Flexible wearable spectrum sensor for leaf physiological feature monitoring and application method thereof

Abstract

The invention discloses a flexible wearable spectrum sensor for leaf physiological feature monitoring and an application method thereof, and relates to the field of long-term in-situ monitoring of crop growth physiological features. Comprising the following steps: the wearable spectrum sensor adopts two layers of flexible polyimide substrate materials, and the two layers of flexible polyimide substrate materials are respectively combined with a copper wire through acrylic adhesive to form a thin single-layer flexible circuit board; an AS7343 spectrum sensor, an infrared light module, an ultraviolet light module and a visible light module are arranged on the single-layer flexible circuit board, and the collected spectrum range is 405-855 nm; the thickness of the wearable spectrum sensor is 0.11 mm, the weight of the wearable spectrum sensor is 0.1894 g, and the size of the wearable spectrum sensor is 1.2 cm * 1.1 cm. The wearable spectrum sensor disclosed by the invention can effectively monitor key physiological information in a crop growth process, especially long-term in-situ monitoring of chlorophyll content and leaf water content.

Description

Technical Field

[0001] The present invention relates to the technical field of long-term in-situ monitoring of physiological characteristics of crop growth, and in particular to a flexible wearable spectral sensor for monitoring leaf physiological characteristics and an application method thereof. Background Art

[0002] With the rapid growth of the global population and the severe challenges brought by climate change, agriculture is facing unprecedented pressure. In this context, improving crop production efficiency, ensuring food safety and achieving sustainable development have become urgent tasks worldwide. According to the forecast of the Food and Agriculture Organization of the United Nations (FAO), by 2050, global food demand will increase by more than 70% compared with 2010. This trend forces agricultural production to adapt to new requirements and achieve precision agricultural management through scientific and technological innovation.

[0003] In agricultural production, the growth status of crops is closely related to their physiological characteristics, especially the physiological information of leaves, such as leaf SPAD value (relative value of chlorophyll content) and water content, which are crucial to improving crop yield and quality. Effective monitoring of these physiological characteristics and proper management can not only promote crop photosynthesis and transpiration, but also optimize overall growth and development. However, traditional measurement methods, such as chemical analysis and drying methods, have many shortcomings. These methods are complex and time-consuming, and are prone to damage to plants, which may interfere with the normal growth of crops and directly affect the efficiency and effectiveness of agricultural management.

[0004] In response to the limitations of traditional measurement methods, researchers are constantly exploring more efficient, portable, and non-destructive monitoring technologies. In recent years, spectral analysis technology has gradually emerged in the field of plant physiological characteristics detection due to its high-speed, non-destructive, and real-time monitoring characteristics. Through non-contact measurements in the near-infrared, visible, and ultraviolet spectral regions, researchers can quickly obtain the chlorophyll content, moisture content, and other physiological indicators of crops, thereby achieving timely feedback and scientific decision-making on the health status of crops. However, existing traditional spectral instruments are usually bulky and complex to operate, making them difficult to use in the field. At the same time, they are often limited by the operating requirements of the equipment and are often unable to meet the needs of long-term crop monitoring.

[0005] Faced with this situation, wearable sensors, as a new generation of monitoring tools, have shown great development potential. Their lightweight and flexible features allow the sensors to fit on the surface of plants, reducing the impact on plant growth while providing stable monitoring results under different environmental conditions.

[0006] Therefore, it is an urgent problem for those skilled in the art to propose a flexible wearable spectral sensor for monitoring leaf physiological characteristics and an application method thereof to solve the difficulties existing in the prior art. Summary of the invention

[0007] In view of this, the present invention provides a flexible wearable spectral sensor for monitoring leaf physiological characteristics and an application method thereof, so as to realize long-term in-situ monitoring of key physiological information during crop growth, especially chlorophyll content and leaf moisture content.

[0008] In order to achieve the above object, the present invention adopts the following technical solution:

[0009] A flexible wearable spectral sensor for monitoring leaf physiological characteristics, comprising:

[0010] The wearable spectral sensor uses two layers of flexible polyimide base material, which are respectively combined with copper wires through acrylic glue to form a thin single-layer flexible circuit board; the single-layer flexible circuit board is equipped with AS7343 spectral sensor, infrared light module, ultraviolet light module and visible light module, and the collected spectrum range is 405-855nm; the wearable spectral sensor is 0.11mm thick, weighs 0.1894g, and measures 1.2cm×1.1cm.

[0011] A method for applying a flexible wearable spectral sensor for monitoring leaf physiological characteristics comprises the following steps:

[0012] S1. Collect chlorophyll concentration and water content measurement data of leaves at different parts of different plants based on random sampling method;

[0013] S2. Acquire spectral data of the leaves at the same position as in S1 based on a wearable spectral sensor;

[0014] S3. Based on the leaf spectral data obtained in S2, a data prediction model is constructed to obtain the prediction results of the chlorophyll concentration and the moisture content of the leaves;

[0015] S4. Compare and analyze the chlorophyll concentration prediction results and moisture content prediction results of the leaves in S3 with the chlorophyll concentration measurement data and moisture content measurement data of the leaves obtained in S1 to evaluate the feasibility of wearable spectral sensor measurement.

