time window.

[0054] As an improvement to the above solution, the calculation module is specifically used for:

[0055] Construct a spatiotemporal relationship map between the predicted flowering time window of each green plant and its spatial location coordinates;

[0056] A spatiotemporal clustering algorithm based on kernel density estimation is used to calculate the spatiotemporal relationship map to obtain the average flowering period center point and the standard deviation of the flowering period distribution of the population.

[0057] Calculate the spatiotemporal offset between the earliest flowering time of each plant and the center point of the average flowering period of the group, and divide the offset by the standard deviation of the flowering period distribution to generate the flowering lag coefficient.

[0058] Determine whether the absolute value of the flowering lag coefficient exceeds the preset flowering period specific threshold of the green plant. If it does, add the spatial location coordinates of the green plant to the control target set.

[0059] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0060] This invention utilizes a visible-near-infrared dual-channel imaging device to scan and acquire multispectral image sequences of plant canopies and ambient light spectral distribution data. By extracting features such as chlorophyll fluorescence intensity and anthocyanin reflectance, a plant physiological state vector is generated, providing precise data for subsequent regulation. A flower-time prediction model optimized through transfer learning is used, integrating ambient light spectral data and historical phenological data to accurately predict the flowering time window for each plant. Then, spatiotemporal overlay analysis is employed to screen plants with abnormal flowering times, calculate compensating spectral parameters to generate spectral regulation commands, and finally, directional irradiation is performed using a rotatable array spectral emitter. As the above analysis shows, this invention effectively solves the problem of accurately synchronizing plant flowering times in existing technologies by dynamically sensing and differentially regulating the individual physiological states of different plants, thus improving the accuracy of synchronous regulation of plant flowering times. Attached Figure Description

[0061] Figure 1 This is a flowchart illustrating a method for synchronously controlling the flowering period of green plants based on dynamic sensing, provided in an embodiment of the present invention.

[0062] Figure 2 This is a schematic diagram of a dynamic sensing-based synchronous control system for the flowering period of green plants, provided in an embodiment of the present invention. Detailed Implementation

[0063] 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.

[0064] See Figure 1 This is a flowchart illustrating a method for synchronously controlling the flowering period of green plants based on dynamic sensing, according to an embodiment of the present invention. The method includes the following steps:

[0065] S10, control the visible light-near infrared dual-channel imaging device to scan the green plants in the target green plant area, and obtain the canopy multispectral image sequence of the green plants in different spatial locations and the ambient light spectral distribution data of the target green plant area;

[0066] S11, Based on the canopy image, extract the chlorophyll fluorescence intensity value and anthocyanin reflectance characteristics of each plant to generate a plant physiological state vector containing spatial location markers.

[0067] S12, the plant physiological state vector is input into the flowering period convolution prediction model optimized by transfer learning, and combined with the ambient light spectral distribution and historical phenological data, the predicted flowering time window for each plant is obtained.

[0068] S13, perform spatiotemporal overlay analysis on the predicted flowering time windows of each of the green plants, calculate the average flowering period of the population, and screen out green plants with abnormal flowering periods to the control target set based on the predicted flowering time windows of each of the green plants and the average flowering period of the population.

[0069] S14, For the green plants at each spatial location in the control target set, calculate the main wavelength and photon flux density of the required compensation spectrum, and generate the spectral control command for the green plants at the corresponding spatial location.

[0070] S15, based on the spectral modulation command, control the rotatable array spectral emitter to directionally irradiate the green plants corresponding to the spatial position.

[0071] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0072] This invention utilizes a visible-near-infrared dual-channel imaging device to scan and acquire multispectral image sequences of plant canopies and ambient light spectral distribution data. By extracting features such as chlorophyll fluorescence intensity and anthocyanin reflectance, a plant physiological state vector is generated, providing precise data for subsequent regulation. A flower-time prediction model optimized through transfer learning is used, integrating ambient light spectral data and historical phenological data to accurately predict the flowering time window for each plant. Then, spatiotemporal overlay analysis is employed to screen plants with abnormal flowering times, calculate compensating spectral parameters to generate spectral regulation commands, and finally, directional irradiation is performed using a rotatable array spectral emitter. As the above analysis shows, this invention effectively solves the problem of accurately synchronizing plant flowering times in existing technologies by dynamically sensing and differentially regulating the individual physiological states of different plants, thus improving the accuracy of synchronous regulation of plant flowering times.

[0073] As an example, in step S10, a visible-near-infrared dual-channel imaging device with high resolution and wide spectral response can be selected. The imaging device is mounted on a movable mechanical gimbal or track system to ensure flexible and stable scanning of the target green area. For example, in a green area under a bridge, the imaging device can be fixed above the area and driven by a motor to scan different areas of the greenery. Before scanning, the imaging device can be calibrated, including the exposure time and gain of the visible and near-infrared channels, to ensure clear and accurate images. When scanning begins, the imaging device scans the greenery in the target area line by line according to a preset scanning path, such as from left to right or from top to bottom. During the scanning process, the visible light channel captures the reflected light information of the plant canopy in the visible light band (400-700nm), generating visible light images reflecting the plant's morphology, color, and other characteristics. The near-infrared channel acquires the reflected light information of the plant canopy in the near-infrared band (700-1100nm), generating corresponding images. By scanning plants at different spatial locations, a series of multispectral image sequences of the plant canopy at different spatial locations can be obtained. Simultaneously, the ambient light spectral sensor on the imaging device collects real-time ambient light spectral distribution data of the target plant area, such as the spectral composition of solar radiation, thereby achieving a complete scanning of the target plant area and obtaining the required image sequences and ambient light spectral distribution data.

[0074] As an improvement to the above embodiment, the step of extracting the chlorophyll fluorescence intensity value and anthocyanin reflectance features of each plant based on the canopy image to generate a plant physiological state vector containing spatial location markers includes the following sub-steps:

[0075] The acquired canopy multispectral image sequence was subjected to spectral reflectance feature separation processing to extract the fluorescence response spectrum of the red channel and the reflectance feature spectrum of the near-infrared channel, respectively.

[0076] Based on the fluorescence response spectrum, the chlorophyll fluorescence intensity value is calculated using a dynamic threshold segmentation algorithm. At the same time, based on the reflectance feature spectrum, the anthocyanin reflectance feature is analyzed by a convolutional neural network.

[0077] Using the spatial coordinates of each plant as the primary index key, the corresponding chlorophyll fluorescence intensity value and anthocyanin reflectance feature are fused to generate a plant physiological state vector containing spatial location markers.

