Method and system for acquiring dark adaptation chlorophyll fluorescence signal under bright adaptation condition

Abstract

The invention relates to the technical field of chlorophyll fluorescence detection, in particular to a method and a system for acquiring a dark adaptation chlorophyll fluorescence signal under a bright adaptation condition. According to the method, an InceptionTime-LSTM model is constructed, the model comprises an InceptionTime module, an LSTM module, a first global average pooling layer and a full connection layer, a chlorophyll fluorescence kinetics curve (OJIPL) under bright adaptation is input into the InceptionTime module and the LSTM module in parallel, and InceptionTime output features and global OJIP time dependence features are obtained respectively; wherein the Inception Time module comprises a plurality of Inception modules which are connected in sequence, and the output of the corresponding Inception module is in residual connection with the OJIPL curve every other preset step length; after passing through the first global average pooling layer, the output features of the InceptionTime are spliced with the global OJIP time dependent features and then input into the full connection layer, and a chlorophyll fluorescence kinetics curve (OJIPD) under dark adaptation and a plurality of chlorophyll fluorescence parameters (ChlFParmsD) are output. According to the method, the OJIPD curve and the ChlFParmsD value can be estimated without dark adaptation, and the method has important theoretical and practical significance.

Description

Technical Field

[0001] The present invention relates to the technical field of chlorophyll fluorescence detection, and in particular to a method and system for acquiring dark-adapted chlorophyll fluorescence signals under light-adapted conditions. Background Art

[0002] Photosynthesis is the most important chemical reaction on earth, which can convert light energy into stable chemical energy. The complexity of the process makes the study of photosynthesis challenging. When photosynthesis occurs, the antenna molecules in the photosynthetic membrane absorb light energy, and the absorbed energy is transferred in the form of excited electrons. Part of it is used for photochemical reactions, part of it is dissipated in the form of heat, and a very small part is dissipated in the form of chlorophyll a fluorescence. These three parts show a competitive coupling relationship, and changes in one will cause changes in the other two. Therefore, the changes in the chlorophyll a fluorescence kinetic curve obtained under measurable dark adaptation conditions become an effective probe for photosynthesis. The many fluorescence parameters extracted from the dark-adapted chlorophyll fluorescence kinetic curve have a wide range of applications. They can be used to characterize photosynthetic efficiency indicators, reveal the physiological state of plants, and evaluate the effects of various environmental stresses on crops.

[0003] However, the acquisition of chlorophyll fluorescence kinetic curves obtained under dark adaptation conditions requires the tested plants to be dark adapted for at least 15 minutes. The waiting time for dark adaptation seriously affects the test efficiency. Therefore, it is of great theoretical and practical significance to use the chlorophyll fluorescence kinetic curves measured under light adaptation conditions to determine the chlorophyll fluorescence kinetic curves and fluorescence parameters under dark adaptation conditions. However, there are great differences between the characteristic points of the chlorophyll fluorescence kinetic curves obtained under dark adaptation conditions and those obtained under light adaptation conditions, which will lead to large errors in the chlorophyll a fluorescence parameters calculated under light and dark adaptation. If a large number of samples need to be collected, it will inevitably consume a lot of manpower, time and leaf clips; and the test of different batches of leaves will be aggravated due to the waiting time, and the natural environment and plant physiological state when chlorophyll fluorescence is collected are not uniform. Since the chlorophyll fluorescence kinetic curves obtained under dark adaptation conditions and the chlorophyll fluorescence kinetic curves obtained under light adaptation conditions are both from the photochemical reaction of the same leaf, they both correspond to a common biological system, so there is a close connection between the two.

[0004] Since the mapping relationship between the two may be a very complex nonlinear relationship, and the deep learning technology that has developed rapidly in recent years and has been widely used may be able to achieve the OJIP L OJIP is obtained by direct mapping of kinetic curves D Kinetic curve. In the prior art, a researcher used the least squares support vector machine (LSSVM) to obtain the fluorescence parameter F under dark adaptation from the chlorophyll fluorescence kinetic curve data measured under light adaptation conditions.V / F M , and achieved good results, but this was only for one chlorophyll fluorescence parameter. The chlorophyll fluorescence kinetic curve obtained under dark-adapted conditions and many other chlorophyll fluorescence parameters with wide applications were not involved. Therefore, despite the rapid development of artificial intelligence algorithms, it is still a difficult problem to obtain the chlorophyll fluorescence kinetic curve and fluorescence parameters under dark-adapted conditions using the chlorophyll fluorescence kinetic curve measured under light-adapted conditions. Summary of the invention

[0005] To this end, the technical problem to be solved by the present invention is to overcome the defect in the prior art of lacking a method for directly obtaining the chlorophyll fluorescence kinetic curve and fluorescence parameters under dark adaptation conditions by using the chlorophyll fluorescence kinetic curve measured under light adaptation conditions.

[0006] In order to solve the above technical problems, the present invention provides a method for obtaining dark-adapted chlorophyll fluorescence signals under light-adapted conditions, comprising the following steps:

[0007] Constructing an InceptionTime-LSTM model, the InceptionTime-LSTM model includes: an InceptionTime module, an LSTM module, a first global average pooling layer and a fully connected layer;

[0008] The chlorophyll fluorescence kinetics curve under light adaptation is input into the InceptionTime module and the LSTM module in parallel, multi-scale features are extracted and long-term and short-term dependencies are processed, and the InceptionTime output features and the global OJIP time dependency features are obtained respectively, wherein the InceptionTime module includes a plurality of sequentially connected Inception modules, and at every preset step, the output of the corresponding Inception module is residually connected with the chlorophyll fluorescence kinetics curve under light adaptation;

[0009] After the InceptionTime output feature passes through the first global average pooling layer, it is spliced ​​with the global OJIP time-dependent feature along the channel dimension and input into the fully connected layer for feature processing and mapping, and outputs the chlorophyll fluorescence kinetic curve under dark adaptation and multiple chlorophyll fluorescence parameters.