[0016] Optionally, in S1, the chlorophyll content of the leaves is measured by a SPAD measuring instrument to obtain chlorophyll concentration measurement data;

[0017] The leaves were placed on a balance to measure their fresh weight. After recording the fresh weight, they were immediately placed in a drying oven for drying. That is, the leaves were placed in a 105°C oven for 30 minutes of withering treatment, and then the leaves were dried at 80°C. After multiple measurements, when the weight no longer changed, the moisture content was calculated to obtain the moisture content measurement data.

[0018] Optionally, the moisture content is calculated as follows:

[0019]

[0020] Among them, FW is fresh weight, DW is dry weight, and RWC is moisture content.

[0021] Optionally, in S2, the wearable spectral sensor is installed on the back of the leaf by magnetic connection, and the average value of each band is taken as the final spectral data, so as to obtain the spectral data of the leaf in the same part as in S1.

[0022] Optionally, in S3, based on the leaf spectral data obtained in S2, a data prediction model is constructed to obtain the leaf chlorophyll concentration prediction result and the water content prediction result. The specific contents are:

[0023] S31. performing noise reduction preprocessing on the acquired spectral data;

[0024] S32. dividing the preprocessed spectral data into a training set and a test set;

[0025] S33. Input the training set into the data prediction model to train the data prediction model. After several trainings, a trained data prediction model is obtained.

[0026] S34. Input the test set into the trained data prediction model to obtain the chlorophyll concentration prediction results and water content prediction results of the leaves.

[0027] Optionally, the noise reduction preprocessing performed on the acquired spectral data in S31 includes multivariate scattering correction processing, standard normal transformation processing, wavelet denoising processing, smoothing processing, normalization processing, continuous projection method, first-order derivative processing, and detrending processing.

[0028] Optionally, in S32, the first 70% of the preprocessed spectral data is divided into a training set, and the last 30% is divided into a test set.

[0029] Optionally, use the training set determination coefficient Test set determination coefficient The training set root mean square error RMSEC and the test set root mean square error RMSEP are used as the basis for evaluating the performance of the data prediction model. The training set determination coefficient Test set determination coefficient The higher the value, the better the effect of the data prediction model; the lower the root mean square error RMSEC of the training set and the root mean square error RMSEP of the test set, the stronger the stability of the data prediction model;

[0030]

[0031] Among them, YPi is the predicted value of chlorophyll or leaf water content of the i-th leaf, Y i is the measured value of chlorophyll or leaf water content of the i-th leaf, is the average value of leaf chlorophyll or leaf water content in the corresponding model set, n c is the number of calculations in the training set, n P is the number of computations in the test set.

[0032] It can be seen from the above technical solutions that, compared with the prior art, the present invention provides a flexible wearable spectral sensor for monitoring leaf physiological characteristics and an application method thereof, which has the following beneficial effects:

[0033] (1) The wearable spectral sensor of the present invention shows significant advantages in monitoring the physiological characteristics of plant leaves, especially in the ability to achieve non-destructive and real-time in-situ monitoring. The wearable spectral sensor can effectively measure the SPAD value and moisture content of leaves, and has better long-term monitoring effect than traditional monitoring methods. The research results show that the wearable spectral sensor is very suitable for real-time acquisition of plant physiological information, providing a new technical support for agricultural management.

[0034] (2) The wearable spectral sensor is designed to be lightweight, easy to operate, and flexible to arrange, so that it can be non-invasively attached to the back of plant leaves. This feature significantly reduces the impact on plant growth, especially in important physiological processes such as photosynthesis and transpiration. The sensor is fixed by magnetic attraction to ensure the stability and reliability of long-term data acquisition. The experimental results show that different regression models have obvious performance differences in predicting SPAD values ​​and moisture content. The prediction effects of Gaussian process regression and gradient boosting regression are particularly outstanding. In particular, their prediction accuracy has been significantly improved after data preprocessing. These results demonstrate the effectiveness of sensor integrated design and lay a solid foundation for promoting on-site agricultural monitoring.

[0035] (3) At the data processing level, the present invention adopts a variety of noise reduction preprocessing methods for the collected spectral data to improve the data quality. By introducing noise reduction strategies such as multivariate scattering correction and standard normal transformation, the predictive ability of the model is significantly improved, which not only enhances the reliability of the data, but also provides the necessary adaptability for data collection in complex environments.

[0036] (4) When selecting a machine learning model, the present invention conducts an extensive analysis of a variety of regression algorithms, including linear regression, random forest, and gradient boosting regression. Through the method of multi-model fusion, the prediction performance of leaf SPAD value and moisture content is further improved. In particular, when the VotingRegressor and StackingRegressor fusion model is adopted, the prediction effect of leaf physiological indicators is effectively improved. The comprehensive use of this algorithm not only enhances the prediction ability, but also reduces the dependence on a single data set, so that the algorithm shows good performance in a wider range of application scenarios.

