Medicinal plant growth cycle evaluation method based on AI image recognition

Abstract

The invention relates to the field of intelligent evaluation and management of a medicinal plant growth cycle, and provides a medicinal plant growth cycle evaluation method based on AI image recognition, which comprises the following steps: acquiring a multi-dimensional image sequence of a plant growth process through a multispectral imaging device, and synchronously recording environment illumination intensity and temperature and humidity data; performing space-time registration on the image sequence, and eliminating pixel offset caused by plant swing and shooting angle change; de-noising by adopting an adaptive hybrid filtering algorithm, and reserving leaf edge texture details; performing multi-scale contrast enhancement on the de-noised image, and reconstructing a high-frequency feature map; constructing a double-branch deep network, and respectively extracting a leaf morphological dynamic feature set and a physiological parameter change feature set; performing cross-modal fusion on the two feature sets, and calculating a dynamic growth index matrix in combination with environmental parameters; and based on matching degree analysis of the dynamic growth index matrix and a preset growth stage model, outputting a stage identifier and a stage transition probability curve of the current growth cycle.

Description

Technical Field

[0001] The present invention relates to the field of intelligent evaluation and management of the growth cycle of medicinal plants. More specifically, the present invention relates to a method for evaluating the growth cycle of medicinal plants based on AI image recognition. Background Art

[0002] The evaluation and management of the growth cycle of medicinal plants are important links in agricultural production and drug research and development. Traditional evaluation methods mainly rely on manual observation and empirical judgment, making it difficult to achieve precise monitoring and dynamic analysis of the growth state of plants. With the development of multispectral imaging technology and deep learning algorithms, the method for evaluating the growth cycle of plants based on image recognition has gradually become a research hotspot. However, there are still many deficiencies in the existing technology in terms of multi-dimensional image acquisition, feature extraction, and growth stage division. For example, the spatio-temporal registration accuracy of multispectral images is relatively low, making it difficult to eliminate the influence of plant swaying and shooting angle changes; during the feature extraction process, the fusion effect of leaf morphological features and physiological parameter features is not good, resulting in insufficient accuracy of the growth index matrix; in addition, the existing growth stage models have poor adaptability to complex environmental factors, making it difficult to achieve precise stage division and prediction.

[0003] In the process of implementing the embodiments of the present invention, the inventors found that there are at least the following problems or defects in the existing technology: insufficient multi-dimensional image acquisition and registration accuracy, poor feature extraction and fusion effects, and poor adaptability of the growth stage model to complex environments. Summary of the Invention

[0004] The present invention provides a method for evaluating the growth cycle of medicinal plants based on AI image recognition, including: S1. Collect a multi-dimensional image sequence of the growth process of medicinal plants through a multispectral imaging device, and synchronously record the environmental light intensity and temperature and humidity data; S2. Perform spatio-temporal registration operations on the multi-dimensional image sequence to eliminate pixel offsets caused by plant swaying and shooting angle changes; S3. Use an adaptive hybrid filtering algorithm to eliminate noise from the registered images and retain the leaf edge texture details; S4. Perform multi-scale contrast enhancement on the denoised images, and reconstruct the high-frequency feature maps through Laplacian pyramid decomposition; S5. Construct a two-branch deep network, where the first branch extracts the dynamic morphological feature set of the leaves, and the second branch extracts the feature set of physiological parameter changes; S6. Perform cross-modal fusion on the morphological dynamic feature set and the physiological parameter change feature set, and calculate the dynamic growth index matrix in combination with the environmental parameters; S7. Output the current growth cycle stage identifier and the stage transition probability curve based on the matching degree analysis between the dynamic growth index matrix and the preset growth stage model.

[0005] Further, the specific steps of extracting the dynamic morphological feature set in step S5 include: S51. Perform dynamic deformation perception convolution operation on the visible light band image sequence. The convolution kernel adaptively adjusts the spatial sampling position according to the leaf growth direction to generate the leaf contour deformation field. S52. Calculate the three-dimensional curvature integral value based on the deformation field to establish the leaf curl degree. S53. Jointly encode the leaf curl degree index and the vein fractal dimension to generate the morphological dynamic feature set.

[0006] Further, the specific steps of extracting the physiological parameter change feature set in step S5 include: S54. Construct a spectral reflectance dynamic decoupling model to separate the chlorophyll absorption band and the overtone vibration signal of water molecules. S55. Calculate the metabolic activity index according to the spectral fingerprint library of medicinal plant species. S56. Perform time-domain cross-correlation analysis on the metabolic activity index and the transpiration rate volatility to generate the physiological parameter change feature set.

[0007] Further, the cross-modal fusion in step S6 adopts: S61. Construct a growth trend coupling model to map the morphological dynamic feature set to the physiological parameter space and calculate the feature compatibility matrix. S62. Generate a growth constraint factor based on environmental parameters. The factor includes the modulation coefficient of light intensity on the photosynthetic phase and the damping coefficient of temperature and humidity on the metabolic rate. S63. Perform multi-source feature constraint optimization as shown in the following formula:

[0008] Among them, is the fused dynamic growth index matrix; is the morphological dynamic feature set; is the physiological parameter change feature set; . is the adaptive balance parameter; is the environmental constraint function; is the growth index matrix to be optimized; is the set of environment-related factors.

[0009] Further, the calculation of the environmental constraint function includes: S611. Establish the phase modulation equation of photosynthetically active radiation as shown in the following formula:

[0010] where, is the light phase modulation factor; L is the real-time light intensity; is the light saturation intensity of the species; is the day-night cycle parameter; t is the current time; S612. Construct the metabolic damping term as shown in the following formula:

[0011] where, is the temperature-humidity damping factor; T is the environmental temperature; is the optimum growth temperature; is the temperature tolerance range; H is the environmental humidity; is the humidity critical threshold.

[0012] Furthermore, the specific implementation of the step S7 includes: S71. Construct a quantization model for the growth stage, map the dynamic growth index matrix to the Hilbert space, and generate discrete energy level basis vectors; S72. Calculate the coherence degree between the current dynamic growth index matrix and each energy level basis vector; S73. Generate a stage transition probability curve according to the coherence degree, and use the Hilbert space basis vector as the topological constraint of the state transition path.