[0078] In this embodiment, the target green plant area is first scanned using a visible-near-infrared dual-channel imaging device to acquire canopy multispectral image sequences and real-time ambient light spectral distribution data for each plant, providing raw information for subsequent analysis. Next, the canopy multispectral image sequences are separated by spectral reflectance features to distinguish between red and near-infrared channels, enabling targeted data processing. Then, a dynamic threshold segmentation algorithm is used to calculate chlorophyll fluorescence intensity from the fluorescence response spectrum of the red channel, and a convolutional neural network is used to analyze anthocyanin reflectance features from the reflectance feature spectrum of the near-infrared channel. These two different but compatible techniques are employed to extract key physiological indicators. Finally, using the spatial coordinates of the green plant as the index key, the chlorophyll fluorescence intensity and anthocyanin reflectance features are fused to generate a spatially labeled plant physiological state vector. This gives the data both spatial and physiological attributes, providing a precise and structured data foundation for subsequent flowering period prediction and regulation. Compared to existing technologies that struggle to accurately acquire and integrate plant physiological and spatial information, this embodiment achieves precise quantification and spatial positioning of the physiological state of each plant, effectively improving the accuracy, completeness, and usability of plant physiological state data, and providing a data foundation for subsequent precise prediction and synchronous regulation of flowering time based on individual differences.

[0079] As an example, the working process of this embodiment is as follows:

[0080] For the first sub-step, in the environment under the bridge, due to the dim lighting, numerous shadow areas, and extremely uneven light distribution, traditional spectral separation methods are insufficient. Therefore, this implementation adopts the following intelligent dynamic spectral analysis and adaptive separation scheme. First, a visible-near-infrared dual-channel imaging device scans the target green area from all directions and multiple angles every 30 seconds, acquiring a sequence of canopy multispectral images. Considering that the lighting under the bridge is affected by various factors such as weather, time, and obstruction by surrounding buildings, an environmental perception lighting compensation system is constructed. This system integrates real-time data from a weather station (including rainfall, wind speed, air quality index, etc.), 3D modeling data of surrounding buildings (used to simulate shadow obstruction), and light intensity and spectral distribution data collected by lighting sensors installed at different locations under the bridge. The lighting compensation formula is: ,in, This is the compensated spectral value. For pixel coordinates, Indicates spectral channels; These are the original spectral values; These are standard light spectrum values ​​for the same time period and geographical location; It is based on the angle between the imaging device and the greenery. The determined angle correction parameters are obtained by establishing a three-dimensional angle-light attenuation model. Based on the height of the green plants The height compensation parameter takes into account the differences in the influence of sunlight on green plants at different heights; It is the average spectral value measured under the current environment.

[0081] In the spectral separation stage, a pre-trained Transformer model is first used to extract global features from the image, while a convolutional neural network is combined to extract local detail features. The features output by both models are then concatenated and fused. Based on this, an improved hybrid pixel decomposition model is constructed: ,in, It is a pixel In spectral channels The observed spectrum; It is the number of endmembers, determined by an adaptive method based on information entropy; It is the first Each terminal element in a pixel The abundance of is predicted by a generative adversarial network with an attention mechanism, and further enhanced by introducing a regularization term. To constrain the sparsity of abundance; It is the first Spectral curves of each endmember; This is the residual term. By alternately optimizing the abundance matrix and endmember matrix, combined with a reinforcement learning-based parameter adjustment strategy, the decomposition process is continuously optimized, ultimately accurately separating the fluorescence response spectrum of the red channel and the reflectance feature spectrum of the near-infrared channel.

[0082] For the second sub-step, addressing the issues of weak chlorophyll fluorescence signals and high noise interference from the green plants under the bridge, a spatiotemporal-spectral multidimensional information neighborhood is first constructed for each pixel. This involves selecting three consecutive frames of images in the temporal dimension and selecting a neighborhood centered on that pixel in the spatial dimension. The neighborhood is determined by selecting three adjacent channels along the spectral dimension to form a three-dimensional information neighborhood. Pixel point calculation... Dynamic threshold: ,in, and These are the mean and standard deviation of the spatiotemporal-spectral multidimensional information neighborhood of the pixel, respectively. It is the mean adjustment coefficient, which was determined to be 1.2 through training with a large amount of sample data of green plants under the bridge; and These are adjustment coefficients related to time variation and spectral variation, respectively, which were experimentally set to 0.6 and 0.5, respectively. It is the amount of grayscale change of the pixel over time, reflecting the fluctuation of the fluorescence signal over time; It represents the spectral difference in the spectral dimension, reflecting the characteristic differences between different spectral channels. When the pixel value is greater than... When the region is identified as a valid fluorescence region, the pixel values ​​within the valid region are summed using a Gaussian weighted average (the weights are calculated based on the distance from the pixel to the neighborhood center and the spectral similarity), and then normalized to obtain the chlorophyll fluorescence intensity value.

[0083] To analyze anthocyanin reflectance features, a hierarchical multi-scale attention fusion convolutional neural network (HM-MAFCNN) was constructed. The network consists of four layers, each containing multiple parallel convolutional modules at different scales to extract reflectance feature map information at different scales. Within each layer, a bidirectional channel-spatial attention mechanism is introduced, first calculating the attention weights along the channel dimension: ,in, It is a feature map. and It is a learnable weight matrix. It is the ReLU activation function. It's the sigmoid function. Then, we calculate the attention weights for the spatial dimension: ,in, This is a convolution operation. The channel attention weights and spatial attention weights are multiplied to obtain the final attention weights, which are then multiplied with the feature map for feature enhancement. At the end of the network, the anthocyanin reflectance features are calculated through a fully connected layer.

[0084] For the third sub-step, a simplified visual positioning and coordinate calculation method is used to obtain the spatial coordinates of the greenery. Several visually distinctive markers are laid at fixed intervals on the ground in the greenery area under the bridge, and the actual three-dimensional coordinates of these markers are precisely measured beforehand. When a visible light-near-infrared dual-channel imaging device scans the greenery, images containing both the markers and the greenery are simultaneously captured. Image recognition technology is used to identify the positions of the markers and the greenery in the images. Utilizing the correspondence between the pixel coordinates of the markers in the image and their actual three-dimensional coordinates, and based on the principle of perspective transformation, a transformation matrix between the pixel coordinates of the greenery in the image and their actual three-dimensional coordinates is calculated. Then, based on this transformation matrix, the pixel coordinates of the greenery in the image are converted into their actual three-dimensional spatial coordinates. In the fusion phase, an ordered data fusion model is created, using the three-dimensional spatial coordinates of each plant. A data record unit is created using the index as the primary key. The calculated chlorophyll fluorescence intensity value, anthocyanin reflectance characteristic value, and data acquisition timestamp, etc., are sequentially stored in this data record unit according to a fixed order and data format. In this way, the spatial location information and physiological characteristic information of each plant are tightly integrated to generate a plant physiological state vector containing spatial location markers, providing data support for subsequent prediction and regulation of plant flowering periods.

[0085] As an improvement to the above embodiment, the step of inputting the plant physiological state vector into the flowering period convolutional prediction model optimized by transfer learning, and combining it with the ambient light spectral distribution and historical phenological data to obtain the predicted flowering time window for each plant, includes the following sub-steps:

[0086] A dual-branch fusion neural network is constructed, wherein the main branch of the dual-branch fusion neural network is used to process the plant physiological state vector, and the auxiliary branch of the dual-branch fusion neural network is used to analyze the temporal fluctuation characteristics of the ambient light spectrum.