[0037] (5) The in-situ tracking technology of wearable spectral sensors provides a new path for the precise management of modern agriculture. By real-time monitoring of the physiological characteristics of plant leaves, agricultural producers can obtain the growth status of crops in a timely manner and formulate scientific and effective management strategies. In the face of challenges such as climate change and resource constraints, the promotion of this technology will be of great significance to the sustainable development of agriculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0039] Figure 1 A schematic diagram of the core structure of a flexible wearable spectral sensor for monitoring leaf physiological characteristics provided by the present invention;

[0040] Figure 2 A flow chart of an application method of a flexible wearable spectral sensor for monitoring leaf physiological characteristics provided by the present invention;

[0041] Figure 3 A schematic diagram of collecting spectral data of a combined light source by a spectrometer (monitoring wavelength range 200-1200nm) provided by the present invention;

[0042] Figure 4 A schematic diagram of measuring the chlorophyll content of leaves by the chlorophyll content measuring instrument SPAD-502 provided by the present invention;

[0043] Figure 5 A schematic diagram of the final SPAD value of the blade collection position provided by the present invention;

[0044] Figure 6 A schematic diagram of a wearable spectral sensor provided by the present invention being placed on the back of a leaf;

[0045] Figure 7 A schematic diagram of obtaining spectral data of leaves at the same position based on a wearable spectral sensor provided by the present invention;

[0046] Figure 8 The present invention provides a method for drying leaves;

[0047] Fig. 9 A schematic diagram of the moisture content distribution of all leaf samples provided by the present invention;

[0048] Among them, 1-sample leaf, 2-wearable spectral sensor, 3-small cylindrical strong magnet, 4-AS7343 spectral sensor, 5-infrared light module, 6-ultraviolet light module, 7-visible light module. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] Reference Figure 1 As shown, the present invention discloses a flexible wearable spectral sensor for monitoring leaf physiological characteristics, comprising:

[0051] The wearable spectral sensor 2 uses two layers of flexible polyimide base material, and the two layers of flexible polyimide base material are respectively combined with copper wires through acrylic glue to form a thin single-layer flexible circuit board; the single-layer flexible circuit board is provided with an AS7343 spectral sensor 4, an infrared light module 5, an ultraviolet light module 6 and a visible light module 7, and the collected spectrum range is 405-855nm; the wearable spectral sensor 2 has a thickness of 0.11mm, a weight of 0.1894g, and a size of 1.2cm×1.1cm.

[0052] Reference Figure 2 As shown, the present invention discloses an application method of a flexible wearable spectral sensor for monitoring leaf physiological characteristics, comprising the following steps:

[0053] S1. Collect chlorophyll concentration and water content measurement data of leaves at different parts of different plants based on random sampling method;

[0054] S2. Acquire spectral data of the leaves at the same position as in S1 based on the wearable spectral sensor 2;

[0055] S3. Based on the leaf spectral data obtained in S2, a data prediction model is constructed to obtain the prediction results of the chlorophyll concentration and the moisture content of the leaves;

[0056] S4. Compare and analyze the chlorophyll concentration prediction results and moisture content prediction results of the leaves in S3 with the chlorophyll concentration measurement data and moisture content measurement data of the leaves obtained in S1 to evaluate the feasibility of the wearable spectral sensor 2.

[0057] Furthermore, in S1, the chlorophyll content of the leaves is measured by a SPAD measuring instrument to obtain chlorophyll concentration measurement data;

[0058] The leaves were placed on a balance to measure their fresh weight. After recording the fresh weight, they were immediately placed in a drying oven for drying. That is, the leaves were placed in a 105°C oven for 30 minutes of withering treatment, and then the leaves were dried at 80°C. After multiple measurements, when the weight no longer changed, the moisture content was calculated to obtain the moisture content measurement data.

[0059] Furthermore, the calculation formula of moisture content is as follows:

[0060]

[0061] Among them, FW is fresh weight, DW is dry weight, and RWC is moisture content.

[0062] Furthermore, in S2, the wearable spectral sensor is installed on the back of the leaf by magnetic connection, and the average value of each band is taken as the final spectral data, so as to obtain the spectral data of the leaf in the same part as in S1.

[0063] Furthermore, in S3, based on the leaf spectral data obtained in S2, a data prediction model is constructed to obtain the leaf chlorophyll concentration prediction results and the water content prediction results. The specific contents are as follows:

[0064] S31. performing noise reduction preprocessing on the acquired spectral data;

[0065] S32. dividing the preprocessed spectral data into a training set and a test set;

[0066] S33. Input the training set into the data prediction model to train the data prediction model. After several trainings, a trained data prediction model is obtained.

[0067] S34. Input the test set into the trained data prediction model to obtain the chlorophyll concentration prediction results and water content prediction results of the leaves.

[0068] Furthermore, in S31, the acquired spectral data is subjected to denoising preprocessing including multivariate scattering correction processing, standard normal transformation processing, wavelet denoising processing, smoothing processing, normalization processing, continuous projection method, first-order derivative processing, and detrending processing.