[0010] Preferably, each Inception module includes: a first branch, a second branch, a second global average pooling layer, and an SE attention module;

[0011] The first branch includes: a bottleneck layer, a one-dimensional convolution block with multiple convolution kernels of different sizes, and the second branch includes: a maximum pooling layer and a one-dimensional convolution block.

[0012] Preferably, the input feature sequence is input into the Inception module to extract the target feature sequence, including the following steps:

[0013] The input feature sequence is input into two branches in parallel. The input feature sequence passes through the first branch to extract features from the input feature sequence, and the first branch feature sequence corresponding to the one-dimensional convolution block of different sizes of convolution kernels is output. The input feature sequence passes through the maximum pooling layer of the second branch, and performs downsampling operation to obtain the feature sequence after maximum pooling processing. The feature sequence after maximum pooling processing passes through the one-dimensional convolution block, and the second branch feature sequence is output;

[0014] The second branch feature sequence is concatenated with the first branch feature sequence corresponding to the one-dimensional convolution blocks with different sizes of convolution kernels along the channel dimension to obtain an initial feature sequence;

[0015] After the initial feature sequence passes through the second global average pooling layer, the SE attention module adjusts the weight of each channel to obtain the SE attention feature sequence;

[0016] The initial feature sequence and the SE attention feature sequence are concatenated along the channel dimension to obtain the target feature sequence.

[0017] Preferably, the InceptionTime module comprises 9 Inception modules connected in sequence.

[0018] Preferably, after the InceptionTime output feature passes through the first global average pooling layer, it is spliced ​​with the global OJIP time-dependent feature along the channel dimension and then input into the fully connected layer for feature processing and mapping, and the chlorophyll fluorescence kinetic curve under dark adaptation and multiple chlorophyll fluorescence parameters are output. The specific formula is:

[0019]

[0020] Among them, F out represents the chlorophyll fluorescence kinetic curve and multiple chlorophyll fluorescence parameters under dark adaptation, g fc represents the fully connected layer, g avg represents the first global average pooling layer, F inceptiontime Indicates the InceptionTime output feature, Represents a concat operation.

[0021] Preferably, the training process of the InceptionTime-LSTM model is:

[0022] Acquire a chlorophyll fluorescence kinetic curve data set, wherein the chlorophyll fluorescence kinetic curve data set includes: a plurality of chlorophyll fluorescence kinetic curves under dark adaptation and chlorophyll fluorescence kinetic curves under light adaptation;

[0023] After preprocessing the chlorophyll fluorescence dynamics curve dataset, it is divided into training set, validation set and test set in proportion;

[0024] The chlorophyll fluorescence kinetic curve under light adaptation in the training set is used as the input of the InceptionTime-LSTM model, and the chlorophyll fluorescence kinetic curve under dark adaptation and multiple chlorophyll fluorescence parameters are output. The mean square error loss function and Adam optimizer are used to train the InceptionTime-LSTM model. The trained InceptionTime-LSTM model is verified by the validation set to obtain the target InceptionTime-LSTM model.

[0025] Preferably, when acquiring the chlorophyll fluorescence kinetic curve data set, the data acquisition device is any one of the FluorPen handheld chlorophyll fluorometer and the FluorX-FX001 handheld chlorophyll fluorometer.

[0026] Preferably, the chlorophyll fluorescence kinetics curve data set is preprocessed, including performing Z-Score normalization processing on the chlorophyll fluorescence kinetics curve data set.

[0027] Preferably, multiple chlorophyll fluorescence parameters include: maximum variable fluorescence, relative variable fluorescence at 2ms, relative variable fluorescence at 30ms, maximum photochemical efficiency of PSII, maximum rate of reduction of primary electron acceptor of PSII, maximum photochemical efficiency of PSII, quantum yield for electron transfer, probability of captured excitons transferring electrons to other electron acceptors in the electron transfer chain that exceed primary electron acceptors of PSII, light energy absorbed by unit reaction center, light energy captured by unit reaction center, light energy used for electron transfer by unit reaction center, energy dissipated by unit reaction center, energy absorbed by unit cross-sectional area, energy captured by unit cross-sectional area, quantum yield of electron transfer by unit cross-sectional area, energy dissipated by unit cross-sectional area, number of reaction centers per unit area, quantum yield of electron transfer to the end of PSI, efficiency of electron transfer to the end of PSI, performance index based on absorbed light energy, and comprehensive performance index of chlorophyll fluorescence parameters.