[0087] The flowering phenology records of the same species and latitude as the plant were retrieved from the historical phenology database. The convolution kernel weight parameters of the main branch were corrected using a transfer learning algorithm to obtain the corrected main branch.

[0088] The real-time ambient light spectral distribution is input into the auxiliary branch to generate a spectral influence feature tensor.

[0089] The plant physiological state vector is input into the corrected main branch, and the spectral influence feature tensor is fused through a cross-modal attention mechanism to obtain the spatiotemporal feature fusion matrix.

[0090] Perform time series prediction decoding on the spatiotemporal feature fusion matrix to generate a probability-weighted prediction flowering time window.

[0091] In this embodiment, a dual-branch fusion neural network is first constructed. With clearly defined main and auxiliary branches, the main and auxiliary branches respectively process the temporal fluctuation characteristics of the plant physiological state vector and the ambient light spectrum, laying a structural foundation for subsequent analysis. Next, flowering phenological records of the same species at the same latitude are retrieved from a historical phenological database. A transfer learning algorithm is used to correct the weight parameters of the main branch's convolution kernel, enabling the model to learn from historical experience and enhance the accuracy of predicting the current flowering period of green plants. Then, the real-time ambient light spectral distribution is input into the auxiliary branch to generate a spectral influence feature tensor. This tensor is then deeply fused with the corrected plant physiological state vector from the main branch through a cross-modal attention mechanism, forming a spatiotemporal feature fusion matrix, achieving effective integration of multi-source data. Finally, time-series prediction decoding is performed on the spatiotemporal feature fusion matrix, outputting a probability-weighted predicted flowering time window, completing the entire prediction process from data input to result output. Compared to traditional single-data or simple model-based flowering period prediction methods, this embodiment, by constructing a dual-branch fusion neural network and introducing transfer learning and a cross-modal attention mechanism, fully utilizes multi-source data information such as plant physiological state, ambient light spectrum, and historical phenology to deeply explore the potential correlations between data. It effectively solves the problems of insufficient accuracy in predicting flowering time and difficulty in adapting to different environments and species in existing technologies, and improves the accuracy and reliability of predicting the flowering time window of each green plant.

[0092] As an example, the working process of this embodiment is as follows:

[0093] (1) To accurately process the multi-source data required for predicting the flowering period of green plants under bridges, a dual-branch fusion neural network was constructed. The main branch is dedicated to processing plant physiological state vectors, and it adopts an improved lightweight convolutional neural network structure. A feature enhancement module is introduced after the traditional convolutional layer. This module enhances the feature map output by the convolutional layer based on the correlation of plant physiological features using the following formula. Enhancement: ,in, It is a learnable weight matrix. It is a bias vector. This is a sigmoid activation function. This module can highlight the key role of chlorophyll fluorescence intensity and anthocyanin reflectance characteristics in the plant physiological state vector, and suppress noise interference caused by the complex environment under the bridge.

[0094] The auxiliary branch is used to analyze the temporal fluctuation characteristics of the ambient light spectrum, employing a temporal-frequency hybrid convolutional architecture. First, a one-dimensional convolutional layer is used to perform preliminary feature extraction on the ambient light spectral distribution data in the temporal dimension, obtaining a preliminary feature sequence. Next, a frequency analysis module based on Fourier transform is introduced to... Transform to the frequency domain to extract periodic fluctuation features from the spectral data. The final input features It is formed by fusing time-dimensional features and frequency features: This architecture can simultaneously capture the changing trends and periodic fluctuations of ambient light spectra over time, providing more comprehensive ambient light information for subsequent analysis.

[0095] (2) After determining the species and latitude of the green plants in the area under the bridge, flowering phenological records of the same species at the same latitude are retrieved from the historical phenological database. Since the environment under the bridge (e.g., insufficient sunlight, high humidity) differs from the conventional environment, an environmental adaptation transfer learning algorithm is used to optimize the main branch. First, the degree of difference between the environmental parameters under the bridge and those in historical data is calculated. Environmental parameters include light intensity, temperature, and humidity, and the calculation formula is as follows: ,in, The number of types of environmental parameters. It is the current number under the bridge Various environmental parameter values, It is the first in historical data The average value of various environmental parameters It is the first The weighting coefficients of the influence of environmental parameters on the flowering period of green plants were determined through extensive experiments and machine learning training.

[0096] Then, based on the degree of difference The following formula is used to weight the convolution kernel of the main branch. Make corrections: ,in, These are the original weights. These are the corrected weights. It is the learning rate, which is dynamically adjusted based on the training progress using an adaptive algorithm; The loss function gradient is calculated based on a limited amount of known flowering period data from under the bridge. This approach allows the main branch of the convolutional neural network to better adapt to the unique environment under the bridge, improving the accuracy of processing plant physiological state vectors.

[0097] (3) The ambient light spectral distribution data acquired in real time by the spectral acquisition equipment deployed under the bridge is organized into a data sequence according to time order and then input into the auxiliary branch. The data is first normalized to ensure that its values ​​are distributed within a suitable range, which is convenient for network learning. In the temporal-frequency hybrid convolutional architecture of the auxiliary branch, the data is first processed by a one-dimensional convolutional layer for feature extraction. The convolution process can be represented as: ,in, It is the output of the convolutional layer. It is the input data sequence. These are the convolution kernel weights. It's a bias. It is the kernel size. It is an index of the data sequence. After multiple convolution and pooling operations, combined with the features extracted by the frequency analysis module, the features are mapped to a high-dimensional space through a fully connected layer, ultimately generating a three-dimensional spectral influence feature tensor. Each dimension corresponds to a different feature category, spatial location, and time series information, comprehensively characterizing the potential impact of ambient light spectrum on the flowering period of green plants.

[0098] (4) Input the plant physiological state vector, which includes the chlorophyll fluorescence intensity value, anthocyanin reflectance features, and spatial location markers of each plant, into the main branch after transfer learning optimization. The main branch performs feature extraction and transformation on the plant physiological state vector to obtain the plant physiological feature matrix. To achieve effective fusion of plant physiological characteristics and environmental light spectral characteristics, a bidirectional cross-modal attention mechanism is employed. First, the plant physiological characteristic matrix is ​​calculated. With spectral influence characteristic tensor Attention weights between : ,in, and These are the plant physiological characteristic matrices. and the influence of spectral characteristics tensor The weight matrix for linear transformation. It is a scaling factor used to prevent values ​​from becoming too large during calculation.

[0099] Then, based on attention weights Plant physiological characteristic matrix and the influence of spectral characteristics tensor Perform weighted fusion: ,in, This is the spatiotemporal feature fusion matrix, which fully combines information about the plant's physiological state and the ambient light spectrum, providing a rich and closely related data foundation for subsequent flowering period prediction.