[0069] Furthermore, in S32, the first 70% of the preprocessed spectral data is divided into a training set, and the last 30% is divided into a test set.

[0070] Furthermore, the coefficients are determined using the training set Test set determination coefficient The training set root mean square error RMSEC and the test set root mean square error RMSEP are used as the basis for evaluating the performance of the data prediction model. The training set determination coefficient Test set determination coefficient The higher the value, the better the effect of the data prediction model; the lower the root mean square error RMSEC of the training set and the root mean square error RMSEP of the test set, the stronger the stability of the data prediction model;

[0071]

[0072] Among them, Y Pi is the predicted value of chlorophyll or leaf water content of the i-th leaf, Y i is the measured value of chlorophyll or leaf water content of the i-th leaf, is the average value of leaf chlorophyll or leaf water content in the corresponding model set, n c is the number of calculations in the training set, n P is the number of computations in the test set.

[0073] In a specific embodiment, the following are included

[0074] 1. Design of wearable spectral sensor

[0075] In order to achieve long-term in-situ monitoring of the physiological characteristics of crop growth, an innovative wearable spectral sensor 2 is designed. The wearable spectral sensor 2 aims to minimize the impact on crop growth while ensuring its stable long-term monitoring capability. Figure 1The core structure of the wearable spectral sensor 2 is shown. The wearable spectral sensor 2 uses two layers of flexible polyimide substrate material (PI). The two layers of flexible polyimide substrate material (PI) are combined with copper wires (CU) by acrylic glue (AD) to form a thin single-layer flexible circuit board (FPC); the single-layer flexible circuit board (FPC) is provided with an AS7343 spectral sensor 4, an infrared light module 5, an ultraviolet light module 6 and a visible light module 7, and the collected spectrum range is 405-855nm; the wearable spectral sensor 2 has a thickness of 0.11mm, a weight of 0.1894g, and a size of 1.2cm×1.1cm. This design can adapt to plants with leaves of different sizes and meet the needs of crop monitoring in various growth periods. The infrared light module 5 is an infrared LED lamp, the ultraviolet light module 6 is an ultraviolet LED lamp, and the visible light module 7 is a visible light LED lamp. In order to cover the monitoring band of the AS7343 spectral sensor 4, the wearable spectral sensor 2 uses a light source combination of infrared, ultraviolet and visible light LED lamps. Figure 3 As shown, the spectral data of the combined light source is collected by a spectrometer (monitoring band 200-1200nm), and the data shows that the combined light source can provide a full spectrum from the near-ultraviolet range to visible light and even to the near-infrared. The design of this combined light source meets the monitoring requirements of the AS7343 spectral sensor 4 at a spectral width of 405-855nm, thereby ensuring the effective capture of diverse spectral information during plant growth. The overall weight of the wearable spectral sensor 2 measured using a high-precision 10,000-point balance is only 0.1894g. This lightweight design significantly improves its wearability, reduces interference with plant growth, and ensures the reliability and accuracy of long-term monitoring. Through its non-invasive characteristics, the wearable spectral sensor 2 has important advantages in field applications, providing a reliable solution for real-time and dynamic monitoring of physiological characteristics, enabling researchers to more comprehensively and accurately grasp the growth status of crops.

[0076] (II) Sample preparation

[0077] The tomato experimental samples selected by the present invention were sown and seedled in Weifang City, Shandong Province, China on June 7, 2024. After sufficient growth and management, the samples were transplanted to Xinjiang Agricultural University's Xinjiang Intelligent Agricultural Equipment Key Laboratory KY-03 on July 18 for growth to ensure its plant growth under controlled conditions. The sample selection used Provence tomatoes as the experimental variety, which has good growth characteristics and experimental adaptability. After the growth period, the experimental samples were sampled from August 16 to August 17. To ensure the representativeness of the samples, multiple parts of different plants were randomly selected for leaf collection. During the sampling process, the chlorophyll content of leaf sample 1 was first measured by a SPAD measuring instrument, and then the spectral data of leaf sample 1 was collected to obtain the spectral characteristics of leaf sample 1. After the sampling was completed, the fresh weight of leaf sample 1 was measured immediately, and then dried until constant weight was reached, and then the dry weight was measured to calculate the moisture content and composition of each sample. The entire experimental process was completed in the laboratory, and there was no sample transfer time during the sample preparation process, ensuring the stability of the operating conditions and the reliability of the data.

[0078] (III) Collecting chlorophyll concentration measurement data of leaf samples

[0079] 80 pots of tomato plants of different growth conditions were selected for SPAD measurement of chlorophyll content, and a total of 200 leaf samples were collected. During the measurement process, a leaf sample 1 without obvious pests and diseases was randomly selected. During the measurement, the thick veins in leaf sample 1 were deliberately avoided to avoid measurement errors. Figure 4 The right side is an enlarged view of the left side, which was measured by Konica Minolta's SPAD-502 chlorophyll content meter. Figure 4 The position of the right collection point is measured five times cumulatively, and the average value is taken as the final SPAD value of the collection position of the leaf sample 1 (such as Figure 5 shown).