[0013] Furthermore, the calculation method of the stage transition probability curve includes: S74. Calculate the barrier height and the thermal fluctuation energy ; S75. Establish a quantum tunneling rate equation as shown in the following formula:

[0014] where, is the stage transition rate; is the intrinsic transition rate; is the barrier height; is the thermal fluctuation energy; is the Boltzmann constant; T is the environmental temperature.

[0015] Furthermore, the adaptive hybrid filtering algorithm of the step S3 includes: S31. Perform leaf-guided anisotropic diffusion filtering; S32. Perform morphological wavelet packet decomposition in the frequency domain and retain the sub-band components that match the leaf edge vibration frequency.

[0016] Further, the dynamic deformation perception convolution operation includes: S511. Generate the convolution kernel deformation offset according to the leaf growth direction field; S512. Apply the offset to the convolution kernel coordinates through differentiable sampling to generate a deformed feature map.

[0017] Further, the growth stage quantization model includes: S711. Construct a growth Hilbert space, the basis vectors of which are composed of photosynthesis eigenstates; S712. Introduce the environmental parameter perturbation Hamiltonian to split the energy levels of the Hilbert space basis vectors; S713. Solve the eigenstates of the Hamiltonian through the imaginary time evolution algorithm to determine the stage transition path.

[0018] According to the above embodiments of the present invention, it has at least the following beneficial effects: The present invention can improve the accuracy and efficiency of the growth cycle assessment of medicinal plants. Through the combination of multispectral imaging technology and deep learning algorithms, the morphological and physiological parameter changes of plants can be monitored in real time, so as to accurately identify their growth stages. This method can avoid the subjectivity and errors of traditional manual observation and provide a reliable basis for the scientific management of medicinal plants. In addition, through the multi-dimensional analysis of environmental parameters, the growth cycle model can be dynamically adjusted to adapt to the growth needs of plants under different climate and soil conditions, further improving the applicability and practicality of the assessment results.

[0019] The present invention can also optimize the harvesting time of medicinal plants and the drug R & D process. Through accurate growth stage division and dynamic growth index analysis, the best period for the accumulation of plant active ingredients can be determined, thereby improving the quality and efficacy of medicinal materials. At the same time, based on the prediction function of the growth cycle model, the planting and harvesting plans can be planned in advance, reducing resource waste and improving production efficiency. This technology can provide high-quality plant raw materials for drug R & D and accelerate the new drug development process, promoting the sustainable utilization of medicinal plant resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the drawings, several embodiments of the present invention are shown in an exemplary rather than restrictive manner, wherein: Figure 1 It is a schematic flowchart of a method for evaluating the growth cycle of medicinal plants based on AI image recognition provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and then implement the present invention, rather than limiting the scope of the present invention in any way. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to be able to fully convey the scope of the present invention to those skilled in the art.

[0022] Those skilled in the art know that the embodiments of the present invention can be implemented as a system, device, equipment, method, or computer program product. Therefore, the present invention can be specifically implemented in the following forms, namely: completely hardware, completely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0023] It should be noted that any number of elements in the drawings is for illustration rather than limitation, and any naming is only for distinction and does not have any limiting meaning.

[0024] The following refers to Figure 1 , Figure 1 which is a schematic flow chart of a method for evaluating the growth cycle of medicinal plants based on AI image recognition provided for an embodiment of the present invention. As Figure 1 shown, a method for evaluating the growth cycle of medicinal plants based on AI image recognition includes: S1. Collect a multi-dimensional image sequence of the growth process of medicinal plants through a multi-spectral imaging device, and synchronously record environmental light intensity and temperature and humidity data; S2. Perform spatio-temporal registration operations on the multi-dimensional image sequence to eliminate pixel offsets caused by plant swaying and shooting angle changes; S3. Use an adaptive hybrid filtering algorithm to eliminate noise from the registered images and retain leaf edge texture details; S4. Perform multi-scale contrast enhancement on the denoised images, and reconstruct the high-frequency feature map through Laplacian pyramid decomposition; S5. Construct a two-branch deep network, where the first branch extracts a set of dynamic morphological features of the leaves, and the second branch extracts a set of features of physiological parameter changes; S6. Perform cross-modal fusion on the set of dynamic morphological features and the set of physiological parameter changes, and calculate a dynamic growth index matrix in combination with environmental parameters; S7. Based on the matching degree analysis between the dynamic growth index matrix and a preset growth stage model, output the current growth cycle stage identifier and the stage transition probability curve.

[0025] It should be noted that in the present invention, a multi-spectral imaging device is used to collect multi-dimensional image sequences of the growth process of medicinal plants, and the ambient light intensity and temperature and humidity data are recorded synchronously. The multi-spectral imaging device is a device that can simultaneously obtain image information in multiple bands. It can capture the image characteristics of medicinal plants in different bands, thus providing richer information for subsequent analysis. The multi-dimensional image sequences refer to the image sequences obtained at different times, different angles, and different lighting conditions. These image sequences can comprehensively reflect the growth state of medicinal plants. The ambient light intensity refers to the light intensity in the plant growth environment. The light intensity has an important impact on the photosynthesis and growth and development of plants. The temperature and humidity data refer to the temperature and humidity data in the environment. Temperature and humidity also have an important impact on the growth and development of plants. By synchronously recording these data, the relationship between plant growth and environmental factors can be better analyzed.