[0100] (5) For the obtained spatiotemporal feature fusion matrix An improved temporal decoding network is employed for time series prediction decoding. This network consists of a multi-layer Long Short-Term Memory (LSTM) network and fully connected layers. The LSTM layers process the spatiotemporal feature fusion matrix. At that time, through the Gate of Oblivion Input gate and output gate The formula for calculating the transmission and updating of control information is as follows: , , ,in, These are the weight matrices for the forget gate, input gate, and output gate, respectively. It is the corresponding bias vector. It is the hidden state from the previous moment. It is the input at the current moment. The Sigmoid activation function is used. After multiple LSTM layers fully extract the time-series information, the final hidden state is input into a fully connected layer. Through linear transformation and the Softmax activation function of the fully connected layer, the features are mapped to the probability distribution of plant flowering at different time points: ,in, It is the weight matrix of the fully connected layer. It is a bias vector. This is the final hidden state of the LSTM layer. Is The probability vector of when the green plant blooms at any given time.

[0101] Finally, based on the probability distribution, a continuous time interval in which the cumulative probability value reaches a set threshold (such as 0.8) is selected as the probability-weighted prediction window for the flowering time of the plant, thereby achieving accurate prediction of the flowering time of each plant and providing a reliable basis for subsequent synchronous regulation of flowering period.

[0102] As an improvement to the above embodiment, the step of performing spatiotemporal overlay analysis on the predicted flowering time windows of each of the green plants, calculating the average flowering period of the population, and selecting green plants with abnormal flowering periods to the control target set based on the predicted flowering time windows of each of the green plants and the average flowering period of the population, includes the following sub-steps:

[0103] Construct a spatiotemporal relationship map between the predicted flowering time window of each green plant and its spatial location coordinates;

[0104] A spatiotemporal clustering algorithm based on kernel density estimation is used to calculate the spatiotemporal relationship map to obtain the average flowering period center point and the standard deviation of the flowering period distribution of the population.

[0105] Calculate the spatiotemporal offset between the earliest flowering time of each plant and the center point of the average flowering period of the group, and divide the offset by the standard deviation of the flowering period distribution to generate the flowering lag coefficient.

[0106] Determine whether the absolute value of the flowering lag coefficient exceeds the preset flowering period specific threshold of the green plant. If it does, add the spatial location coordinates of the green plant to the control target set.

[0107] In this embodiment, the predicted flowering time window and spatial coordinates of each plant are first integrated to construct a relationship graph containing spatiotemporal information, giving the data a structured and visual foundation. Then, a spatiotemporal clustering algorithm based on kernel density estimation is used to process the graph, deeply mining the data distribution patterns to obtain the average flowering period center point and the standard deviation of the flowering period distribution, clarifying the overall trend and dispersion of the group's flowering. Subsequently, the spatiotemporal offset between the earliest flowering time of each plant and the average flowering period center point of the group is calculated and divided by the standard deviation of the flowering period distribution to obtain the flowering lag coefficient, quantifying the difference between the flowering period of a single plant and the group. Finally, the absolute value of the flowering lag coefficient is compared with a preset flowering period specificity threshold for the plant. When the threshold is exceeded, the spatial coordinates of the plant are included in the control target set, completing the screening process. This embodiment, by constructing a spatiotemporal relationship graph, using clustering algorithms, and employing quantitative evaluation methods, fully analyzes the spatiotemporal characteristics of plant flowering periods, improving the objectivity and accuracy of screening plants with abnormal flowering periods.

[0108] As an example, the working process of this embodiment is as follows:

[0109] First, based on the above method of using cameras and markers to identify the spatial location of plants, the precise spatial coordinates of each plant are obtained. Simultaneously, using the flower-time convolutional prediction model optimized through transfer learning in the above embodiments, the predicted flowering time window for each plant is obtained, denoted as... , To predict the earliest flowering time, To predict the latest flowering time, a data node structure containing various information was created, including the spatial coordinates of each plant. As a unique identifier, the corresponding predicted flowering time window will be used as the primary key. Data collection timestamps, environmental data such as light intensity, humidity, and temperature collected by surrounding environmental monitoring equipment are used as node attributes. Based on the graph data structure, connections are constructed according to the spatial proximity between plants and the overlap of predicted flowering times. Spatial distance thresholds are set. (e.g., 1 meter) and time overlap threshold (If the predicted flowering time overlaps by more than 3 days), if the spatial distance between the two plants is less than... And the predicted flowering time overlap exceeds Then, an edge connection is established between the two in the graph. The weight of the edge is calculated based on the product of the reciprocal of the spatial distance and the degree of temporal overlap, ultimately forming a spatiotemporal relationship graph that intuitively presents the spatiotemporal flowering prediction and environmental correlation of the green plant group under the bridge.

[0110] Next, considering the complexity of the spatiotemporal data on vegetation under the bridge, an improved adaptive bandwidth kernel density estimation spatiotemporal clustering algorithm is adopted. The spatiotemporal kernel function is defined as follows: ,in, These are the spatiotemporal coordinates of the point to be estimated; They are space Bandwidth in both the dimensional and time dimensions; A Gaussian kernel function is selected to smooth the data distribution. The adaptive bandwidth adjustment formula is as follows: , , , ,in, It is the first in the neighborhood The spatiotemporal coordinates of each data point This represents the number of data points within the neighborhood. The adjustment coefficients (0.8, 0.8, 0.9, and 1.2, respectively) were obtained through training and optimization using a large amount of data on vegetation under bridges, and were used to balance the data distribution range and estimation accuracy. Based on the above kernel function and adaptive bandwidth, the kernel density estimate was calculated for each data point in the spatiotemporal relationship map. A density threshold was set. (e.g., 0.3), filter out those with a density greater than Using the peak points as cluster centers, density peak clustering algorithm is employed for clustering to obtain multiple spatiotemporal clusters. For each cluster, the average predicted flowering time window of all plants within the cluster is calculated along the time dimension to obtain the center point of the average flowering period of the population. ; through formula (in For the first in the cluster Predicting the flowering time of a single green plant. Calculate the predicted flowering time of each plant relative to the number of plants in the cluster. The standard deviation of the flowering period distribution is obtained. This allows us to quantify the central tendency and dispersion of flowering time among the green plants under the bridge.

[0111] Then, the predicted flowering time window for each plant is... Its earliest flowering time value is First calculate Center point of average flowering period of the population time offset Then calculate the spatial coordinates of the green plant. Average coordinates of all green space Spatial Euclidean distance Introducing spatiotemporal weighting coefficients The time offset and spatial offset are merged into a spatiotemporal offset: Finally, the spatiotemporal offset... Divide by the standard deviation of the flowering period distribution The flowering lag coefficient was obtained. This coefficient comprehensively reflects the degree of deviation of each plant from the average flowering situation of the group in the spatiotemporal dimensions.

[0112] Finally, taking into account the biological characteristics of different plant varieties and the special environmental conditions under the bridge, historical data and machine learning algorithms were used to preset flowering period-specific thresholds for each plant. For example, for light-sensitive plant varieties, the flowering time specificity threshold was set to 0.8 under the low-light environment under the bridge; for shade-tolerant plant varieties, the threshold was set to 1.2. The flowering lag coefficient of each plant was checked individually. ,like If the flowering period of the plant is abnormal, its spatial coordinates will be determined. Added to the control target set, subsequent measures such as spectral regulation will be precisely implemented for these green plants with abnormal flowering periods, based on their physiological state and environmental needs, to achieve synchronized management of the flowering periods of green plants under the bridge.