[0028] The present invention also provides a dark-adapted chlorophyll fluorescence signal acquisition system under light-adapted conditions, comprising:

[0029] A model construction module, used to construct an InceptionTime-LSTM model, wherein the InceptionTime-LSTM model includes: an InceptionTime module, an LSTM module, a first global average pooling layer and a fully connected layer;

[0030] A feature extraction module is used to input the chlorophyll fluorescence kinetics curve under light adaptation into the InceptionTime module and the LSTM module in parallel, and obtain the InceptionTime output feature and the global OJIP time-dependent feature respectively, wherein the InceptionTime module includes a plurality of sequentially connected Inception modules, and at every preset step, the output of the corresponding Inception module is residually connected with the chlorophyll fluorescence kinetics curve under light adaptation;

[0031] The output module is used to output the InceptionTime feature after passing through the first global average pooling layer, splicing it with the global OJIP time-dependent feature along the channel dimension, and then input it into the fully connected layer for feature processing and mapping, and output the chlorophyll fluorescence kinetic curve under dark adaptation and multiple chlorophyll fluorescence parameters.

[0032] The above technical solution of the present invention has the following beneficial effects compared with the prior art:

[0033] The present invention discloses a method and system for acquiring dark-adapted chlorophyll fluorescence signals under light-adapted conditions, and constructs an InceptionTime-LSTM model, wherein the InceptionTime-LSTM model comprises: an InceptionTime module, an LSTM module, a first global average pooling layer and a fully connected layer; the chlorophyll fluorescence kinetic curve under light-adapted conditions is input into the InceptionTime module and the LSTM module in parallel, and the InceptionTime output features and the global OJIP time-dependent features are obtained respectively, wherein the InceptionTime module comprises a plurality of sequentially connected Inception modules, and at every preset step length, the output of the corresponding Inception module is residually connected with the chlorophyll fluorescence kinetic curve under light-adapted conditions; after the InceptionTime output features pass through the first global average pooling layer, they are spliced ​​with the global OJIP time-dependent features along the channel dimension and then input into the fully connected layer, and feature processing and mapping are performed, and the chlorophyll fluorescence kinetic curve under dark-adapted conditions and a plurality of chlorophyll fluorescence parameters are output. The present invention realizes OJIP by using the constructed InceptionTime-LSTM model. L Kinetic curve to OJIP D Kinetic curves and ChlFParms D Complex mapping, that is, OJIP of plant leaves can be accurately estimated without dark adaptation D Kinetic curves and ChlFParms DIt saves a lot of time for related tests, and also provides a mitigation solution for the inconsistency between the natural environment and the plant physiological state when chlorophyll fluorescence is collected due to the waiting time during dark adaptation, which has important practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below according to specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:

[0035] Figure 1 This is a structural diagram of a method and system for acquiring dark-adapted chlorophyll fluorescence signals under light-adapted conditions according to the present invention. Figure 1 (a) is the chlorophyll fluorescence dynamics curve under light adaptation. Figure 1 (b) shows the chlorophyll fluorescence kinetic curve and multiple chlorophyll fluorescence parameters under dark adaptation. Figure 1 (c) in the figure is the structure diagram of the InceptionTime-LSTM model. Figure 1 (d) in the figure is the structure diagram of the InceptionTime module. Figure 1 (e) in the figure is the structure diagram of the Inception module.

[0036] Figure 2 This is the error analysis diagram of different models in estimating the characteristic points of chlorophyll fluorescence dynamics curve under dark adaptation. Figure 2 (a) represents the LS-SVR model. Figure 2 (b) in the figure represents the LSTM model. Figure 2 (c) in the figure represents the CNN-LSTM model. Figure 2 (d) in the figure represents the iTransformer model. Figure 2 (e) in the figure represents the PatchTST model. Figure 2 (f) in the figure represents the DLinear model. Figure 2 (g) in the figure represents the TSMixer model. Figure 2 (h) in the figure represents the InceptionTime-LSTM model.

[0037] Figure 3 It is a performance graph of multiple dark-adapted chlorophyll fluorescence parameter estimates on different models. Figure 3 (a) shows the R values ​​corresponding to different chlorophyll fluorescence parameters. 2 value, Figure 3 (b) shows the RMSPE values ​​corresponding to different chlorophyll fluorescence parameters. Figure 3 (c) in the figure shows the RPD values ​​corresponding to different chlorophyll fluorescence parameters. Figure 3 (d) in the figure shows the SMAPE values ​​corresponding to different chlorophyll fluorescence parameters.

[0038] Figure 4 This is a schematic diagram of the estimation results of the chlorophyll fluorescence dynamics curve under dark adaptation by the InceptionTime-LSTM model. Figure 4 (a) is the chlorophyll fluorescence kinetic curve measured under light and dark adaptation conditions. Figure 4 (b) in the figure shows the measured chlorophyll fluorescence kinetic curve under dark adaptation and the estimated chlorophyll fluorescence kinetic curve under dark adaptation.

[0039] Figure 5 This is the estimated result of the key characteristic points of the chlorophyll fluorescence kinetics curve under dark adaptation.

[0040] Figure 6 This is the error analysis diagram of chlorophyll fluorescence parameters of different plant samples under multiple dark adaptation conditions on the InceptionTime-LSTM model. Figure 6 (a) in the figure shows the error analysis diagram of Osmanthus fragrans on the InceptionTime-LSTM model. Figure 6 (b) in the figure shows the error analysis diagram of Euonymus japonicus on the InceptionTime-LSTM model. Figure 6 (c) in the figure shows the error analysis diagram of Camellia on the InceptionTime-LSTM model. Figure 6 (d) in the figure shows the error analysis diagram of green pepper on the InceptionTime-LSTM model. Figure 6 (e) shows the error analysis diagram of Sakura Oshima on the InceptionTime-LSTM model. Figure 6 (f) in the figure shows the error analysis diagram of rice on the InceptionTime-LSTM model. Figure 6 (g) in the figure shows the error analysis diagram of Wutacai on the InceptionTime-LSTM model. Figure 6 (h) in the figure shows the error analysis diagram of Dwarf Yellow Cabbage on the InceptionTime-LSTM model.