[0026] Specifically, the multi-spectral imaging device can include various types of cameras such as visible light cameras, near-infrared cameras, and mid-infrared cameras. These cameras can respectively obtain image information in different bands. For example, a visible light camera can obtain images of medicinal plants in the visible light band for observing the morphological characteristics of the plants; a near-infrared camera can obtain images of plants in the near-infrared band for analyzing the physiological state of the plants. When collecting image sequences, different shooting times and shooting angles can be set according to the growth cycle and growth characteristics of the medicinal plants. For example, in the initial stage of plant growth, images can be taken at regular intervals to observe the germination and growth of the plants; in the middle stage of plant growth, the shooting frequency can be increased to better monitor the growth dynamics of the plants. The ambient light intensity can be measured by a light sensor, and the temperature and humidity data can be measured by temperature and humidity sensors. These sensors can work synchronously with the multi-spectral imaging device to ensure the accuracy and consistency of the data. The measurement range of the light sensor can be adjusted according to the actual ambient light intensity. For example, in an indoor planting environment, the light intensity range may be between 100 and 1000 lux; in an outdoor planting environment, the light intensity range may be between 1000 and 100000 lux. The measurement range of the temperature and humidity sensor is usually -20°C to 50°C for temperature and 0% to 100% RH for humidity. These parameters can be optimized according to the specific planting environment.

[0027] Preferably, in order to improve the accuracy and reliability of image acquisition, an autofocus function can be introduced into the multispectral imaging device to ensure that clear images can be obtained in each shot. At the same time, the acquired image sequence can be preprocessed, for example, the random noise in the image can be removed through a denoising algorithm, and the color deviation of the image can be corrected through a color correction algorithm to improve the quality and usability of the image. In addition, in order to better synchronously record the ambient light intensity and temperature and humidity data, a timestamp synchronization mechanism can be set between the sensor and the imaging device to ensure the consistency of the acquired image data and environmental data in time.

[0028] In some embodiments, the specific steps of extracting the morphological dynamic feature set in step S5 include: S51. Perform a dynamic deformation perception convolution operation on the visible light band image sequence, and the convolution kernel adaptively adjusts the spatial sampling position according to the leaf growth direction to generate a leaf contour deformation field; S52. Calculate the three-dimensional curvature integral value based on the deformation field, and establish a leaf curl index as shown in the following formula:

[0029] where, is the leaf curl index; are the first principal curvature and the second principal curvature of the leaf surface respectively; S is the leaf surface integral region; is the area differential; S53. Jointly encode the leaf curl index and the vein fractal dimension to generate the morphological dynamic feature set.

[0030] It should be noted that the specific steps of extracting the morphological dynamic feature set in the present invention include performing a dynamic deformation perception convolution operation on the visible light band image sequence, and the convolution kernel of this operation will adaptively adjust the spatial sampling position according to the leaf growth direction, thereby generating a leaf contour deformation field. The leaf contour deformation field refers to a feature field that reflects the dynamic growth changes of the leaf by calculating the shape changes of the leaf at different time points. Based on this deformation field, the three-dimensional curvature integral value is calculated, and a leaf curl index is established to quantify the curling degree of the leaf. The vein fractal dimension is used to describe the complexity and distribution characteristics of the veins. The leaf curl index and the vein fractal dimension are jointly encoded to finally generate the morphological dynamic feature set for subsequent analysis and processing.

[0031] Specifically, the dynamic deformation perception convolution operation is a special convolution operation that can dynamically adjust the sampling position of the convolution kernel according to the growth direction of the leaf. In the visible light band image sequence, the growth direction of the leaf can be extracted through image processing algorithms, such as edge detection and morphological analysis. The sampling position of the convolution kernel will be adjusted according to the leaf growth direction, enabling the convolution operation to better capture the morphological changes of the leaf. The three-dimensional curvature integral value is obtained by calculating the curvature of the leaf surface, which can reflect the degree of leaf bending. In the calculation process, the first principal curvature and the second principal curvature of the leaf surface are key parameters, which can be obtained through geometric analysis of the image. The calculation of the vein fractal dimension involves the analysis of the complexity of the vein structure and can be achieved through relevant algorithms in fractal theory. The process of joint encoding is to fuse the two features of the leaf curl index and the vein fractal dimension to form a comprehensive morphological dynamic feature set for more comprehensively describing the morphological features of the leaf.

[0032] Preferably, when performing the dynamic deformation perception convolution operation, a more refined leaf growth direction detection algorithm can be introduced, such as combining the convolutional neural network (CNN) in deep learning to more accurately identify the growth direction of the leaf. This can further improve the accuracy of the convolution kernel sampling position, thereby better capturing the morphological changes of the leaf. When calculating the three-dimensional curvature integral value, a multi-scale analysis method can be adopted to calculate the curvature of the leaf at different scales to more comprehensively reflect the degree of leaf bending. In addition, for the calculation of the vein fractal dimension, different fractal analysis methods, such as the box-counting method or the correlation fractal dimension method, can be considered to adapt to different types of vein structures. In the process of joint encoding, more advanced feature fusion techniques, such as the fusion method based on the attention mechanism, can be adopted to better highlight important features and improve the quality and effectiveness of the morphological dynamic feature set.

[0033] In some embodiments, the specific steps of extracting the physiological parameter change feature set in step S5 include: S54. Construct a spectral reflectance dynamic decoupling model to separate the chlorophyll absorption band and the overtone vibration signal of water molecules; S55. According to the spectral fingerprint library of the medicinal plant species, calculate the metabolic activity index as shown in the following formula:

[0034] where, is the metabolic activity index; is the spectral reflectance function at time t; is the species-specific absorption template; 、 is the characteristic spectral range; S56. Perform a time-domain cross-correlation analysis on the metabolic activity index and the transpiration rate volatility to generate the physiological parameter change feature set.

[0035] It should be noted that the specific steps for extracting the physiological parameter change feature set in the present invention include constructing a spectral reflectance dynamic decoupling model to separate the chlorophyll absorption band and the overtone vibration signal of water molecules. The spectral reflectance dynamic decoupling model is a model for analyzing the change of plant spectral reflectance. It can decompose complex spectral reflectance signals into different components, so as to extract features related to the physiological state of plants. The chlorophyll absorption band refers to the specific wavelength bands in the spectrum that chlorophyll absorbs light. The change of reflectance in these bands is closely related to the photosynthesis state of plants. The overtone vibration signal of water molecules is related to the water state of plants. By separating these signals, the physiological state of plants can be evaluated more accurately. Based on the spectral fingerprint library of medicinal plant species, calculate the metabolic activity index, which is used to quantify the metabolic activity level of plants. Finally, perform a time-domain cross-correlation analysis on the metabolic activity index and the transpiration rate volatility to generate the physiological parameter change feature set for subsequent growth cycle assessment.