[0080] (IV) Obtaining spectral data of leaves in the same position based on wearable spectral sensors

[0081] After the chlorophyll measurement, the leaf sample 1 Figure 6 The leaf sample 1 was measured by SPAD and the same spectral data was collected. The AS7343 spectral sensor 4 was connected to the small cylindrical strong magnet 3 through PDMS transparent high elastic glue and baked in an oven at 80°C for 4 hours to ensure a stable connection. The wearable spectral sensor 2 was placed on the back of the leaf sample 1 and fixed with a magnet to complete the installation of the device. The spectral data was collected 10 times, and the average value of each band was taken as the final spectral data (such as Figure 7 shown).

[0082] (V) Collecting leaf moisture content measurement data

[0083] After completing the spectral data collection, the leaf sample 1 was placed on a balance (accuracy 0.0001g) for fresh weight measurement. The weight at this time was the fresh weight FW. After recording the fresh weight, it was immediately placed in Figure 8 The leaf sample 1 was dried in a drying oven, and the leaf sample 1 was placed in a 105°C oven for 30 minutes. Then the leaf sample 1 was dried at 80°C. When the weight no longer changed after three measurements, the weight was recorded as the dry weight DW. The moisture content RWC calculation formula of leaf sample 1 is as follows. The moisture content distribution of all leaf samples is as follows Fig. 9 As shown;

[0084]

[0085] Among them, FW is fresh weight, DW is dry weight, and RWC is moisture content.

[0086] (vi) constructing a data prediction model based on the acquired leaf spectral data to obtain a chlorophyll concentration prediction result and a water content prediction result of leaf sample 1;

[0087] (1) Perform noise reduction preprocessing on the acquired spectral data;

[0088] In the process of spectral data collection, various noises may appear due to improper equipment installation or environmental lighting. In order to improve the accuracy of subsequent modeling and analysis, multiplicative scatter correction (MSC) is used to eliminate the scattering effect and make the spectral data of the sample closer to the real sample characteristics. Standard Normal Variate (SNV) can eliminate the influence of spectral scattering, thereby improving the comparability between different samples. Wavelet Denoising (WD) uses wavelet transform to decompose spectral data and effectively remove noise components. Smoothing (SMT) reduces the random fluctuations of spectral data through smoothing to improve the stability of the signal. Normalization (NORM) normalizes the data to facilitate comparison between different data sets. Continuum Removal (CR) is used to eliminate baseline drift in spectral data, thereby improving the accuracy of subsequent analysis. First-order derivative (Derivative Spectroscopy, DS), by calculating the first-order derivative of the spectrum, enhances the recognizability of the characteristic signal. Detrending (Detrending, DT), is used to eliminate the trend component in the spectral data to ensure the purity of the data signal. A variety of effective noise reduction techniques are used to pre-process the spectral data for noise reduction.

[0089] (2) Dividing the preprocessed spectral data into a training set and a test set;

[0090] (3) Inputting the training set into the data prediction model to train the data prediction model. After several trainings, a trained data prediction model is obtained.

[0091] Fifteen regression algorithms were used for comprehensive evaluation. These algorithms include Linear Regression (LR), which is suitable for linear relationships, can quickly obtain prediction results and is easy to explain; Ridge Regression (RR), which effectively handles multicollinearity and reduces model complexity, especially suitable for high-dimensional data; Huber Regression (HR), which is robust to outliers and suitable for data sets containing a small number of outliers; K-Nearest Neighbors Regression (KNN), whose simple and easy-to-understand characteristics make it unnecessary to assume data distribution and can capture complex nonlinear relationships; Random Forest Regression (RFR), which adapts to the analysis of complex nonlinear relationships by processing high-dimensional data and reducing overfitting; AdaBoost Regression (ABR), which improves prediction accuracy by weighted combination of multiple weak learners and shows high robustness; Gradient Boosting Regression (GBR), which effectively captures complex nonlinear relationships through good prediction performance and is suitable for processing large-scale data; Bagging Regression (BGR), which effectively captures complex nonlinear relationships through good prediction performance and is suitable for processing large-scale data; Regression (BR), which enhances robustness by reducing model variance, is suitable for irregular data distribution; Gaussian Process Regression (GPR), provides uncertainty estimation, is particularly suitable for small sample data and can adapt to different function shapes; Bayesian Ridge Regression (BRR), provides parameter uncertainty estimation for multicollinearity and noisy data; Partial Least Squares Regression (PLSR), effectively extracts information when facing highly correlated independent variables, is particularly suitable for high-dimensional data; Support Vector Regression (SVR), can handle complex nonlinear relationships by finding the best hyperplane; Transformed Target Regression (TTR), improves the linear characteristics of the model by transforming the target variable, thereby improving the prediction performance; Lasso Regression (LAR), selects features and reduces model complexity through L1 regularization, is suitable for generating sparse solutions; ElasticNet Regression (ElasticNet ENR) combines L1 and L2 regularization and can make effective predictions in the case of multicollinearity.In the scenarios of spectrum, leaf SPAD and moisture content prediction, these algorithms each have their own advantages, providing diverse options for achieving accurate predictions.