[0041] Figure 7 It is a data distribution diagram of the measured and estimated values ​​of chlorophyll fluorescence parameters under multiple dark adaptation conditions.

[0042] Figure 8 This is a performance chart of chlorophyll fluorescence parameters of different plant samples under multiple dark adaptation conditions on the InceptionTime-LSTM model. Figure 8 (a) shows the chlorophyll fluorescence parameters of different plant samples corresponding to R 2 value, Figure 8 (b) shows the RMSPE values ​​of chlorophyll fluorescence parameters of different plant samples. Figure 8 (c) shows the RPD values ​​of chlorophyll fluorescence parameters of different plant samples. Figure 8 (d) in the figure shows the SMAPE values ​​of chlorophyll fluorescence parameters of different plant samples.

[0043] Fig. 9 It is a flow chart of the steps of a method for obtaining dark-adapted chlorophyll fluorescence signals under light-adapted conditions.

[0044] Fig.10 It is a structural diagram of a dark-adapted chlorophyll fluorescence signal acquisition system under light-adapted conditions. DETAILED DESCRIPTION

[0045] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0046] Reference Figure 1 , 9 As shown, Embodiment 1 of the present invention provides a method for acquiring dark-adapted chlorophyll fluorescence signals under light-adapted conditions, comprising the following steps:

[0047] Step S1: constructing an InceptionTime-LSTM model, wherein the InceptionTime-LSTM model includes: an InceptionTime module, an LSTM module, a first global average pooling layer and a fully connected layer;

[0048] Step S2: Inputting the chlorophyll fluorescence kinetics curve under light adaptation into the InceptionTime module and the LSTM module in parallel, extracting multi-scale features and processing long-term and short-term dependencies, and obtaining the InceptionTime output features and the global OJIP time dependency features respectively, wherein the InceptionTime module includes a plurality of sequentially connected Inception modules, and at every preset step, the output of the corresponding Inception module is residually connected with the chlorophyll fluorescence kinetics curve under light adaptation;

[0049] Step S3: After the InceptionTime output feature passes through the first global average pooling layer, it is spliced ​​with the global OJIP time-dependent feature along the channel dimension and input into the fully connected layer for feature processing and mapping, and outputs the chlorophyll fluorescence kinetic curve under dark adaptation and multiple chlorophyll fluorescence parameters.

[0050] In this embodiment, preferably, each Inception module includes: a first branch, a second branch, a second global average pooling layer, and an SE attention module;

[0051] The first branch includes: a bottleneck layer, a one-dimensional convolution block with multiple convolution kernels of different sizes, and the second branch includes: a maximum pooling layer and a one-dimensional convolution block.

[0052] In this embodiment, preferably, the input feature sequence is input into the Inception module, and the target feature sequence is extracted by inputting the input feature sequence into the Inception module, including the following steps:

[0053] The input feature sequence is input into two branches in parallel. The input feature sequence passes through the first branch to extract features from the input feature sequence, and the first branch feature sequence corresponding to the one-dimensional convolution block of different sizes of convolution kernels is output. The input feature sequence passes through the maximum pooling layer of the second branch, and performs downsampling operation to obtain the feature sequence after maximum pooling processing. The feature sequence after maximum pooling processing passes through the one-dimensional convolution block, and the second branch feature sequence is output;

[0054] The second branch feature sequence is concatenated with the first branch feature sequence corresponding to the one-dimensional convolution blocks with different sizes of convolution kernels along the channel dimension to obtain an initial feature sequence;

[0055] After the initial feature sequence passes through the second global average pooling layer, the SE attention module adjusts the weight of each channel to obtain the SE attention feature sequence;

[0056] The initial feature sequence and the SE attention feature sequence are concatenated along the channel dimension to obtain the target feature sequence.

[0057] In this embodiment, preferably, the InceptionTime module includes 9 sequentially connected Inception modules. As the depth of the network increases, the computational cost and training time will increase significantly. The use of 9 Inception modules ensures the performance of the InceptionTime-LSTM model while ensuring the computational efficiency of the InceptionTime-LSTM model.

[0058] In this embodiment, preferably, after the InceptionTime output feature passes through the first global average pooling layer, it is spliced ​​with the global OJIP time-dependent feature along the channel dimension and then input into the fully connected layer for feature processing and mapping, and the chlorophyll fluorescence kinetic curve under dark adaptation and multiple chlorophyll fluorescence parameters are output. The specific formula is:

[0059]

[0060] Among them, F out represents the chlorophyll fluorescence kinetic curve and multiple chlorophyll fluorescence parameters under dark adaptation, g fc represents the fully connected layer, g avgrepresents the first global average pooling layer, F inceptiontime Indicates the InceptionTime output feature, Represents a concat operation.