[0036] Specifically, the construction of the spectral reflectance dynamic decoupling model requires a large amount of spectral data, which can be obtained from multispectral imaging devices. The model separates the chlorophyll absorption band and the overtone vibration signal of water molecules by analyzing the change of spectral reflectance in different bands. The chlorophyll absorption band is usually located in the red and near-infrared wavelength bands, while the overtone vibration signal of water molecules appears in specific mid-infrared wavelength bands. The calculation of the metabolic activity index requires the use of the spectral fingerprint library of medicinal plant species, which contains the spectral characteristics of different species in different physiological states. By comparing with the data in the fingerprint library, the metabolic activity index can be calculated, which reflects the intensity of the metabolic activity of plants at a specific time. The transpiration rate volatility refers to the change rate of plant transpiration over time, which can be calculated by measuring the water evaporation amount of plant leaves. The time-domain cross-correlation analysis is a statistical method for analyzing the correlation between two time series. Through this method, the metabolic activity index and the transpiration rate volatility can be combined to generate the physiological parameter change feature set.

[0037] Preferably, in order to improve the accuracy of the spectral reflectance dynamic decoupling model, machine learning algorithms such as support vector machine (SVM) or convolutional neural network (CNN) in deep learning can be introduced to train and optimize the model. These algorithms can learn more accurate decoupling patterns based on a large amount of sample data, so as to more effectively separate the chlorophyll absorption band and the overtone vibration signal of water molecules. When calculating the metabolic activity index, more spectral feature parameters such as spectral slope and spectral curvature can be considered to more comprehensively reflect the metabolic state of plants. In addition, for the measurement of transpiration rate volatility, more advanced sensor technologies such as high-precision humidity sensors and temperature sensors can be used to improve the accuracy and reliability of the measurement. In the time-domain cross-correlation analysis, more complex statistical models such as autoregressive moving average model (ARMA) or wavelet analysis method can be used to more accurately capture the dynamic relationship between the metabolic activity index and the transpiration rate volatility, so as to generate a higher-quality set of physiological parameter change features.

[0038] In some embodiments, the cross-modal fusion in step S6 adopts: S61. Construct a growth trend coupling model, map the morphological dynamic feature set to the physiological parameter space, and calculate the feature compatibility matrix; S62. Generate a growth constraint factor based on environmental parameters, where the factor includes the modulation coefficient of light intensity on the photosynthetic phase and the damping coefficient of temperature and humidity on the metabolic rate; S63. Perform multi-source feature constraint optimization as shown in the following formula:

[0039] Where, is the fused dynamic growth index matrix; is the morphological dynamic feature set; is the set of physiological parameter change features; , is the adaptive balance parameter; is the environmental constraint function; is the growth index matrix to be optimized; is the set of environmental-related factors.

[0040] It should be noted that in the present invention, the cross-modal fusion step is achieved by constructing a growth trend coupling model. This model maps the morphological dynamic feature set to the physiological parameter space, calculates the feature compatibility matrix, so as to evaluate the matching degree between the morphological features and physiological parameters. The growth constraint factors generated by environmental parameters include the modulation coefficient of light intensity on the photosynthetic phase and the damping coefficient of temperature and humidity on the metabolic rate, and these factors are used to reflect the influence of environmental conditions on plant growth. Finally, through multi-source feature constraint optimization, the morphological dynamic feature set, the physiological parameter change feature set, and the environmental constraint factors are comprehensively processed to generate a fused dynamic growth index matrix for evaluating the growth state of medicinal plants.

[0041] Specifically, the growth trend coupling model is a mathematical model used to convert the information in the morphological dynamic feature set into the physiological parameter space for comprehensive analysis. The feature compatibility matrix is obtained by calculating the similarity or correlation between the morphological features and physiological parameters, and it reflects the matching degree between the two features. The growth constraint factors generated by environmental parameters are calculated based on environmental conditions such as light intensity, temperature, and humidity. The modulation coefficient of light intensity on the photosynthetic phase reflects the influence of light intensity on plant photosynthesis, while the damping coefficient of temperature and humidity on the metabolic rate reflects the inhibitory or promoting effect of temperature and humidity on plant metabolic rate. Multi-source feature constraint optimization is a comprehensive processing process. By balancing the relationship between the morphological dynamic feature set, the physiological parameter change feature set, and the environmental constraint factors, a fused dynamic growth index matrix is obtained, and this matrix can more comprehensively reflect the growth state of medicinal plants.

[0042] Preferably, in order to improve the accuracy of the growth trend coupling model, deep learning methods such as neural networks can be used to learn the complex mapping relationship between the morphological dynamic feature set and physiological parameters. When calculating the feature compatibility matrix, more feature similarity measurement methods such as cosine similarity and Euclidean distance can be introduced to more accurately evaluate the matching degree between features. For the growth constraint factors generated by environmental parameters, more environmental parameters such as carbon dioxide concentration and soil humidity can be considered to more comprehensively reflect the influence of the environment on plant growth. In the process of multi-source feature constraint optimization, more advanced optimization algorithms such as genetic algorithms or particle swarm optimization algorithms can be used to more effectively balance the relationship between different feature sets, so as to obtain a more accurate dynamic growth index matrix.

[0043] In some embodiments, the calculation of the environmental constraint function includes: S611. Establish a phase modulation equation for photosynthetically active radiation as shown in the following formula:

[0044] Among them, is the light phase modulation factor; L is the real-time light intensity; is the light saturation intensity of the species; is the day-night cycle parameter; t is the current time; S612. Construct a metabolic damping term as shown in the following formula:

[0045] Among them, is the temperature-humidity damping factor; T is the environmental temperature; is the optimal growth temperature; is the temperature tolerance range; H is the environmental humidity; is the humidity critical threshold.