[0092] (4) The test set is input into the trained data prediction model to obtain the chlorophyll concentration prediction result and water content prediction result of leaf sample 1.

[0093] (VII) Data prediction model evaluation;

[0094] Using the training set determination coefficient Test set determination coefficient The training set root mean square error RMSEC and the test set root mean square error RMSEP are used as the basis for evaluating the performance of the data prediction model. The training set determination coefficient Test set determination coefficient The higher the value, the better the effect of the data prediction model; the lower the root mean square error RMSEC of the training set and the root mean square error RMSEP of the test set, the stronger the stability of the data prediction model;

[0095]

[0096]

[0097] Among them, Y Pi is the predicted value of chlorophyll or leaf water content of the i-th leaf, Y i is the measured value of chlorophyll or leaf water content of the i-th leaf, is the average value of leaf chlorophyll or leaf water content in the corresponding model set, n c is the number of calculations in the training set, n P is the number of computations in the test set.

[0098] (VIII) Results and Discussion

[0099] (1) Raw spectral data analysis

[0100] The data prediction models presented by different model frameworks for processing measured data may achieve different results. In order to obtain the best data prediction model, a variety of different model frameworks are selected for comparative processing and analysis. The data is divided into 7% for training set and 3% for test set. Finally, the best model is determined as the final prediction model through the coefficient of determination and root mean square error. Machine learning is used to complete the regression analysis of the SPAD value and moisture content of tomato leaves on the original spectral data. The results are shown in Table 1. The analysis shows that in the case of all spectral wavelength combinations of chlorophyll, the three regression algorithms of LAR, GBR and BR can predict SPAD well, among which GBR has the best prediction effect. On the training set, GBR's is 0.9741, and RMSEc is 0.6526; on the test set, GBR The RR, GBR and TTR regression algorithms can predict the moisture content well in all spectral wavelength combinations, among which GBR has the best prediction effect. is 0.9850, and RMSEc is 0.0014; on the test set, GBR It is 0.4946 and the RMSep is 0.0087.

[0101] Table 1 Original spectrum analysis

[0102]

[0103]

[0104] (2) Analysis of original relative spectral data

[0105] The Clear band of the AS7343 spectral sensor 4 has a positive effect in improving measurement accuracy and data comparability, and helps to accurately identify background light and improve the signal-to-noise ratio. It can be used as a correction benchmark for other bands. The Clear band effectively compensates for changes in ambient light, sensor characteristics, and object reflectivity, and measures the overall intensity of the incident light. The original relative intensity spectrum is obtained by dividing the intensity of the Clear band by the intensity of other bands. This method effectively standardizes the data and reduces the impact of different lighting conditions on the measurement results, thereby enhancing the consistency of the results. Machine learning is used to complete the regression analysis of the SPAD value and moisture content of tomato leaves on the original relative spectral data. The results are shown in Table 2. The analysis found that when chlorophyll is at 640nm, the three regression algorithms of ABR, HR, and GPR can all predict SPAD well, among which GPR has the best prediction effect. On the training set, GPR’s is 0.8260, and RMSEc is 1.7445; on the test set, GPR is 0.7112, and the RMSEp is 2.0525. When the moisture content is in the 855nm band of the spectrum, the three regression algorithms RFR, GBR, and BR can predict the moisture content well, among which GBR has the best prediction effect. is 0.9401, and RMSEc is 0.0028; on the test set, GBR It is 0.6667 and the RMSep is 0.0067.

[0106] Table 2 Analysis of original relative spectral data

[0107]

[0108]

[0109] (3) Spectral data noise reduction analysis

[0110] In order to effectively extract information from spectral data, a variety of spectral preprocessing methods were used to achieve data denoising. Spectral data are often affected by noise due to improper equipment installation, ambient lighting and other factors, which in turn reduces the clarity and resolvability of the signal. Therefore, the implementation of a variety of noise reduction strategies is crucial to improving spectral quality. Through systematic denoising, the interference signal is suppressed, so that the original spectrum and original relative spectral data after denoising can be improved in signal strength and resolvability, making the effective information in the spectrum more prominent, which helps to improve the significance and reliability of spectral features, and shows good stability and robustness under multiple experimental conditions.