[0061] The Inception module has the ability to process features of different scales and reduce the number of parameters. However, the traditional Inception module uses a two-dimensional convolution kernel design, which is not suitable for the OJIP time series estimation task. Therefore, this paper constructs a new Inception Block framework such as Figure 1 As shown in (e), its advantage lies in the design of a 1D convolution kernel and the adjustment of the parallel processing structure of the convolution kernel size from 3, 5, 7... to 51. This design improves the estimation accuracy of the OJIP time series, and the multi-scale convolution kernel can better extract the biochemical reaction information across various time scales during photosynthesis. In addition, the SE attention module is introduced. The SE mechanism further optimizes the feature expression by adaptively adjusting the weight of each channel to improve OJIP. D Kinetic curves and ChlFParms D The estimation accuracy.

[0062] To further improve the model performance, this embodiment connects 9 layers of Inception Block in series in the InceptionTime-LSTM model, and introduces residual connections after the 3rd, 6th and 9th layers of the input layer to alleviate the gradient vanishing problem in the deep network and promote the effective flow of information in the network, thereby constructing the InceptionTime structure as shown in FIG. Figure 1 (d) as shown.

[0063] Estimating the evolution of important information in the photosynthesis process requires not only paying attention to local features, but also capturing the overall dynamic information of chlorophyll fluorescence during photosynthesis. Therefore, a long short-term memory (LSTM) module is introduced to handle the temporal dependency in the photosynthesis process.

[0064] Finally, the output of InceptionTime is passed to the fully connected layer for model fusion after global average pooling operation and the global OJIP time-dependent features captured by the LSTM module, and further feature processing and mapping are performed to output OJIP. D Kinetic curves and ChlFParms D .

[0065] In this embodiment, preferably, the training process of the InceptionTime-LSTM model is:

[0066] Acquire a chlorophyll fluorescence kinetic curve data set, wherein the chlorophyll fluorescence kinetic curve data set includes: a plurality of chlorophyll fluorescence kinetic curves under dark adaptation and chlorophyll fluorescence kinetic curves under light adaptation;

[0067] After preprocessing the chlorophyll fluorescence dynamics curve dataset, it is divided into training set, validation set and test set in proportion;

[0068] The chlorophyll fluorescence kinetic curve under light adaptation in the training set is used as the input of the InceptionTime-LSTM model, and the chlorophyll fluorescence kinetic curve under dark adaptation and multiple chlorophyll fluorescence parameters are output. The mean square error loss function and Adam optimizer are used to train the InceptionTime-LSTM model. The trained InceptionTime-LSTM model is verified by the validation set to obtain the target InceptionTime-LSTM model.

[0069] In this embodiment, preferably, when acquiring the chlorophyll fluorescence kinetic curve data set, the data acquisition device is any one of the FluorPen handheld chlorophyll fluorometer and the FluorX-FX001 handheld chlorophyll fluorometer.

[0070] In this embodiment, preferably, preprocessing the chlorophyll fluorescence kinetics curve data set includes: performing Z-Score normalization processing on the chlorophyll fluorescence kinetics curve data set. The formula is:

[0071]

[0072] Among them, X n represents the input sequence, Y n represents the normalized output sequence. Assume that each data size is an m×n matrix, where m represents the number of samples, n represents the length of each sample, and the vector X mean , X std They represent the mean and standard deviation of each column of the input sequence, both of size 1×n.

[0073] In this embodiment, preferably, the multiple chlorophyll fluorescence parameters include: maximum variable fluorescence F V , Relative variable fluorescence V at 2ms J , Relative variable fluorescence V at 30ms I , PSII maximum photochemical efficiency F V / F M , Q A Maximum reduction rate M0, maximum photochemical efficiency of PSII Quantum yield for electron transfer The trapped excitons deliver electrons to the electron transport chain beyond the PSII primary electron acceptor Q A The probability of other electron acceptors Ψ0, the light energy absorbed per reaction center ABS / RC, the light energy captured per reaction center TR0 / RC, the light energy used for electron transfer per reaction center ET0 / RC, the energy dissipated per reaction center DI0 / RC, the energy absorbed per unit cross-sectional area ABS / CS0, the energy captured per unit cross-sectional area TR0 / CS0, the quantum yield of electron transfer per unit cross-sectional area ET0 / CS0, the energy dissipated per unit cross-sectional area DI0 / CS0, the number of reaction centers per unit area RC / CS0, the quantum yield of electron transfer to the end of PSI Efficiency of electron transfer to the PSI terminus Performance index PI based on absorbed light energy abs , chlorophyll fluorescence parameter comprehensive performance index PI total .

[0074] As shown in Table 1, Table 1 is the parameter definition and calculation formula of the commonly used chlorophyll fluorescence kinetic curve OJIP characteristic points and fluorescence parameter ChlFParms.

[0075] Table 1

[0076]

[0077]

[0078]

[0079] Among them, ABS represents the amount of light absorbed, RC represents the reaction center, CS0 represents the initial cross-sectional area, TR0 represents the initial captured energy flux, DI0 represents the initial dissipated flux, and ET0 represents the initial electron transfer flux.

[0080] In Example 2 of the present invention, various plant samples were collected from the campus of Jiangnan University and the greenhouse of Wuxi Zhihui Tianyuan Agricultural Development Co., Ltd. The chlorophyll fluorescence kinetic curve dataset contains a total of 8 plant samples: rice, camellia, osmanthus, holly euonymus, green pepper, Oshima cherry, dwarf yellow cabbage, and wutacai. The experimental data collection time is from the summer of 2019 to the winter of 2023. A total of 15,891 copies of OJIP kinetic curve data under dark adaptation and OJIP kinetic curve data under light adaptation were collected. L With OJIP D The kinetic curves correspond one to one.