[0046] It should be noted that the calculation of the environmental constraint function in the present invention is realized based on the phase modulation equation of photosynthetically active radiation and the metabolic damping term. The phase modulation equation of photosynthetically active radiation is used to describe the influence of light intensity on the photosynthesis phase of plants, while the metabolic damping term is used to describe the inhibitory effect of environmental temperature and humidity on the metabolic rate of plants. Through these two equations, environmental constraint factors including the modulation coefficient of light intensity on the photosynthesis phase and the damping coefficient of temperature and humidity on the metabolic rate can be generated, and these factors can quantify the influence of environmental conditions on plant growth.

[0047] Specifically, the phase modulation equation of photosynthetically active radiation is a mathematical model that takes into account factors such as real-time light intensity, light saturation intensity of the species, day-night cycle parameter, and current time, and is used to calculate the light phase modulation factor. This factor reflects the regulatory effect of light intensity changes on the photosynthesis phase of plants. For example, when the light intensity is low, the photosynthesis phase of plants may be delayed. The metabolic damping term is a function related to environmental temperature and humidity, which takes into account factors such as environmental temperature, optimal growth temperature, temperature tolerance range, environmental humidity, and humidity critical threshold, and is used to calculate the temperature-humidity damping factor. This factor reflects the inhibitory or promoting effect of temperature and humidity on the metabolic rate of plants. For example, when the temperature is too high or the humidity is too low, the metabolic rate of plants may decrease. The environmental constraint factors calculated by these two equations can provide important environmental parameter support for subsequent growth constraint analysis.

[0048] Preferably, when calculating the phase modulation equation of photosynthetically active radiation, more environmental light parameters, such as the spectral distribution of light, can be introduced to more accurately reflect the impact of light on plant photosynthesis. For the metabolic damping term, other environmental factors such as soil moisture can be considered to more comprehensively evaluate the impact of the environment on plant metabolism. In addition, machine learning methods can be used to model and optimize the environmental constraint function. By training the model with a large amount of environmental data and plant growth data, it can more accurately predict the environmental constraint factors. For example, algorithms such as random forest or neural network can be used to learn the complex relationship between environmental parameters and plant growth based on historical data, thereby improving the accuracy and reliability of the environmental constraint function.

[0049] In some embodiments, the specific implementation of step S7 includes: S71. Construct a quantization model for the growth stage, map the dynamic growth index matrix to the Hilbert space, and generate discrete energy level basis vectors; S72. Calculate the coherence degree between the current dynamic growth index matrix and each energy level basis vector as shown in the following formula:

[0050] where, is the coherence degree of the k-th energy level; is the growth state density matrix; is the projection operator of the k-th energy level; represents the matrix trace operation; S73. Generate a phase transition probability curve based on the coherence degree, and use the Hilbert space basis vector as the topological constraint of the state transition path.

[0051] It should be noted that the generation of the phase transition probability curve in the present invention is realized based on the quantization model of the Hilbert space. The Hilbert space is a mathematical space used to describe the evolution and interaction of quantum states. In the present invention, by mapping the dynamic growth index matrix to the Hilbert space, discrete energy level basis vectors are generated. These energy level basis vectors are used to represent different stages of plant growth, and the coherence degree is used to measure the similarity between the current growth state and each energy level basis vector. By calculating the coherence degree, a phase transition probability curve can be generated, which reflects the possibility of a plant transitioning from one growth stage to another.

[0052] Specifically, the construction of the Hilbert space is based on the eigenstates of photosynthesis, which reflect the photosynthesis characteristics of plants at different growth stages. The environmental parameter perturbation Hamiltonian is used to introduce the influence of environmental factors on plant growth. By splitting the energy levels of the Hilbert space basis vectors, the growth state of plants under different environmental conditions can be more accurately reflected. The imaginary-time evolution algorithm is a method for solving the eigenstates of the Hamiltonian. Through this algorithm, the phase transition path can be determined, that is, the possible path for plants to transition from one growth stage to another. The calculation of the coherence degree is achieved through matrix trace operation, which reflects the similarity between the growth state density matrix and the energy level projection operator. The generation of the phase transition probability curve is based on the calculation results of the coherence degree. By analyzing the barrier height and thermal fluctuation energy between adjacent energy levels, the calculation of the phase transition probability curve can be further optimized.

[0053] Preferably, to improve the accuracy of the Hilbert space model, more environmental parameters can be introduced, such as soil humidity, carbon dioxide concentration, etc., to more comprehensively reflect the influence of the environment on plant growth. When constructing the quantization model of the growth stage, more advanced quantum mechanics methods, such as the quantum Monte Carlo method, can be adopted to more accurately solve the eigenstates of the Hamiltonian. In addition, the calculation of the phase transition probability curve can be combined with machine learning algorithms, such as Bayesian networks or Markov chain Monte Carlo methods, to more accurately predict the transition probability of plant growth stages. For example, by analyzing the growth stage transition patterns in historical data, a machine learning model can be trained to improve the prediction accuracy of the phase transition probability curve.

[0054] In some embodiments, the calculation method of the phase transition probability curve includes: S74. Calculating the barrier height between adjacent energy levels based on the Hilbert space basis vectors and the thermal fluctuation energy ; S75. Establishing a quantum tunneling rate equation as shown in the following formula:

[0055] where, is the phase transition rate; is the eigen-transition rate; is the barrier height; is the thermal fluctuation energy; is the Boltzmann constant; T is the environmental temperature.

[0056] It should be noted that the calculation of the stage transition probability curve in the present invention is completed based on the basis vectors of the Hilbert space. The basis vectors of the Hilbert space are a concept in the quantization model, which are used to represent different stages of plant growth. By calculating the barrier height between adjacent energy levels and the thermal fluctuation energy, a quantum tunneling rate equation can be established. This equation is used to describe the rate at which a plant transitions from one growth stage to another, taking into account the influence of factors such as environmental temperature on the transition process. Through these calculations, the transition probability of the plant growth stage can be predicted more accurately.

[0057] Specifically, the basis vectors of the Hilbert space are generated by the growth stage quantization model, and these basis vectors reflect the characteristic states of the plant at different growth stages. The barrier height between adjacent energy levels represents the energy barrier required for the plant to transition from one stage to another, while the thermal fluctuation energy is related to the environmental temperature and reflects the promoting effect of thermal energy on the transition of the plant growth stage. The quantum tunneling rate equation is an equation based on the principles of quantum mechanics, which combines factors such as the barrier height, thermal fluctuation energy, and environmental temperature to calculate the rate of plant growth stage transition. The intrinsic transition rate is a constant related to the plant species, representing the rate of plant growth stage transition under ideal conditions. Through the calculation and analysis of these parameters, the stage transition probability curve can be obtained, providing an important basis for the evaluation of the plant growth cycle.