[0111] In order to evaluate the effectiveness of the denoising preprocessing method used for the original spectra, the three regression algorithms that showed the highest accuracy in SPAD prediction in the original spectra were selected: LAR, GBR and BR; at the same time, the three best performing regression algorithms were also used for the prediction of leaf moisture content: RR, GBR and TTR. Detailed analysis and prediction were performed on each wavelength and all wavelength combinations to systematically evaluate the impact of the preprocessed spectral data on the prediction model. The analysis results are shown in Table 3. In the SPAD prediction part, it was observed that the four denoising methods MSC, NORM, CR and DS all showed a trend of improving prediction accuracy. In particular, the effects of MSC and NORM were the most significant, and their determination coefficients on the test set were The values ​​of 0.6782, 0.7070 and 0.6832 respectively show the positive effect of pre-processing on SPAD data. However, under the denoising condition of NORM, Lasso regression failed to successfully predict the SPAD value. In the prediction analysis of leaf moisture content, the four denoising methods SNV, WD, NORM and DS also showed a trend of enhancing prediction accuracy. Among them, the improvement effect of WD and NORM is particularly obvious, with a determination coefficient of 0.11 on the test set. They are 0.5456, 0.5780 and 0.6227 respectively, indicating that denoising plays a decisive role in improving model performance.

[0112] Table 3. Noise reduction analysis of raw spectral data

[0113]

[0114] (4) Noise reduction analysis of raw relative spectral data

[0115] The effectiveness of the denoising preprocessing method used for the original relative spectra was evaluated, and the three regression algorithms that showed the highest accuracy in SPAD prediction in the original relative spectra were selected: ABR, HR, and GPR; at the same time, the three best performing regression algorithms were also applied to the prediction of leaf moisture content: RFR, GBR, and BR. Detailed analysis and prediction were performed on each wavelength and all wavelength combinations to systematically evaluate the impact of the preprocessed spectral data on the prediction model. The analysis results are shown in Table 4. In the SPAD prediction part, it was observed that both MSC and CR denoising methods showed a trend of improving prediction accuracy. Its determination coefficient on the test set The results are 0.7113, 0.7155 and 0.7128 respectively, indicating the positive effect of preprocessing on SPAD data. In the prediction analysis of leaf moisture content, the SNV and DT denoising methods are particularly obvious compared with other denoising methods, but the prediction performance is similar to that of the original relative spectral data, and the determination coefficient on the test set is They are 0.6604, 0.6667 and 0.6554 respectively.

[0116] Table 4. Noise reduction analysis of original relative spectral data

[0117]

[0118]

[0119] (5) Multi-model fusion analysis

[0120] The method of multi-model fusion analysis is adopted to improve the prediction accuracy of leaf moisture content and chlorophyll content. Two fusion models, VotingRegressor and StackingRegressor, are selected to fuse the regression models with high correlation between leaf moisture content and chlorophyll in the original spectral data, original relative spectral data and denoised spectra to achieve more accurate prediction of the target variable. VotingRegressor combines the advantages of each model by weighted voting on the outputs of multiple basic regression models to reduce the possible deviations of individual models and enhance the robustness of the overall model. StackingRegressor constructs a multi-layer model and inputs the prediction results of different regression models as new features into a higher-level model, thereby achieving more complex feature combinations and more accurate regression predictions. The analysis results are shown in Table 5. For the original spectral data, the prediction of SPAD is completed. The VR and SR fusion methods use lasso, gbr and br as the basic learning models, which can effectively predict and analyze SPAD through all spectral combinations, and improve the best prediction performance of the original spectrum. The determination coefficient on the test set is At the same time, the VR and SR fusion method uses ridge, gbr and ttr as the basic learning model, which can effectively predict and analyze the leaf moisture content through all spectral combinations and single 855nm band spectra, and improves the best prediction performance of the original spectrum. The determination coefficient on the test set is For the original denoised spectral data, under the NORM denoised original spectral data, the SR fusion method uses ridge and ttr as the basic learning model, which can effectively predict and analyze the leaf moisture content through the spectral data of the 855nm band, and the best prediction performance of the original denoised spectrum is improved. The determination coefficient on the test set is For the relative spectral data, the SR fusion method uses ada, huber and gpr as the basic learning model, which can effectively predict and analyze the blade SPAD through the spectral data of the 640nm band, and improves the best prediction performance of the relative spectrum. The determination coefficient on the test set is The relative spectral denoising data is used for leaf moisture content and SPAD prediction, and the effect is not significantly improved.

[0121] Table 5 Multi-model fusion analysis of spectral data

[0122]

[0123] IX. Conclusion

[0124] The present invention verifies the feasibility of the designed wearable spectral sensor for plant leaves, especially its application in monitoring plant physiological characteristics. By integrating the light source with the AS7343 spectral sensor, an integrated solution of active near-ultraviolet, visible light and near-infrared multi-band light sources is realized. This design not only improves the flexibility of the sensor, but also enhances its adaptability in different spectral ranges. During the experiment, the wearable spectral sensor was installed on the back of the leaf by magnetic connection, so as to minimize the impact on leaf photosynthesis and respiration during long-term data collection. The original spectral data of leaf reflection collected by the AS7343 spectral sensor is combined with the information of the Clear band to calculate the original relative spectral data. For the original spectral data and the original relative spectral data, a variety of processing methods are used, including machine learning, noise reduction preprocessing and multi-model fusion, to complete the regression prediction analysis of leaf moisture content and chlorophyll concentration. Among them, the prediction effect of Gaussian process regression is the best in the multivariate scattering correction of leaf chlorophyll in the original relative spectrum, and the prediction effect is the best on the training set. is 0.8261, and RMSEc is 1.7444; on the test set The leaf moisture content is 0.7155 and the RMSEp is 2.0374. At the same time, the gradient boosting regression can effectively predict the leaf moisture content under various data preprocessing conditions of the relative spectrum. is 0.9401, RMSEc is 0.0028; on the test set It is 0.6667 and the RMSep is 0.0067.