[0113] As an improvement to the above embodiment, the step of calculating the dominant wavelength and photon flux density of the required compensation spectrum for the green plants at each spatial location in the control target set, and generating spectral control instructions for the green plants at the corresponding spatial locations, includes the following sub-steps:

[0114] The spectral modulation mode of the green plant at the corresponding spatial location is selected according to the sign direction of the flowering lag coefficient. If it is positive, the spectral modulation mode adopts the photo-promoting main wavelength, and if it is negative, the spectral modulation mode adopts the photo-inhibiting main wavelength.

[0115] Based on the chlorophyll fluorescence intensity value of the green plant at the corresponding spatial location, the photosynthetic enzyme activity index is obtained by matching in the photophysiological response rule base.

[0116] The absolute value of the flowering lag coefficient and the photosynthetic enzyme activity index are input into the photomorphodynamic equation to calculate the photon flux compensation coefficient of the plant at the corresponding spatial location.

[0117] By combining the species' light saturation point parameters, the target photon flux density of the green plants at the corresponding spatial locations is calculated using a piecewise function;

[0118] By integrating spatial location coordinates, the dominant wavelength, the target photon flux density, and the pre-dawn irradiation period, a spectral control command containing wavelength-intensity-temporal parameters is generated for the green plant at the corresponding spatial location.

[0119] In this embodiment, firstly, based on the sign of the flowering lag coefficient, a photo-promoting (dominant wavelength is red light that promotes photosynthesis) or photo-inhibiting (dominant wavelength is far-red light that delays growth) spectral regulation mode is selected for the target plants. Then, based on the chlorophyll fluorescence intensity value of the plants, the photosynthetic enzyme activity index is obtained by matching in the photophysiological response rule base. Next, the absolute value of the flowering lag coefficient and the photosynthetic enzyme activity index are substituted into the photomorphodynamic equation to calculate the photon flux compensation coefficient. Then, combined with the species light saturation point parameter, the target photon flux density is determined by a piecewise function. Finally, the spatial location coordinates, dominant wavelength, target photon flux density, and pre-dawn irradiation period are integrated to generate a spectral regulation instruction containing wavelength-intensity-temporal parameters. This embodiment achieves full automation and intelligence from the judgment of abnormal flowering period of green plants to the generation of precise spectral regulation instructions. It fully considers the physiological state of green plants (chlorophyll fluorescence intensity, photosynthetic enzyme activity), the degree of deviation of flowering period (flowering lag coefficient), and species characteristics (light saturation point). It can generate customized spectral regulation instructions for green plants with abnormal flowering period in different spatial locations, ensuring that the dominant wavelength, photon flux density, and irradiation time of spectral regulation are highly matched with the needs of green plants. It effectively solves the problem of synchronous regulation of flowering period of green plants in complex environments such as under bridges, and improves the accuracy and effectiveness of flowering period regulation.

[0120] As an example, the working process of this embodiment is as follows:

[0121] First, obtain the flowering lag coefficient for each plant. Then, the system automatically determines its sign direction. If A positive value indicates that the flowering period of this plant is delayed compared to the average flowering period of the population, in which case a light-promoting spectral control mode should be selected. For plants growing under bridges, the light-promoting spectral mode mainly uses red light with a predominant wavelength of 660nm, because red light can effectively promote photosynthesis, accelerate chlorophyll synthesis, and thus promote flower bud differentiation and flowering. If A negative value indicates that the plant's flowering period is premature, requiring the use of a light-suppression spectral control mode. This mode selects 730nm far-red light as the dominant wavelength, as far-red light can inhibit photomorphogenesis and slow down plant growth, thus delaying flowering. The system associates the spectral control mode with the corresponding spatial location of the plant by querying a preset spectral mode configuration table, providing a foundation for subsequent precise control.

[0122] Next, the chlorophyll fluorescence intensity values ​​of the plants under the bridge are obtained, and the system uses this as key information for retrieval. The photophysiological response rule base is constructed based on extensive plant physiology experiments and monitoring data of the plants under the bridge. The base stores the correspondence between different chlorophyll fluorescence intensity value ranges and photosynthetic enzyme activity indices. This correspondence is not a simple linear mapping, but rather derived through nonlinear fitting of experimental data using machine learning algorithms. For example, when the chlorophyll fluorescence intensity value... When the value is in the range of [100-150], the trained decision tree model will output the corresponding photosynthetic enzyme activity index range of [0.6-0.8]. The system will then determine the specific value based on the given information. The value was used to determine the accurate photosynthetic enzyme activity index using linear interpolation. This matching method fully considers the impact of the complex environment under the bridge on the physiological indicators of green plants, and improves the accuracy of obtaining the photosynthetic enzyme activity index.

[0123] Then, based on the following photomorphodynamic equation, the photon flux compensation coefficient of the green plant at the corresponding spatial location is calculated: ,in, The photon flux compensation coefficient; It is the absolute value of the flowering lag coefficient, reflecting the degree to which the flowering period of green plants deviates from the average flowering period of the population; The photosynthetic enzyme activity index reflects the current physiological activity status of the green plant. These parameters were obtained through extensive experiments regulating greenery under bridges and optimization using a genetic algorithm. By introducing an exponential relationship, the complex nonlinear relationship between flowering time deviation, physiological activity, and photon flux compensation is described more accurately. For example, for greenery with severe flowering lag and low photosynthetic enzyme activity, the parameters calculated using this equation... A larger value means that the photon flux needs to be adjusted more significantly.

[0124] Then, the target photon flux density of the plant at the corresponding spatial location is calculated using a piecewise function. The light saturation point parameter for each plant species is pre-stored in the system's species parameter database; this parameter represents the light intensity at which the plant reaches its maximum photosynthetic rate. The piecewise function is defined as follows: ,in, For the target photon flux density, This refers to the light saturation point parameter of the plant. When... When the value is greater than or equal to 0, it indicates a need to promote plant growth and flowering. Based on the light saturation point, the photon flux compensation coefficient should be used. Increase the target photon flux density proportionally; when A value less than 0 indicates that growth needs to be suppressed, and the target photon flux density should be reduced accordingly. This segmented calculation method ensures effective regulation of plants with abnormal flowering periods while avoiding the problems of decreased photosynthetic efficiency due to excessively high light intensity exceeding the light saturation point, or insufficient regulation due to excessively low light intensity.

[0125] Finally, the system will assign the three-dimensional coordinates of the greenery to the corresponding spatial location. (If only the xy coordinates of the plane are needed, z can be set to 0 or the z coordinate can be omitted.) Selected dominant wavelength Calculated target photon flux density The system integrates preset pre-dawn illumination periods (4:00-6:00 AM daily, a time when there is less environmental interference under the bridge and the plants are sensitive to light regulation). A unified format of spectral control commands is generated. These commands are transmitted as control signals to a rotatable array-type spectral transmitter. Based on the parameters in the commands, the transmitter precisely applies directional spectral illumination to the plants at the corresponding spatial locations, achieving precise control over the flowering period of the plants under the bridge and ensuring synchronization of the flowering period of the plant population.