[0058] Preferably, when calculating the barrier height between adjacent energy levels, more plant physiological parameters, such as the metabolic activity index of the plant, can be considered to more comprehensively reflect the changes in the plant growth state. For the calculation of the thermal fluctuation energy, specific environmental conditions, such as humidity and light intensity, can be combined to more accurately evaluate the influence of environmental factors on the transition of the plant growth stage. In addition, the parameters in the quantum tunneling rate equation can be calibrated and optimized through experimental data to improve the accuracy and reliability of the model. For example, by conducting growth experiments on medicinal plants of different species and collecting relevant data on stage transitions, the optimal values of parameters such as the intrinsic transition rate can be determined.

[0059] In some embodiments, the adaptive hybrid filtering algorithm in step S3 includes: S31. Perform leaf-guided anisotropic diffusion filtering, and the diffusion tensor is shown in the following formula:

[0060] where D is the diffusion tensor; ∇ is the gradient operator; I is the image gray value; 、 are the smoothing intensity parameters in the directions of the main vein and secondary veins of the leaf, respectively; S32. Perform morphological wavelet packet decomposition in the frequency domain and retain the sub-band components matching the vibration frequency of the leaf margin.

[0061] It should be noted that the adaptive hybrid filtering algorithm in step S3 of the present invention includes two main operations: leaf-guided anisotropic diffusion filtering and morphological wavelet packet decomposition in the frequency domain. Leaf-guided anisotropic diffusion filtering is an image processing technique aimed at smoothing according to the structural characteristics of the leaf while retaining the texture details of the leaf edge. Morphological wavelet packet decomposition is a frequency domain analysis method used to extract sub-band components matching the vibration frequency of the leaf edge, thereby further enhancing the feature information of the image. By combining these two methods, important image details can be retained while removing noise, providing high-quality input for subsequent image analysis.

[0062] Specifically, leaf-guided anisotropic diffusion filtering is achieved by defining a diffusion tensor, which adjusts the smoothing intensity according to the directions of the main veins and secondary veins of the leaf. The smoothing intensity parameter determines the degree of smoothing in different directions. For example, the smoothing intensity along the direction of the main vein of the leaf may be higher to reduce noise, while the smoothing intensity perpendicular to the main vein remains lower to retain edge details. Morphological wavelet packet decomposition is carried out in the frequency domain. By performing multi-layer decomposition on the image, sub-band components in different frequency ranges are extracted. The vibration frequency of the leaf edge refers to the tiny vibration frequency that may occur in the natural growth process of the leaf edge. By retaining the sub-band components matching these frequencies, the feature information of the leaf edge in the image can be enhanced, making it more suitable for subsequent morphological analysis.

[0063] Preferably, when implementing leaf-guided anisotropic diffusion filtering, an adaptive method can be used to dynamically adjust the smoothing intensity parameter. For example, according to the local structural characteristics of the leaf in the image, the smoothing intensity is automatically adjusted to better adapt to the noise level and texture characteristics of different regions. In morphological wavelet packet decomposition, more decomposition layers can be introduced to more finely extract the feature information in different frequency ranges. In addition, machine learning methods, such as convolutional neural networks (CNNs), can be combined to perform feature extraction and analysis on the decomposed sub-band components, further improving the quality of the image features. For example, by training a CNN model to identify and enhance the features related to the growth state of the leaf, more accurate image data can be provided for subsequent growth cycle assessment.

[0064] In some embodiments, the dynamic deformation perception convolution operation includes: S511. Generate the convolution kernel deformation offset according to the leaf growth direction field as shown in the following formula:

[0065] Wherein, is the convolution kernel offset; is the deformation sensitivity parameter; is the vein gradient field; is the random perturbation term; S512. Apply the offset to the convolutional kernel coordinates through differentiable sampling to generate a deformed feature map.

[0066] It should be noted that in the present invention, the dynamic deformation perception convolution operation generates a deformed feature map by generating a convolutional kernel deformation offset according to the leaf growth direction field and applying the offset to the convolutional kernel coordinates. Dynamic deformation perception convolution is an advanced image processing technology aimed at dynamically adjusting the shape and position of the convolutional kernel according to the growth direction of the leaf, so as to more accurately capture the morphological changes of the leaf. The convolutional kernel deformation offset is calculated based on the leaf growth direction field, which reflects the growth direction differences of the leaf at different positions. By applying the offset to the convolutional kernel coordinates through differentiable sampling, a feature map consistent with the leaf growth direction can be generated, thus better reflecting the dynamic changes of the leaf.

[0067] Specifically, the leaf growth direction field refers to the distribution of the growth directions of the leaf at different positions, which can be extracted from the multi-spectral image through an image processing algorithm. The calculation of the convolutional kernel deformation offset involves a deformation sensitivity parameter and the vein gradient field. The deformation sensitivity parameter is a regulating factor used to control the degree of convolutional kernel deformation; the vein gradient field reflects the distribution and direction of the veins inside the leaf and is an important basis for calculating the offset. By applying the offset to the convolutional kernel coordinates, the dynamic deformation of the convolutional kernel can be realized, enabling it to better adapt to the growth direction of the leaf. Differentiable sampling is a numerical calculation method that allows dynamically adjusting the sampling position in the convolution operation to generate a deformed feature map. This feature map can more accurately reflect the morphological changes of the leaf and provide higher-quality input for subsequent morphological feature extraction.