[0125] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0126] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A flexible wearable spectral sensor for monitoring leaf physiological characteristics, characterized in that: include: The wearable spectral sensor uses two layers of flexible polyimide base material, which are respectively combined with copper wires through acrylic glue to form a thin single-layer flexible circuit board; the single-layer flexible circuit board is equipped with AS7343 spectral sensor, infrared light module, ultraviolet light module and visible light module, and the collected spectrum range is 405-855nm; the wearable spectral sensor is 0.11mm thick, weighs 0.1894g, and measures 1.2cm×1.1cm.

2. A flexible wearable spectral sensor application method for monitoring leaf physiological characteristics, characterized in that: The following steps are involved: S1. Collect chlorophyll concentration and water content measurement data of leaves at different parts of different plants based on random sampling method; S2. Acquire spectral data of the leaves at the same position as in S1 based on a wearable spectral sensor; S3. Based on the leaf spectral data obtained in S2, a data prediction model is constructed to obtain the prediction results of the chlorophyll concentration and the moisture content of the leaves; S4. Compare and analyze the chlorophyll concentration prediction results and moisture content prediction results of the leaves in S3 with the chlorophyll concentration measurement data and moisture content measurement data of the leaves obtained in S1 to evaluate the feasibility of wearable spectral sensor measurement.

3. The method for applying a flexible wearable spectral sensor for monitoring leaf physiological characteristics according to claim 2, characterized in that: In S1, the chlorophyll content of the leaves is measured by a SPAD measuring instrument to obtain chlorophyll concentration measurement data; The leaves were placed on a balance to measure their fresh weight. After recording the fresh weight, they were immediately placed in a drying oven for drying. That is, the leaves were placed in a 105°C oven for 30 minutes of withering treatment, and then the leaves were dried at 80°C. After multiple measurements, when the weight no longer changed, the moisture content was calculated to obtain the moisture content measurement data.

4. The method for applying a flexible wearable spectral sensor for monitoring leaf physiological characteristics according to claim 3 is characterized in that: The calculation formula for moisture content is as follows: Among them, FW is fresh weight, DW is dry weight, and RWC is moisture content.

5. The method for applying a flexible wearable spectral sensor for monitoring leaf physiological characteristics according to claim 2, characterized in that: In S2, the wearable spectral sensor is installed on the back of the leaf by magnetic connection, and the average value of each band is taken as the final spectral data, so as to obtain the spectral data of the same part of the leaf as in S1.

6. The method for applying a flexible wearable spectral sensor for monitoring leaf physiological characteristics according to claim 2, characterized in that: In S3, based on the leaf spectral data obtained in S2, a data prediction model is constructed to obtain the leaf chlorophyll concentration prediction results and the water content prediction results. The specific contents are as follows: S31. performing noise reduction preprocessing on the acquired spectral data; S32. dividing the preprocessed spectral data into a training set and a test set; S33. Input the training set into the data prediction model to train the data prediction model. After several trainings, a trained data prediction model is obtained. S34. Input the test set into the trained data prediction model to obtain the chlorophyll concentration prediction results and water content prediction results of the leaves.

7. The method for applying a flexible wearable spectral sensor for monitoring leaf physiological characteristics according to claim 6, characterized in that: The spectral data acquired in S31 are subjected to noise reduction preprocessing including multivariate scattering correction processing, standard normal transformation processing, wavelet denoising processing, smoothing processing, normalization processing, continuous projection method, first-order derivative processing, and detrending processing.

8. The method for applying a flexible wearable spectral sensor for monitoring leaf physiological characteristics according to claim 6, characterized in that: In S32, the first 70% of the preprocessed spectral data is divided into a training set, and the last 30% is divided into a test set.

9. The method for applying a flexible wearable spectral sensor for monitoring leaf physiological characteristics according to claim 6, characterized in that: Using the training set determination coefficient Test set determination coefficient The training set root mean square error RMSEC and the test set root mean square error RMSEP are used as the basis for evaluating the performance of the data prediction model. The training set determination coefficient Test set determination coefficient The higher the value, the better the effect of the data prediction model; the lower the root mean square error RMSEC of the training set and the root mean square error RMSEP of the test set, the stronger the stability of the data prediction model; in, is the predicted value of chlorophyll or leaf water content of the i-th leaf, Y i is the measured value of chlorophyll or leaf water content of the i-th leaf, is the average value of leaf chlorophyll or leaf water content in the corresponding model set, n c is the number of calculations in the training set, n P is the number of computations in the test set.

Images