[0126] As an improvement to the above embodiment, the step of controlling the rotatable array-type spectral emitter to provide directional illumination to the greenery corresponding to the spatial location based on the spectral modulation command includes the following sub-steps:

[0127] The spatial position coordinates in the spectral control command are analyzed to calculate the illumination azimuth angle of the rotatable array spectral emitter.

[0128] Configure the spectral waveform output mode of the rotatable array spectral emitter according to the dominant wavelength parameter;

[0129] During the pre-dawn illumination period each day, a rotatable array-type spectral emitter is controlled to directionally illuminate the green plants at the corresponding spatial locations according to the illumination azimuth angle, with the directional illumination parameters of the spectrum configured according to the target photon flux density.

[0130] In this embodiment, firstly, based on the sign direction of the flowering lag coefficient, it is determined whether the flowering period of the green plant is advanced or delayed, and then the dominant wavelength of photopromoting or photoinhibiting is selected to determine the basic light quality direction for spectral regulation. Then, based on the chlorophyll fluorescence intensity value of the green plant, the photosynthetic enzyme activity index is matched in the photophysiological response rule base to transform the plant physiological characteristics into quantifiable regulatory parameters. Next, the absolute value of the flowering lag coefficient and the photosynthetic enzyme activity index are substituted into the photomorphodynamic equation to calculate the photon flux compensation coefficient, quantifying the light intensity requirements of flowering period differences and physiological states. Then, combined with the species light saturation point parameter, the target photon flux density is calculated through a piecewise function to ensure that the light intensity is within a reasonable and effective range. Finally, the spatial location coordinates, dominant wavelength, target photon flux density, and pre-dawn irradiation period are integrated to generate a spectral regulation command containing wavelength, intensity, and temporal parameters, realizing a complete transformation from data to command. This embodiment effectively solves the problems of lack of specificity in spectral regulation and difficulty in taking into account individual plant differences and environmental adaptability in the prior art. It can accurately determine the light quality, light intensity and irradiation time of spectral regulation according to the abnormal flowering period, physiological state and species characteristics of each plant, thus improving the accuracy and effectiveness of spectral regulation.

[0131] As an example, the working process of this embodiment is as follows:

[0132] First, after receiving the spectral modulation command, the system extracts the spatial coordinates of the target plant from the command. Simultaneously, the system reads the pre-stored installation coordinates of the rotatable array-type spectral emitter itself. To calculate the azimuth angle of the illumination, the system first determines the projection direction of the target vegetation relative to the spectral emitter on the horizontal plane. This is done by calculating the projection direction of the two azimuth angles. shaft and Coordinate difference along the axis and Using the arctangent function Obtain the preliminary azimuth angle However, due to the complex environment under the bridge, with obstructions such as buildings and piers, the system will call upon the LiDAR deployed under the bridge to scan the environment in real time, acquiring 3D point cloud data to construct an obstacle model. Based on the relative positions of the obstacles, the spectral emitter, and the target vegetation, the initial azimuth angle is corrected. If obstruction exists, the system will calculate the optimal direction for light to bypass the obstacle and adjust the correction angle by adding or subtracting a correction angle value. The final azimuth angle is obtained. .

[0133] Next, the spectral modulation command includes the dominant wavelength parameter for the target plant. The system determines the spectral waveform output mode based on this parameter. If the dominant wavelength... For light-promoting (e.g., 660nm red light) applications, the system controls the multi-channel LED array of the spectral emitter, activating red LED channels with wavelengths close to 660nm and appropriately adjusting the brightness of other auxiliary wavelength channels (e.g., 730nm far-red light channels). By adjusting the duty cycle of the pulse width modulation (PWM) signal for each channel, the spectral energy is concentrated mainly around 660nm, forming a spectral waveform with 660nm as the dominant peak. Specifically, the system calculates the PWM duty cycle value of each LED channel based on a pre-set red light-promoting spectral energy distribution curve, precisely controlling the luminous intensity of each channel to output a light-promoting spectral waveform that meets the requirements, thereby promoting photosynthesis and flower bud differentiation in plants. When the dominant wavelength... When the light-suppressing type (such as 730nm far-red light) is used, the system focuses on activating the 730nm far-red LED channel and reducing the brightness of other wavelength channels. Similarly, using PWM technology, based on the light-suppressing spectral energy distribution model, the duty cycle of each channel is adjusted to generate a spectral waveform with 730nm as the main wavelength, thereby inhibiting the growth and development speed of green plants and delaying the flowering period.

[0134] Then, the system monitors the time in real time through its built-in clock module. When the time reaches 4:00 AM daily (the start time of the pre-dawn irradiation period), the system starts the irradiation program. First, the angle control motor of the spectral emitter adjusts according to the previously calculated irradiation azimuth angle. By controlling the parameters, the transmitter is driven to rotate to the accurate angle, ensuring that the light can be accurately directed at the target plant.

[0135] After determining the irradiation direction, the system uses the target photon flux density specified in the spectral modulation command. The parameters configure the luminescence intensity of the spectral emitter. The system's internal power regulation module will adjust the parameters according to the luminescence intensity of the spectral emitter. Compared with the actual output photon flux density of the current transmitter Calculate the power value that needs to be adjusted based on the difference. If The system will increase the duty cycle of the PWM signal in each LED channel to improve the LED luminous intensity, thereby increasing the photon flux density; conversely, if... This reduces the PWM signal duty cycle and decreases the luminous intensity. During the adjustment process, the system monitors the data fed back from the photon flux density sensor in real time and continuously adjusts the output through closed-loop control until the actual output photon flux density reaches the target value. This ensures that the target plants receive spectral illumination that meets the control requirements throughout the entire pre-dawn illumination period (e.g., 4:00-6:00), thereby achieving precise regulation of the flowering period of the plants.

[0136] See Figure 2This is a schematic diagram of a dynamic sensing-based synchronous control system for the flowering period of green plants, provided in an embodiment of the present invention. The dynamic sensing-based synchronous control system for the flowering period of green plants includes:

[0137] The scanning module 10 is used to control the visible light-near infrared dual-channel imaging device to scan the green plants in the target green plant area, and obtain the canopy multispectral image sequence of the green plants in different spatial locations and the ambient light spectral distribution data of the target green plant area.

[0138] The feature extraction module 11 is used to extract the chlorophyll fluorescence intensity value and anthocyanin reflectance feature of each plant based on the canopy image, and generate a plant physiological state vector containing spatial location markers.

[0139] Prediction module 12 is used to input the plant physiological state vector into the flowering period convolution prediction model optimized by transfer learning, and combine it with the ambient light spectral distribution and historical phenological data to obtain the predicted flowering time window for each plant.

[0140] The calculation module 13 is used to perform spatiotemporal overlay analysis on the predicted flowering time windows of each of the green plants, calculate the average flowering period of the group, and filter out green plants with abnormal flowering periods to the control target set based on the predicted flowering time windows of each of the green plants and the average flowering period of the group.