[0068] Preferably, when calculating the convolutional kernel deformation offset, more image features can be introduced, such as the texture features and color features of the leaf, to more comprehensively reflect the growth state of the leaf. For example, combining texture features can better capture the subtle changes on the leaf surface, while color features help distinguish different types of leaf tissues. When implementing differentiable sampling, more advanced numerical calculation methods, such as Gaussian sampling or adaptive sampling, can be adopted to improve the accuracy and efficiency of sampling. In addition, deep learning technologies, such as convolutional neural networks (CNNs), can also be combined to optimize the dynamic deformation perception convolution operation. By training the CNN model, the optimal convolutional kernel deformation strategy can be automatically learned, further improving the accuracy and robustness of feature extraction.

[0069] In some embodiments, the growth stage quantization model includes: S711. Construct a growth Hilbert space, the basis vectors of which are composed of photosynthesis eigenstates; S712. Introduce the environmental parameter perturbation Hamiltonian , and split the energy levels of the Hilbert space basis vectors; S713. Solve the Hamiltonian 's eigenstates through the imaginary-time evolution algorithm, and determine the phase transition path as shown in the following formula:

[0070] where, is the quantum state at time t; is the initial state; e is the natural exponential function.

[0071] It should be noted that the construction of the quantization model of the growth stage in the present invention is realized by introducing the environmental parameter perturbation Hamiltonian. The Hamiltonian is a basic concept in quantum mechanics, which is used to describe the energy state of a system. In the present invention, the environmental parameter perturbation Hamiltonian is used to consider the influence of environmental factors on the plant growth stage. By solving the eigenstates of the Hamiltonian through the imaginary-time evolution algorithm, the transition path of the plant growth stage can be determined. This method introduces the concept of quantum mechanics into the evaluation of the plant growth stage, providing a new theoretical basis for the dynamic evaluation of the plant growth cycle.

[0072] Specifically, the quantization model of the growth stage is constructed based on the Hilbert space of plant growth, where the basis vectors of the Hilbert space are composed of the eigenstates of photosynthesis. These eigenstates reflect the photosynthesis characteristics of plants at different growth stages. The environmental parameter perturbation Hamiltonian considers the influence of environmental factors such as light intensity, temperature, and humidity on plant growth. By introducing the perturbation Hamiltonian, the energy levels of the basis vectors of the Hilbert space can be split, so as to more accurately reflect the growth state of plants under different environmental conditions. The imaginary-time evolution algorithm is a numerical method for solving the eigenstates of the Hamiltonian. Through this algorithm, the transition path of the plant growth stage can be determined. This method can map the growth state of plants to the evolution process of quantum states, thus providing a new perspective for the dynamic evaluation of the plant growth stage.

[0073] Preferably, when constructing the quantization model of the growth stage, more environmental parameters can be introduced, such as soil humidity, carbon dioxide concentration, etc., to more comprehensively reflect the influence of the environment on plant growth. When introducing the environmental parameter perturbation Hamiltonian, a more complex quantum mechanics model, such as the many-body interaction model, can be considered to more accurately describe the physical and chemical changes in the plant growth process.

[0074] Furthermore, the parameter settings of the imaginary-time evolution algorithm can be optimized according to specific plant species and growth environments. For example, by calibrating the time step and evolution time in the algorithm with experimental data, the accuracy and reliability of the model can be improved. Machine learning techniques, such as generative adversarial networks (GANs) in deep learning, can also be combined to optimize and validate the quantization model of the growth stage. Through these methods, the accuracy and robustness of plant growth stage assessment can be further enhanced.

[0075] The above-mentioned various embodiments of the present invention have the following beneficial effects: The present invention uses a multispectral imaging device to collect multi-dimensional image sequences of the growth process of medicinal plants, and synchronously records environmental light intensity and temperature and humidity data, comprehensively obtaining morphological and physiological information of plant growth, and providing rich data for growth cycle assessment. Pixel offsets are eliminated through spatio-temporal registration operations to improve the accuracy and consistency of image data; an adaptive hybrid filtering algorithm is used to remove noise and retain leaf edge texture details, enhancing the image quality; multi-scale contrast enhancement is used to highlight key features and enhance feature recognizability. A dual-branch deep network is constructed to extract the dynamic feature set of leaf morphology and the change feature set of physiological parameters respectively, reflecting the plant growth state in multiple dimensions; the two feature sets are cross-modal fused and combined with environmental parameters to calculate the dynamic growth index matrix, comprehensively considering morphological, physiological characteristics and environmental impacts, making the growth index more representative and accurate; based on the matching degree analysis between the dynamic growth index matrix and the preset growth stage model, the current growth cycle stage identifier and the stage transition probability curve are output, realizing real-time monitoring and prediction of the growth stage of medicinal plants, and providing support for scientific management and precise regulation.

[0076] When extracting the dynamic feature set of leaf morphology, the present invention generates a leaf contour deformation field through dynamic deformation perception convolution operation, calculates the leaf curling index, and jointly encodes this index with the vein fractal dimension to finely capture the morphological changes of the leaf and enrich the morphological feature information. When extracting the feature set of physiological parameter changes, a spectral reflectance dynamic decoupling model is constructed to separate the chlorophyll absorption band and the water molecule overtone vibration signal, calculate the metabolic activity index, and perform time-domain cross-correlation analysis with the transpiration rate volatility to accurately reflect the changes in the plant physiological state. During the cross-modal fusion process, a growth trend coupling model is constructed to calculate the feature compatibility matrix, generate growth constraint factors, and perform multi-source feature constraint optimization to better fuse morphological and physiological features. At the same time, considering environmental constraints, the dynamic growth index matrix is made more scientific and practical. When calculating the environmental constraint function, a phase modulation equation of photosynthetically active radiation and a metabolic damping term are established to accurately quantify the impact of environmental factors on plant growth. In the calculation of the stage transition probability curve, a growth stage quantization model is constructed to calculate the coherence, generate the stage transition probability curve, and use the Hilbert space basis vector as the topological constraint of the state transition path to accurately predict the plant growth stage transition based on the principles of quantum mechanics. In the adaptive hybrid filtering algorithm, leaf-guided anisotropic diffusion filtering and morphological wavelet packet decomposition in the frequency domain are performed to effectively remove noise and retain the leaf edge texture details. In the dynamic deformation perception convolution operation, the convolution kernel deformation offset is generated according to the leaf growth direction field, and the deformed feature map is generated through differentiable sampling to flexibly capture the morphological changes of the leaf. In the growth stage quantization model, a growth Hilbert space is constructed, the environmental parameter perturbation Hamiltonian is introduced, and the eigenstates of the Hamiltonian are solved through the imaginary time evolution algorithm to determine the stage transition path. Based on the theoretical framework of quantum mechanics, the mechanism of plant growth stage transition is deeply studied, providing a solid theoretical basis for the accurate assessment of the growth cycle.

