Plant growth environment dynamic evaluation method and system based on multi-source perception

Abstract

The invention relates to the field of data processing, in particular to a plant growth environment dynamic evaluation method and system based on multi-source perception, and the method comprises the steps: obtaining an environment parameter matrix and a plant image, and obtaining an environment health index; training a preset growth stage recognition model based on the plant image and the growth stage; calculating the static importance of any column; obtaining a to-be-detected plant image and a to-be-detected environment parameter matrix, inputting the to-be-detected plant image into a preset growth stage recognition model to output a to-be-detected growth stage, taking the static importance of any column in the to-be-detected growth stage as a weight, and multiplying the weight with any column in the to-be-detected environment parameter matrix to obtain a weighting matrix; and inputting the weighting matrix into a preset plant prediction model to output the health index of the environment to be detected. According to the technical scheme of the invention, the precision of the evaluation result of the plant growth environment can be improved.

Description

Technical Field

[0001] This invention relates to the field of data processing. In particular, it relates to a method and system for dynamic assessment of plant growth environment based on multi-source sensing. Background Technology

[0002] Plant growth depends on multiple environmental factors, primarily including light, temperature, humidity, soil nutrients, and air quality. In modern agriculture, horticulture, and plant research, understanding and real-time monitoring of changes in the plant's growth environment is crucial for improving plant growth quality and efficiency. Such monitoring systems are often called environmental sensing systems, which collect environmental data to provide a basis for plant growth analysis. However, because plant growth is influenced by a combination of environmental factors, relying on a single environmental parameter often fails to accurately assess plant growth status, especially under complex conditions where environmental factors are intertwined; assessments from a single data source are frequently inaccurate.

[0003] Traditional methods for monitoring plant growth environments mostly rely on a single type of sensor, such as temperature and humidity sensors, soil moisture sensors, or light intensity sensors. These methods offer advantages such as low equipment cost and ease of installation, but they can only capture single environmental factors and lack a comprehensive assessment of the multi-dimensional environment. A single sensor is significantly affected by its local environment, leading to inaccurate or biased data, which in turn affects the plant growth monitoring results and results in inaccurate plant growth environment assessments. Summary of the Invention

[0004] To address the aforementioned technical problems, the present invention provides solutions in the following aspects.

[0005] In the first aspect, a dynamic assessment method for plant growth environment based on multi-source perception includes: collecting an environmental parameter matrix and plant images of plants at a target stage in history, where the target stage is any growth stage; the rows of the environmental parameter matrix represent environmental parameters of different dimensions at the same sampling time, and the columns of the environmental parameter matrix represent environmental parameters of the same dimension at different sampling times; obtaining the environmental health index corresponding to any row; training a preset growth stage recognition model based on plant images and growth stages; calculating the linearity and nonlinearity of any column under the target stage with the environmental health index, and using the absolute value of the product of the linearity and nonlinearity as the static importance of any column; acquiring the plant image to be detected and the environmental parameter matrix to be detected; inputting the plant image to be detected into the preset growth stage recognition model to output the growth stage to be detected; using the static importance of any column under the growth stage to be detected as a weight, multiplying it with any column in the environmental parameter matrix to obtain a weighted matrix; inputting the weighted matrix into a preset plant prediction model to output the environmental health index to be detected; and completing the environmental assessment based on the environmental health index to be detected.

[0006] Preferably, the preset growth stage identification model is a convolutional neural network, the input of which is a plant image, the output is a predicted value of the growth stage, the label is the true value of the growth stage, and the loss function of the convolutional neural network is cross-entropy loss.

[0007] Preferably, the linearity includes: constructing an environmental health index sequence for each column corresponding to the environmental health index; using the standard deviation of the environmental parameter sequence of any column as the first standard deviation and the standard deviation of the environmental health index sequence as the second standard deviation, and calculating the first product of the first standard deviation and the second standard deviation; calculating the covariance of the environmental parameter sequence and the environmental health index sequence of any column; and using the ratio of the covariance to the first product as the linearity of any column.

[0008] Preferably, the degree of nonlinearity includes: constructing an environmental health index sequence for each column corresponding to the environmental health index; performing regression analysis on the environmental parameter sequence and the environmental health index sequence of any column to obtain a residual value sequence; sorting the residual values; and calculating the Spearman rank correlation coefficient between the environmental parameter sequence and the environmental health index sequence of any column as the degree of nonlinearity of any column.

[0009] Preferably, the preset plant prediction model is a BP network, the input of which is an environmental parameter matrix, the output is the predicted value of the environmental health index, the label is the true value of the environmental health index, and the loss function of the BP network is the mean squared error loss.