[0141] The instruction generation module 14 is used to calculate the main wavelength and photon flux density of the required compensation spectrum for the green plants at each spatial location in the control target set, and generate spectral control instructions for the green plants at the corresponding spatial locations.

[0142] The control module 15 is used to control the rotatable array spectral emitter to directionally irradiate the green plants corresponding to the spatial position based on the spectral control command.

[0143] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0144] This invention utilizes a visible-near-infrared dual-channel imaging device to scan and acquire multispectral image sequences of plant canopies and ambient light spectral distribution data. By extracting features such as chlorophyll fluorescence intensity and anthocyanin reflectance, a plant physiological state vector is generated, providing precise data for subsequent regulation. A flower-time prediction model optimized through transfer learning is used, integrating ambient light spectral data and historical phenological data to accurately predict the flowering time window for each plant. Then, spatiotemporal overlay analysis is employed to screen plants with abnormal flowering times, calculate compensating spectral parameters to generate spectral regulation commands, and finally, directional irradiation is performed using a rotatable array spectral emitter. As the above analysis shows, this invention effectively solves the problem of accurately synchronizing plant flowering times in existing technologies by dynamically sensing and differentially regulating the individual physiological states of different plants, thus improving the accuracy of synchronous regulation of plant flowering times.

[0145] As an improvement to the above scheme, the feature extraction module is specifically used for:

[0146] The acquired canopy multispectral image sequence was subjected to spectral reflectance feature separation processing to extract the fluorescence response spectrum of the red channel and the reflectance feature spectrum of the near-infrared channel, respectively.

[0147] Based on the fluorescence response spectrum, the chlorophyll fluorescence intensity value is calculated using a dynamic threshold segmentation algorithm. At the same time, based on the reflectance feature spectrum, the anthocyanin reflectance feature is analyzed by a convolutional neural network.

[0148] Using the spatial coordinates of each plant as the primary index key, the corresponding chlorophyll fluorescence intensity value and anthocyanin reflectance feature are fused to generate a plant physiological state vector containing spatial location markers.

[0149] As an improvement to the above scheme, the prediction module is specifically used for:

[0150] A dual-branch fusion neural network is constructed, wherein the main branch of the dual-branch fusion neural network is used to process the plant physiological state vector, and the auxiliary branch of the dual-branch fusion neural network is used to analyze the temporal fluctuation characteristics of the ambient light spectrum.

[0151] The flowering phenology records of the same species and latitude as the plant were retrieved from the historical phenology database. The convolution kernel weight parameters of the main branch were corrected using a transfer learning algorithm to obtain the corrected main branch.

[0152] The real-time ambient light spectral distribution is input into the auxiliary branch to generate a spectral influence feature tensor.

[0153] The plant physiological state vector is input into the corrected main branch, and the spectral influence feature tensor is fused through a cross-modal attention mechanism to obtain the spatiotemporal feature fusion matrix.

[0154] Perform time series prediction decoding on the spatiotemporal feature fusion matrix to generate a probability-weighted prediction flowering time window.

[0155] As an improvement to the above solution, the calculation module is specifically used for:

[0156] Construct a spatiotemporal relationship map between the predicted flowering time window of each green plant and its spatial location coordinates;

[0157] A spatiotemporal clustering algorithm based on kernel density estimation is used to calculate the spatiotemporal relationship map to obtain the average flowering period center point and the standard deviation of the flowering period distribution of the population.

[0158] Calculate the spatiotemporal offset between the earliest flowering time of each plant and the center point of the average flowering period of the group, and divide the offset by the standard deviation of the flowering period distribution to generate the flowering lag coefficient.

[0159] Determine whether the absolute value of the flowering lag coefficient exceeds the preset flowering period specific threshold of the green plant. If it does, add the spatial location coordinates of the green plant to the control target set.

[0160] It is understood that the specific embodiments of the above-mentioned dynamic perception-based green plant flowering period synchronization control system can be referred to the relevant embodiments of the above-mentioned dynamic perception-based green plant flowering period synchronization control method, and will not be repeated here.

[0161] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0162] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for synchronously controlling the flowering period of green plants based on dynamic sensing, characterized in that, Includes the following steps: The visible light-near infrared dual-channel imaging device is controlled to scan the green plants in the target green plant area to obtain the canopy multispectral image sequence of the green plants at different spatial locations and the ambient light spectral distribution data of the target green plant area; Based on the canopy image, the chlorophyll fluorescence intensity value and anthocyanin reflectance feature of each plant are extracted to generate a plant physiological state vector containing spatial location markers. The plant physiological state vector is input into the flowering period convolution prediction model optimized by transfer learning, and combined with the ambient light spectral distribution and historical phenological data, the predicted flowering time window for each plant is obtained. Spatiotemporal overlay analysis is performed on the predicted flowering time windows of each of the green plants to calculate the average flowering period of the population. Based on the predicted flowering time windows of each of the green plants and the average flowering period of the population, green plants with abnormal flowering periods are screened into the control target set. For the green plants at various spatial locations in the control target set, calculate the dominant wavelength and photon flux density of the required compensation spectrum, and generate spectral control instructions for the green plants at the corresponding spatial locations. Based on the spectral modulation command, a rotatable array-type spectral emitter is controlled to provide directional illumination to the green plants at the corresponding spatial locations; The step of inputting the plant physiological state vector into a flower period convolutional prediction model optimized through transfer learning, and combining it with the ambient light spectral distribution and historical phenological data to obtain the predicted flowering time window for each plant, includes the following sub-steps: A dual-branch fusion neural network is constructed, wherein the main branch of the dual-branch fusion neural network is used to process the plant physiological state vector, and the auxiliary branch of the dual-branch fusion neural network is used to analyze the temporal fluctuation characteristics of the ambient light spectrum. Retrieve flowering phenology records of the same species and latitude as the plant from the historical phenology database, and use a transfer learning algorithm to correct the convolution kernel weight parameters of the main branch to obtain the corrected main branch. The real-time ambient light spectral distribution is input into the auxiliary branch to generate a spectral influence feature tensor. The plant physiological state vector is input into the corrected main branch, and the spectral influence feature tensor is fused through a cross-modal attention mechanism to obtain the spatiotemporal feature fusion matrix. Perform time series prediction decoding on the spatiotemporal feature fusion matrix to generate a probability-weighted prediction flowering time window.

2. The method for synchronous regulation of green plant flowering period based on dynamic perception as described in claim 1, characterized in that, The step of extracting chlorophyll fluorescence intensity and anthocyanin reflectance features for each plant based on the canopy image to generate a plant physiological state vector containing spatial location markers includes the following sub-steps: The acquired canopy multispectral image sequence was subjected to spectral reflectance feature separation processing to extract the fluorescence response spectrum of the red channel and the reflectance feature spectrum of the near-infrared channel, respectively. Based on the fluorescence response spectrum, the chlorophyll fluorescence intensity value is calculated using a dynamic threshold segmentation algorithm. At the same time, based on the reflectance feature spectrum, the anthocyanin reflectance feature is analyzed by a convolutional neural network. Using the spatial coordinates of each plant as the primary index key, the corresponding chlorophyll fluorescence intensity value and anthocyanin reflectance feature are fused to generate a plant physiological state vector containing spatial location markers.