[0077] Further, the storage medium of the embodiment of the present application stores program instructions capable of implementing all the above methods. Among them, the program instructions can be stored in the above storage medium in the form of a software product, including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media that can store program codes such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs, or terminal devices such as computers, servers, mobile phones, and tablets.

[0078] The above description is only some preferred embodiments of the present invention and an explanation of the technical principles applied. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present invention is not limited to the technical solutions formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) having similar functions disclosed in the embodiments of the present invention.

Claims

1. A method for evaluating the growth cycle of medicinal plants based on AI image recognition, characterized in that, It includes the following steps: S1. Collect a multi-dimensional image sequence of the growth process of medicinal plants through a multi-spectral imaging device, and synchronously record environmental light intensity and temperature and humidity data; S2. Perform spatio-temporal registration operations on the multi-dimensional image sequence to eliminate pixel offsets caused by plant swaying and shooting angle changes; S3. Use an adaptive hybrid filtering algorithm to eliminate noise from the registered images and retain the leaf edge texture details; S4. Perform multi-scale contrast enhancement on the denoised images, and reconstruct the high-frequency feature maps through Laplacian pyramid decomposition; S5. Construct a two-branch deep network, where the first branch extracts the morphological dynamic feature set of the leaves, and the second branch extracts the physiological parameter change feature set; S6. Perform cross-modal fusion on the morphological dynamic feature set and the physiological parameter change feature set, and calculate the dynamic growth index matrix in combination with environmental parameters; S7. Based on the matching degree analysis between the dynamic growth index matrix and the preset growth stage model, output the current growth cycle stage identifier and the stage transition probability curve.

2. The method according to claim 1, characterized in that, The specific steps for extracting the morphological dynamic feature set of the leaves in step S5 include: S51. Perform dynamic deformation perception convolution operations on the visible light band image sequence. The convolution kernel adaptively adjusts the spatial sampling position according to the leaf growth direction to generate a leaf contour deformation field; S52. Calculate the three-dimensional curvature integral value based on the deformation field to establish the leaf curl; S53. Jointly encode the leaf curl index and the vein fractal dimension to generate the morphological dynamic feature set.

3. The method according to claim 1, characterized in that, The specific steps for extracting the physiological parameter change feature set in step S5 include: S54. Construct a spectral reflectance dynamic decoupling model to separate the chlorophyll absorption band and the overtone vibration signal of water molecules; S55. Calculate the metabolic activity index according to the spectral fingerprint library of medicinal plant species; S56. Perform time-domain cross-correlation analysis on the metabolic activity index and the transpiration rate volatility to generate the physiological parameter change feature set.

4. The method according to claim 1, characterized in that, The cross-modal fusion in step S6 adopts: S61. Construct a growth trend coupling model, map the morphological dynamic feature set to the physiological parameter space, and calculate the feature compatibility matrix; S62. Generate a growth constraint factor based on environmental parameters, which includes the modulation coefficient of light intensity on the photosynthetic phase and the damping coefficient of temperature and humidity on the metabolic rate; S63. Perform multi-source feature constraint optimization as shown in the following formula: Among them, is the fused dynamic growth index matrix; is the morphological dynamic feature set; is the physiological parameter change feature set; , are the adaptive balance parameters; is the environmental constraint function; is the growth index matrix to be optimized; is the set of environment-related factors.

5. The method according to claim 4, characterized in that, The described environmental constraint function is calculated to include: S611. Establish a phase modulation equation for photosynthetically active radiation as shown in the following formula: Among them, is the light phase modulation factor; L is the real-time light intensity; is the light saturation intensity of the species; is the day-night cycle parameter; t is the current time; S612. Construct a metabolic damping term as shown in the following formula: Among them, is the temperature-humidity damping factor; T is the ambient temperature; is the optimum growth temperature; is the temperature tolerance range; H is the ambient humidity; is the humidity critical threshold.

6. The method according to claim 1, characterized in that, The specific implementation of step S7 includes: S71. Construct a growth stage quantization model, map the dynamic growth index matrix to the Hilbert space, and generate discrete energy level basis vectors; S72. Calculate the coherence degree between the current dynamic growth index matrix and each energy level basis vector; S73. Generate a stage transition probability curve according to the coherence degree, and use the Hilbert space basis vector as the topological constraint of the state transition path.

7. The method according to claim 6, characterized in that, The calculation method of the stage transition probability curve includes: S74. Calculate the barrier height between adjacent energy levels based on the Hilbert space basis vectors and the thermal fluctuation energy ; S75. Establish a quantum tunneling rate equation as shown in the following formula: Among them, is the stage conversion rate; is the intrinsic transition rate; is the barrier height; is the thermal fluctuation energy; is the Boltzmann constant; T is the ambient temperature.

8. The method according to claim 1, wherein The adaptive hybrid filtering algorithm in step S3 includes: S31. Perform anisotropic diffusion filtering for blade guidance; S32. Implement morphological wavelet packet decomposition in the frequency domain and retain the sub-band components matching the vibration frequency of the leaf margin.

9. The method according to claim 2, wherein The dynamic deformation perception convolution operation includes: S511. Generate the convolution kernel deformation offset according to the blade growth direction field; S512. Apply the offset to the convolution kernel coordinates through differentiable sampling to generate the deformed feature map.

10. The method according to claim 6, wherein The growth stage quantization model includes: S711. Construct a growth Hilbert space whose basis vectors are composed of photosynthesis eigenstates; S712. Introduce the environmental parameter perturbation Hamiltonian to split the energy levels of the basis vectors of the Hilbert space; S713. Solve the eigenstates of the Hamiltonian through the imaginary time evolution algorithm to determine the stage transition path.

Images