[0010] Preferably, the step of completing the environmental assessment based on the environmental health index to be detected includes: in response to the environmental health index to be detected being not less than a preset threshold, the environmental assessment result is good and no adjustment is required; in response to the environmental health index to be detected being less than the preset threshold, the environmental assessment result is poor, the real-time importance of any column of the environmental parameter matrix to be detected is calculated, and the environmental parameters are adjusted according to the real-time importance.

[0011] Preferably, the real-time importance includes: using feature visualization technology to obtain the dynamic importance of any column in the environmental parameter matrix to be detected for a preset plant prediction model; preset static weights and dynamic weights, calculate the second product of static weights and static importance, calculate the third product of dynamic weights and dynamic importance, and use the sum of the second product and the third product as the real-time importance of any column.

[0012] Secondly, a dynamic assessment system for plant growth environment based on multi-source sensing includes: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the system implements any one of the methods for dynamic assessment of plant growth environment based on multi-source sensing.

[0013] The present invention has the following effects: This invention constructs a correlation between an environmental parameter matrix and plant images, utilizing a dynamic environmental health index assessment. This not only reflects the plant's adaptability to the environment at different growth stages in real time but also dynamically adjusts environmental parameters to optimize plant growth conditions. Specifically, by calculating the linear and nonlinear relationships between environmental parameters and the health index, the influence weight of each environmental factor is accurately quantified. In practical applications, environmental variables can be adjusted according to the real-time status of the plant, improving the accuracy of environmental assessment results. Attached Figure Description

[0014] Figure 1 This is a flowchart of a dynamic assessment method for plant growth environment based on multi-source sensing, according to an embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0016] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0017] Reference Figure 1 A method for dynamic assessment of plant growth environment based on multi-source sensing includes steps S1-S3, as detailed below: S1: Collect the environmental parameter matrix and plant images of plants in the target stage from the historical data. The target stage is any growth stage. The rows of the environmental parameter matrix represent environmental parameters of different dimensions at the same sampling time, and the columns of the environmental parameter matrix represent environmental parameters of the same dimension at different sampling times. Obtain the environmental health index corresponding to any row.

[0018] In one embodiment, multiple sensors (such as temperature and humidity sensors, light intensity sensors, carbon dioxide concentration sensors, and pH sensors) are deployed to collect environmental parameters from the plant growth environment, including temperature, humidity, light intensity, carbon dioxide concentration, and nutrient solution pH, while simultaneously acquiring plant images. Subsequently, the raw data undergoes preprocessing operations such as cleaning, missing value imputation, and normalization to eliminate dimensional differences and improve the consistency of model input. Based on this, those skilled in the art, combining theoretical models of plant photosynthetic efficiency with actual growth performance, comprehensively evaluate each collection of multidimensional environmental data and assign an environmental health index between 0 and 1.

[0019] It's important to explain that historical monitoring data is constructed into an environmental parameter matrix: each row corresponds to a specific sampling time, containing the observed values ​​for all environmental dimensions at that time (such as temperature, humidity, light intensity, carbon dioxide concentration, and nutrient solution pH); each column represents the time series of a specific environmental parameter at different sampling times. This matrix representation clearly demonstrates the dynamic trends of various environmental factors over time.

[0020] S2: Train a pre-defined growth stage recognition model based on plant images and growth stages; calculate the linearity and nonlinearity of any column under the target stage with the environmental health index, and use the absolute value of the product of the linearity and nonlinearity as the static importance of any column.

[0021] In one embodiment, a preset growth stage recognition model is trained using historically collected plant images and their corresponding growth stages. This model is a convolutional neural network (CNN). The input to the CNN is the historical plant images, and the output is the predicted growth stage corresponding to each historical plant image. The label is the true value of the growth stage. The loss function of the CNN is cross-entropy loss. Gradient descent is used to update the CNN's network parameters. Training stops when the maximum number of training iterations is reached or the loss value is less than a set threshold. For example, training stops when the number of iterations reaches 200 or the loss value is less than 0.0001.

[0022] It should be noted that plants exhibit significant differences in their sensitivity to and requirements for environmental factors at different growth stages. For example, during the seedling stage, temperature and humidity play a decisive role in root development and seedling survival rate; while during the flowering and fruiting stage, sufficient light and suitable temperature more directly affect photosynthetic efficiency and fruit quality. Therefore, the contribution weight of environmental parameters to the overall environmental health index is not constant. To more accurately assess environmental suitability at each stage, it is necessary to calculate the importance of each dimension at different growth stages.