[0081] The experimental data acquisition equipment used was the FluorPen handheld chlorophyll fluorometer from PSI in the Czech Republic and the FluorX-FX001 handheld chlorophyll fluorometer from Green Vision Technology Co., Ltd. in China. The minimum sampling interval of both instruments is 10 μs, which meets the requirements of OJIP data acquisition. Before the experiment, the excitation light intensity of the instrument needs to be set. In the experiment, the excitation light intensity of the FluorPen chlorophyll fluorometer was set to 2400 μmolphotons·m -2 ·s -1 The excitation intensity of the FluorX-FX001 chlorophyll fluorometer was set to 2700 μmolphotons·m -2 ·s -1 During data collection, OJIP measurements were first performed on plant leaves that were not dark-adapted, and then OJIP measurements were performed on the same sampling position of the plant after 20 minutes of dark adaptation. In this way, a set of OJIP kinetic curve data corresponding to light and dark was obtained through two measurements.

[0082] Due to the large amount of data samples, it is inevitable that the leaf clips are forgotten to be opened and the instrument is removed before the measurement is completed during the data collection process. These invalid tests can be easily eliminated according to the amplitude and continuity of the data values. In this study, the abnormal values ​​generated by these erroneous operations are processed by Python (Version 3.8.10) scripts. Since the variable frequency sampling parameter settings of the chlorophyll fluorescence instruments of the two companies used for data collection are not exactly the same, this embodiment processes the time resolution correspondence of the two types of instrument data points by interpolating and resampling the data collected by FluorX-FX001. In order to facilitate subsequent model training, Z-Score normalization is used in this embodiment.

[0083] The size of the chlorophyll fluorescence kinetic curve data set included in this embodiment is 15891, which includes 8 types of plant samples. During the experiment, the entire data set is randomly divided into a training set, a validation set, and a test set, and it is ensured that the 8 types of plant samples are evenly distributed in the three sets. The division ratio of the training set, the validation set, and the test set is shown in Table 2. Table 2 is a table of the division ratios of the chlorophyll fluorescence kinetic curve data set.

[0084] Table 2

[0085]

[0086] Three data partitioning methods are used in model training, namely, the training set accounts for 60%, 70% and 80%, the validation set accounts for 20%, 15% and 10%, and the test set accounts for 20%, 15% and 10%. a represents the number of samples.

[0087] For model training, the input data size is set in three ways as shown in Table 2 and the performance of the models is compared. In order to make all experimental results comparable, all deep neural network models use the following hyperparameter settings: the loss function is set to mean square error MSELoss, the initial learning rate is 0.0001, and when the loss value no longer improves, the learning rate is adjusted to half of the original after waiting for 5 rounds; Adam is selected as the optimizer, and the number of training rounds is set to 200. Since CNN-LSTM did not converge after 200 rounds of training, the number of training rounds for the InceptionTime-LSTM model is set to 300.

[0088] After each model is trained, the InceptionTime-LSTM model is evaluated with the best performing parameter settings. The evaluation indicators include root mean square percentage error RMSPE, determination coefficient R 2 , residual prediction deviation RPD, symmetric mean absolute percentage error SMAPE. The smaller the RMSPE and SMAPE values, the better, and they are independent of the data size, which is very helpful for comparing the prediction errors between different data sets. 2 The larger the value, the better. The calculation of the four indicators is as follows:

[0089]

[0090]

[0091]

[0092]

[0093] After comparing the performance of various models, the relative error of dark-adapted chlorophyll fluorescence estimated by the model structure with the best performance is calculated as follows:

[0094]

[0095] Among them, RMSPE is the root mean square percentage error, R 2 is the coefficient of determination, RPD is the residual prediction deviation, SMAPE is the symmetric mean absolute percentage error, RE is the relative error of dark-adapted chlorophyll fluorescence, c is the length of the output feature, and Y i is the sample measurement value, is the average measurement value, Y i is the predicted value. Since the constructed InceptionTime-LSTM is a multi-input and multi-output model, Y i It represents only one output feature vector, whose size is the number of input samples.

[0096] The output of the InceptionTime-LSTM model contains two important parts: the estimated OJIP D Kinetic curves (458 dynamic data sequences) and 20 ChlFParms D , the output is denormalized to restore it to the scale of the original input features, and then statistical analysis is performed on the estimated data and the original labels to evaluate the model performance.

[0097] Table 3 shows the performance of each model on OJIP under different data set partition ratios. D Performance of kinetic curve estimation. The InceptionTime-LSTM model constructed in this paper performs best in all data set partitions, with the lowest RMSPE and SMAPE values ​​(0.064 and 0.041 for the 20% test set) and the highest R 2 The value and RPD value (0.988 and 9.206 for the 20% test set) show that the InceptionTime-LSTM model has the best prediction accuracy and good model stability.

[0098] As the test set ratio decreases (from 20% to 10%), the estimated performance of all models varies slightly, which may be due to the imbalance of samples in the dataset. Overall, the performance of each model is best at a test set ratio of 20%. Therefore, this section further evaluates the model with a 20% test set ratio.

[0099] Table 3 shows the performance of each model in OJIP when the test set ratio is 20%, 15% and 10%. D Performance comparison on dynamic curve estimation, with RMSPE, R 2 , RPD and SMAPE.