3. The method for synchronous regulation of green plant flowering period based on dynamic perception as described in claim 2, characterized in that, The step of performing spatiotemporal overlay analysis on the predicted flowering time windows of each of the green plants, calculating the average flowering period of the population, and selecting green plants with abnormal flowering periods to the control target set based on the predicted flowering time windows of each of the green plants and the average flowering period of the population includes the following sub-steps: Construct a spatiotemporal relationship map between the predicted flowering time window of each green plant and its spatial location coordinates; A spatiotemporal clustering algorithm based on kernel density estimation is used to calculate the spatiotemporal relationship map to obtain the average flowering period center point and the standard deviation of the flowering period distribution of the population. Calculate the spatiotemporal offset between the earliest flowering time of each plant and the center point of the average flowering period of the group, and divide the offset by the standard deviation of the flowering period distribution to generate the flowering lag coefficient. Determine whether the absolute value of the flowering lag coefficient exceeds the preset flowering period specific threshold of the green plant. If it does, add the spatial location coordinates of the green plant to the control target set.

4. The method for synchronous regulation of green plant flowering period based on dynamic perception as described in claim 3, characterized in that, The process of calculating the dominant wavelength and photon flux density of the required compensation spectrum for the green plants at various spatial locations within the control target set, and generating spectral control instructions for the green plants at corresponding spatial locations, includes the following sub-steps: The spectral modulation mode of the green plant at the corresponding spatial location is selected according to the sign direction of the flowering lag coefficient. If it is positive, the spectral modulation mode adopts the photo-promoting main wavelength, and if it is negative, the spectral modulation mode adopts the photo-inhibiting main wavelength. Based on the chlorophyll fluorescence intensity value of the green plant at the corresponding spatial location, the photosynthetic enzyme activity index is obtained by matching in the photophysiological response rule base. The absolute value of the flowering lag coefficient and the photosynthetic enzyme activity index are input into the photomorphodynamic equation to calculate the photon flux compensation coefficient of the plant at the corresponding spatial location. By combining the species' light saturation point parameters, the target photon flux density of the green plants at the corresponding spatial locations is calculated using a piecewise function; By integrating spatial location coordinates, the dominant wavelength, the target photon flux density, and the pre-dawn irradiation period, a spectral control command containing wavelength-intensity-temporal parameters is generated for the green plant at the corresponding spatial location.

5. The method for synchronous regulation of green plant flowering period based on dynamic perception as described in claim 4, characterized in that, The step of controlling a rotatable array-type spectral emitter to provide directional illumination to the greenery at the corresponding spatial location based on the spectral modulation command includes the following sub-steps: The spatial position coordinates in the spectral control command are analyzed to calculate the illumination azimuth angle of the rotatable array spectral emitter. Configure the spectral waveform output mode of the rotatable array spectral emitter according to the dominant wavelength parameter; During the pre-dawn illumination period each day, a rotatable array-type spectral emitter is controlled to directionally illuminate the green plants at the corresponding spatial locations according to the illumination azimuth angle, with the directional illumination parameters of the spectrum configured according to the target photon flux density.

6. A system for synchronously controlling the flowering period of green plants based on dynamic sensing, characterized in that, include: The scanning module is used to control the visible light-near infrared dual-channel imaging device to scan the green plants in the target green plant area, and obtain the canopy multispectral image sequence of the green plants at different spatial locations and the ambient light spectral distribution data of the target green plant area; The feature extraction module is used to extract the chlorophyll fluorescence intensity value and anthocyanin reflectance features of each plant based on the canopy image, and generate a plant physiological state vector containing spatial location markers. The prediction module is used to input the plant physiological state vector into the flowering period convolution prediction model optimized by transfer learning, and combine it with the ambient light spectral distribution and historical phenological data to obtain the predicted flowering time window for each plant. The calculation module is used to perform spatiotemporal overlay analysis on the predicted flowering time windows of each of the green plants, calculate the average flowering period of the group, and filter out green plants with abnormal flowering periods to the control target set based on the predicted flowering time windows of each of the green plants and the average flowering period of the group. The instruction generation module is used to calculate the dominant wavelength and photon flux density of the required compensation spectrum for the green plants at various spatial locations in the control target set, and generate spectral control instructions for the green plants at the corresponding spatial locations. The control module is used to control the rotatable array-type spectral emitter to provide directional illumination to the green plants corresponding to the spatial position based on the spectral control command. Specifically, the prediction module is used for: A dual-branch fusion neural network is constructed, wherein the main branch of the dual-branch fusion neural network is used to process the plant physiological state vector, and the auxiliary branch of the dual-branch fusion neural network is used to analyze the temporal fluctuation characteristics of the ambient light spectrum. Retrieve flowering phenology records of the same species and latitude as the plant from the historical phenology database, and use a transfer learning algorithm to correct the convolution kernel weight parameters of the main branch to obtain the corrected main branch. The real-time ambient light spectral distribution is input into the auxiliary branch to generate a spectral influence feature tensor. The plant physiological state vector is input into the corrected main branch, and the spectral influence feature tensor is fused through a cross-modal attention mechanism to obtain the spatiotemporal feature fusion matrix. Perform time series prediction decoding on the spatiotemporal feature fusion matrix to generate a probability-weighted prediction flowering time window.

7. The dynamic sensing-based synchronous control system for green plant flowering period as described in claim 6, characterized in that, The feature extraction module is specifically used for: The acquired canopy multispectral image sequence was subjected to spectral reflectance feature separation processing to extract the fluorescence response spectrum of the red channel and the reflectance feature spectrum of the near-infrared channel, respectively. Based on the fluorescence response spectrum, the chlorophyll fluorescence intensity value is calculated using a dynamic threshold segmentation algorithm. At the same time, based on the reflectance feature spectrum, the anthocyanin reflectance feature is analyzed by a convolutional neural network. Using the spatial coordinates of each plant as the primary index key, the corresponding chlorophyll fluorescence intensity value and anthocyanin reflectance feature are fused to generate a plant physiological state vector containing spatial location markers.

8. The dynamic sensing-based synchronous control system for green plant flowering period as described in claim 7, characterized in that, The calculation module is specifically used for: Construct a spatiotemporal relationship map between the predicted flowering time window of each green plant and its spatial location coordinates; A spatiotemporal clustering algorithm based on kernel density estimation is used to calculate the spatiotemporal relationship map to obtain the average flowering period center point and the standard deviation of the flowering period distribution of the population. Calculate the spatiotemporal offset between the earliest flowering time of each plant and the center point of the average flowering period of the group, and divide the offset by the standard deviation of the flowering period distribution to generate the flowering lag coefficient. Determine whether the absolute value of the flowering lag coefficient exceeds the preset flowering period specific threshold of the green plant. If it does, add the spatial location coordinates of the green plant to the control target set.

Images