[0023] Specifically, when calculating the importance of environmental parameters to the environmental health index, both linear and nonlinear relationships between them and the health index should be considered. This is because, in actual plant growth, some environmental factors (such as temperature or light) may show an approximately linear positive correlation with health within a moderate range, but exceeding a threshold may drastically reduce health, exhibiting a significant nonlinear or even non-monotonic relationship (such as an inverted U-shaped response). Therefore, relying solely on linear models (such as multiple linear regression) will ignore such complex interaction effects, leading to biased importance assessments. To address this, a hybrid method combining linear and nonlinear modeling capabilities can be used to construct a pre-defined plant prediction model, and the contribution of each environmental parameter can be comprehensively evaluated by incorporating feature importance metrics.

[0024] Quantifying the linearity between a specific environmental parameter and an environmental health index involves aligning each column of the environmental parameter matrix (representing the environmental parameter value at different sampling times) with the corresponding environmental health index sequence, forming two time series of equal length: one for the environmental parameter and the other for the environmental health index during the same period. Then, the standard deviations of the environmental parameter sequence (denoted as the first standard deviation) and the environmental health index sequence (denoted as the second standard deviation) are calculated separately and multiplied to obtain the first product. Next, the covariance between the environmental parameter sequence and the health index sequence is calculated to measure the trend of their joint change. Finally, the covariance is divided by the first product; the resulting ratio represents the linearity between the environmental parameter and the health index. Linearity effectively characterizes the linear dependence between various environmental factors and the health index.

[0025] Quantifying the nonlinearity between an environmental parameter and an environmental health index involves aligning each column of the environmental parameter matrix (i.e., the observed values ​​of a certain environmental factor at different sampling times) with the corresponding environmental health index at those times, thus constructing an environmental parameter sequence and an environmental health index sequence. Next, linear regression is performed on these two sequences to obtain predicted values ​​and calculate residual sequences to capture the portion not explained by the linear model, which often contains nonlinear or monotonic nonlinear relationships. Subsequently, instead of directly using the original values, the two original sequences are sorted (i.e., converted to ranks), and their Spearman rank correlation coefficients are calculated. This coefficient measures the strength of the monotonic relationship between two variables, regardless of whether the relationship is linear or nonlinear (such as logarithmic, exponential, or piecewise monotonic relationships), as long as there is a consistent upward or downward trend between the variables, it can be effectively captured.

[0026] The absolute value of the product of linearity and nonlinearity is used as the static importance of any column.

[0027] The static importance of each environmental parameter at any growth stage is obtained by iterating through the parameters.

[0028] S3: Obtain the image of the plant to be detected and the matrix of environmental parameters to be detected. Input the image of the plant to be detected into the preset growth stage recognition model to output the growth stage to be detected. Use the static importance of any column under the growth stage to be detected as a weight and multiply it with any column in the matrix of environmental parameters to obtain a weighted matrix. Input the weighted matrix into the preset plant prediction model to output the health index of the environment to be detected. Complete the environmental assessment based on the health index of the environment to be detected.

[0029] In one embodiment, a preset plant prediction model is trained based on a historical environmental parameter matrix and an environmental health index. The preset plant prediction model is a Back Propagation (BP) network. The input to the BP network is the environmental parameter matrix, and the output is the predicted value of the environmental health index. The label is the true value of the environmental health index, and the loss function of the BP network is the mean squared error loss. The network parameters of the BP network are updated using gradient descent. Training stops when the maximum number of training iterations is reached or the loss value is less than a set threshold. For example, training stops when the number of training iterations reaches 200 or the loss value is less than 0.0001.

[0030] To obtain the image of the plant to be detected and the matrix of environmental parameters to be detected, it should be noted that the historical environmental parameter matrix is... take The matrix, i.e. OK Columns and rows represent environmental parameters in different dimensions at the same sampling time, while columns represent environmental parameters in the same dimension at different sampling times. The matrix of environmental parameters to be detected is 1-dimensional. A matrix, i.e., 1 row Columns represent environmental parameters at different sampling times within the same dimension. The number of columns indicates the number of dimensions of environmental parameters. For example... The value is 5, which represents five dimensions: temperature, humidity, light intensity, carbon dioxide concentration, and pH of the nutrient solution.

[0031] The image of the plant to be detected is input into a pre-trained preset growth stage recognition model, which outputs the growth stage to be detected. The static importance of any column under the growth stage to be detected is used as a weight and multiplied by any column in the environmental parameter matrix to be detected. That is, the static importance of the historical temperature dimension is used as a weight and multiplied by the temperature value in the first column of the environmental parameter matrix to be detected. The environmental parameters in each column of the environmental parameter matrix to be detected are weighted in the same way to obtain a weighted matrix. The weighted matrix is ​​input into a preset plant prediction model to output the environmental health index to be detected. If the environmental health index to be detected is not less than a preset threshold, the environmental assessment result is good and no adjustment is needed; if the environmental health index to be detected is less than the preset threshold, the environmental assessment result is poor. The real-time importance of any column in the environmental parameter matrix to be detected is calculated, and the environmental parameters are adjusted according to the real-time importance.