[0100] Table 3

[0101]

[0102] Since there are 458 data points on the OJIP kinetic curve, it is impossible to show the error of each data point. The most commonly used chlorophyll fluorescence values ​​are 50μs, 300μs, 2ms, 20ms and the maximum fluorescence intensity time point. The corresponding relative ChlF values ​​represent F0, F 300μs 、F J 、F I and F M Five feature points, so about the estimated OJIP DThe dynamic curve only takes these 5 points for analysis. After the model is trained, the loss value has converged to a certain range, and the best training model is saved for OJIP D Relative error analysis of fluorescence values ​​at five important moments on the kinetic curve.

[0103] like Figure 2 As shown, Figure 2 This is an error analysis diagram of different models in estimating characteristic points of chlorophyll fluorescence dynamics curve under dark adaptation. A total of 8 models were compared for their relative error results. Figure 2 (a) represents the LS-SVR model. Figure 2 (b) in the figure represents the LSTM model. Figure 2 (c) in the figure represents the CNN-LSTM model. Figure 2 (d) in the figure represents the iTransformer model. Figure 2 (e) in the figure represents the PatchTST model. Figure 2 (f) in the figure represents the DLinear model. Figure 2 (g) in the figure represents the TSMixer model. Figure 2 (h) in the figure represents the InceptionTime-LSTM model. The data in each model accounts for more than 95.8% of the total data.

[0104] Figure 2 Demonstrated different models in OJIP D Violin plot of error analysis on kinetic curve estimation. The scales in each subplot represent the quantile, 25% quantile, median, 75% quantile, and 95% quantile corresponding to the minimum relative error from bottom to top. The results show that the best performing model is InceptionTime-LSTM. The point with the largest error is F J The median is 3.50%. From the data distribution density of the relative errors of the estimation results of all models, the estimation results of the InceptionTime-LSTM model are the most concentrated, indicating that the estimation results of this model on the entire data set are stable and have good robustness.

[0105] As shown in Table 4, Table 4 shows the chlorophyll fluorescence parameters ChlFParms of different models under multiple dark adaptations D The estimated performance of each model on ChlFParms when the test set ratio is 20%, 15%, and 10%. D The performance comparison on the 2D image is based on RMSPE, R 2 , RPD and SMAPE.

[0106] Table 4

[0107]

[0108]

[0109] According to Table 4, the InceptionTime-LSTM model performs best under different data set partition ratios. 2 The RPD values ​​are the highest (0.911 and 4.330 in the 20% test set), indicating that the InceptionTime-LSTM model has the best goodness of fit and the best estimation ability. The RMSPE and SMAPE values ​​are the lowest (0.083 and 0.051 in the 20% test set), indicating that the model is D Combining Table 3 and Table 4, by comparing different models in OJIP D Kinetic curves and ChlFParms D From the performance comparison of estimation, it can be clearly seen that the InceptionTime-LSTM model performs best in all indicators, has high estimation accuracy and stability, has good generalization ability, and is suitable for different data set division ratios.

[0110] like Figure 3 As shown, Figure 3 It is a performance graph of multiple dark-adapted chlorophyll fluorescence parameter estimates on different models. Figure 3 (a) shows the R values ​​corresponding to different chlorophyll fluorescence parameters. 2 value, Figure 3 (b) shows the RMSPE values ​​corresponding to different chlorophyll fluorescence parameters. Figure 3 (c) in the figure shows the RPD values ​​corresponding to different chlorophyll fluorescence parameters. Figure 3 (d) is the SMAPE value corresponding to different chlorophyll fluorescence parameters

[0111] Figure 3 Analyzed 8 models in 20 common ChlFParms D The estimated performance on the 20% test set is obtained from various ChlFParms D From the above analysis, we can see that the InceptionTime-LSTM model has the highest goodness of fit and the best estimation accuracy, and also has the lowest estimation error. 2 From the perspective of the values ​​and RPD values, all models have good performance on the ChlF parameter V J The goodness of fit and estimation ability of RC / CS0 are the worst, and the estimation ability of RC / CS0 is the best; from the RMSPE and SMAPE values, all models have good fit for the ChlF parameter PI total The estimation error of is the largest, which provides direction for the adjustment of later model parameters.

[0112] like Figure 4 As shown, Figure 4 This is a schematic diagram of the estimation results of the chlorophyll fluorescence dynamics curve under dark adaptation by the InceptionTime-LSTM model. Figure 4 (a) is the chlorophyll fluorescence kinetic curve measured under light and dark adaptation conditions. Figure 4 (b) in the figure shows the measured chlorophyll fluorescence kinetic curve under dark adaptation and the estimated chlorophyll fluorescence kinetic curve under dark adaptation. Figure 4 As shown in (a) and (b), the chlorophyll fluorescence dynamics curve OJIP under light adaptation L After the InceptionTime-LSTM model, the chlorophyll fluorescence dynamics curve OJIP under dark adaptation can be well fitted D Kinetic curve.

[0113] like Figure 5 As shown, Figure 5 This is the estimated result of the key characteristic points of the chlorophyll fluorescence kinetic curve under dark adaptation. The OJIP curve consists of 458 points. In order to further analyze the performance of the model, Figure 5 The five most commonly used OJIP feature points F0, F 300μs 、F J 、F I and F M Correlation and linear regression analysis. Each sub-graph represents OJIP D A characteristic point of the kinetic curve, the horizontal axis represents the measured value under dark adaptation, the vertical axis represents the estimated value, and the red line represents the linear regression line of the sample. Figure 5 Analysis, OJIP D The five key feature points on the kinetic curve are highly correlated with the measured values ​​after model estimation, and their determination coefficients are greater than or equal to 0.98. The RPD values ​​are between 7.69 and 9.86, showing strong estimation ability. The RMSPE values ​​are all below 0.07, and the SMAPE values ​​are all below 0.05, indicating that the InceptionTime-LSTM model is effective in OJIP. D There is a small error in the estimation of the kinetic curve.