[0032] The real-time importance includes: using feature visualization technology to obtain the dynamic importance of any column in the environmental parameter matrix to be detected for the preset plant prediction model; setting static and dynamic weights, calculating the second product of the static weight and the static importance, calculating the third product of the dynamic weight and the dynamic importance, and using the sum of the second and third products as the real-time importance of any column. For example, the static weight is set to 0.3 and the dynamic weight is set to 0.7.

[0033] The system includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement a dynamic assessment method for plant growth environment based on multi-source sensing according to the first aspect of the present invention.

[0034] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0035] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for dynamic assessment of plant growth environment based on multi-source sensing, comprising: Collect environmental parameter matrices and plant images of plants in the target stage from historical data. The target stage is any growth stage. The rows of the environmental parameter matrix represent environmental parameters of different dimensions at the same sampling time, and the columns of the environmental parameter matrix represent environmental parameters of the same dimension at different sampling times. Obtain the environmental health index corresponding to any row. A pre-defined growth stage recognition model is trained based on plant images and growth stages; Calculate the linearity and nonlinearity of any column with respect to the environmental health index under the target stage, and take the absolute value of the product of the linearity and nonlinearity as the static importance of any column; The process involves acquiring images of the plant to be detected and a matrix of environmental parameters to be detected. The plant image is input into a preset growth stage recognition model, which outputs the growth stage to be detected. The static importance of any column under the growth stage to be detected is used as a weight and multiplied by any column in the matrix of environmental parameters to be detected to obtain a weighted matrix. The weighted matrix is ​​input into a preset plant prediction model, which outputs the health index of the environment to be detected. An environmental assessment is then completed based on the health index of the environment to be detected.

2. The method for dynamic assessment of plant growth environment based on multi-source sensing according to claim 1, characterized in that, The preset growth stage recognition model is a convolutional neural network. The input of the convolutional neural network is a plant image, the output is the predicted value of the growth stage, the label is the true value of the growth stage, and the loss function of the convolutional neural network is cross-entropy loss.

3. The method for dynamic assessment of plant growth environment based on multi-source sensing according to claim 1, characterized in that, The degree of linearity includes: Construct an environmental health index sequence for each column corresponding to the environmental health index. Take the standard deviation of any column of environmental parameter series as the first standard deviation, and the standard deviation of the environmental health index series as the second standard deviation, and calculate the first product of the first standard deviation and the second standard deviation. Calculate the covariance between any column of environmental parameter sequences and environmental health index sequences; The ratio of the covariance to the first product is used as the linearity of any column.

4. The method for dynamic assessment of plant growth environment based on multi-source sensing according to claim 1, characterized in that, The degree of nonlinearity includes: Construct an environmental health index sequence for each column corresponding to the environmental health index. Regression analysis is performed on the environmental parameter series and environmental health index series of any column to obtain the residual value series. The residual values ​​are sorted, and the Spearman rank correlation coefficient between the environmental parameter series and the environmental health index series of any column is calculated as the degree of nonlinearity of any column.

5. The method for dynamic assessment of plant growth environment based on multi-source sensing according to claim 1, characterized in that, The preset plant prediction model is a BP network. The input of the BP network is an environmental parameter matrix, the output is the predicted value of the environmental health index, the label is the true value of the environmental health index, and the loss function of the BP network is the mean squared error loss.

6. The method for dynamic assessment of plant growth environment based on multi-source sensing according to claim 1, characterized in that, The environmental assessment based on the environmental health index to be tested includes: If the environmental health index to be tested is not less than the preset threshold, the environmental assessment result is good and no adjustment is required; In response to a poor environmental assessment result due to the health index of the environment being tested being less than a preset threshold, the real-time importance of any column in the environmental parameter matrix is ​​calculated, and the environmental parameters are adjusted according to the real-time importance.

7. The method for dynamic assessment of plant growth environment based on multi-source sensing according to claim 6, characterized in that, The real-time importance includes: The dynamic importance of any column in the matrix of environmental parameters to be detected is obtained by using feature visualization technology on the preset plant prediction model. Preset static and dynamic weights, calculate the second product of static weight and static importance, calculate the third product of dynamic weight and dynamic importance, and use the sum of the second and third products as the real-time importance of any column.

8. A dynamic assessment system for plant growth environment based on multi-source sensing, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a dynamic assessment method for plant growth environment based on multi-source sensing according to any one of claims 1-7.

Images