[0114] like Figure 6 As shown, Figure 6 This is an error analysis diagram of chlorophyll fluorescence parameters of different plant samples under multiple dark adaptation conditions on the InceptionTime-LSTM model. Figure 6 A total of 8 plant samples were included. Figure 6 (a) in the figure shows the error analysis diagram of Osmanthus fragrans on the InceptionTime-LSTM model. Figure 6(b) in the figure shows the error analysis diagram of Euonymus japonicus on the InceptionTime-LSTM model. Figure 6 (c) in the figure shows the error analysis diagram of Camellia on the InceptionTime-LSTM model. Figure 6 (d) in the figure shows the error analysis diagram of green pepper on the InceptionTime-LSTM model. Figure 6 (e) shows the error analysis diagram of Sakura Oshima on the InceptionTime-LSTM model. Figure 6 (f) in the figure shows the error analysis diagram of rice on the InceptionTime-LSTM model. Figure 6 (g) in the figure shows the error analysis diagram of Wutacai on the InceptionTime-LSTM model. Figure 6 (h) in the figure shows the error analysis diagram of dwarf yellow cabbage on the InceptionTime-LSTM model. Each plant sample accounts for more than 98.8% of the class.

[0115] Figure 6 It is clearly shown that each plant sample is in OJIP D The median relative error of the upper estimation of the kinetic curve does not exceed 5%, and the relative error of 75% of the estimated data is less than 10%. The estimation results of the model show that the InceptionTime-LSTM model can adapt to the test results of different plant samples and different environmental conditions. Figure 6 The InceptionTime-LSTM model demonstrates good generalization ability and adaptability.

[0116] like Figure 7 As shown, Figure 7 The data distribution diagram of the measured and estimated values ​​of chlorophyll fluorescence parameters under multiple dark adaptation conditions. Each sub-graph represents a chlorophyll fluorescence parameter, the horizontal axis represents the measured value of the chlorophyll fluorescence parameter, the vertical axis represents the estimated value of the chlorophyll fluorescence parameter, and the red line is the sample regression line. Figure 7 The scatter plots of the measured and estimated values ​​of several chlorophyll fluorescence parameters are plotted in Figure 2, and the R value of each parameter estimate is indicated. 2 , RMSPE, RPD and SMAPE values.

[0117] After model fitting, the 20 chlorophyll fluorescence parameters showed a strong correlation before and after dark adaptation, and their R 2 The parameters with the largest values ​​are TR0 / CS0 and RC / CS0 (R 2 =0.99), the smallest parameter is V J (R 2=0.84); the parameter with the largest RPD value is TR0 / CS0 (RPD=9.27), and the parameter with the smallest RPD value is V J (RPD = 2.51); the parameter with the largest RMSPE and SMAPE values ​​is PI total (RMSPE = 0.36, SMAPE = 0.18), the smallest parameter is (RMSPE = 0.01, SMAPE = 0.01). Therefore, OJIP L After the kinetic curves were mapped by the model, the estimated 20 ChlFParms D Very accurate.

[0118] like Figure 8 As shown, Figure 8 The performance of chlorophyll fluorescence parameters of different plant samples under multiple dark adaptation on the InceptionTime-LSTM model. These four radar charts show the performance of the InceptionTime-LSTM model on 20 ChlFParms of 8 different plant samples. D Performance on. Figure 8 (a) shows the chlorophyll fluorescence parameters of different plant samples corresponding to R 2 value, Figure 8 (b) shows the RMSPE values ​​of chlorophyll fluorescence parameters of different plant samples. Figure 8 (c) shows the RPD values ​​of chlorophyll fluorescence parameters of different plant samples. Figure 8 (d) in the figure shows the SMAPE values ​​of chlorophyll fluorescence parameters of different plant samples.

[0119] from Figure 8 It can be seen that most plants ChlFParms D The estimated R 2 The values ​​are all above 0.9, the RPD values ​​are mostly above 1.2, with the maximum reaching 7.68, the RMSPE values ​​are mostly less than 0.08, and the SMAPE values ​​are mostly less than 0.06, which is a very balanced performance on the entire dataset. D From the above analysis, we can see that PI total The estimated error in all ChlFParms D The maximum value in the estimation (RMSPE value is 0.9, SMAPE value is 0.6), but its R 2 Most of the values, except for green pepper and holly, were within the normal range, and also had a good goodness of fit.

[0120] In order to handle the light and dark mapping relationship of the OJIP dynamic curve, the OJIP LOJIP is estimated from the test value of the kinetic curve D Kinetic curves and ChlFParms D The present invention constructs InceptionTime-LSTM and collects 15891 sets of OJIP kinetic curves corresponding to light and dark in 4 years for training, verification and testing the model. Finally, it is compared with the traditional machine learning algorithm LS-SVR and the recent classic neural network models LSTM, CNN-LSTM, iTransformer, PatchTST, DLinear and TSMixer. The results show that the InceptionTime-LSTM network has the best estimation performance in this task. D Kinetic curve estimation on R 2 , RMSPE, RPD and SMAPE values ​​are 0.988, 0.063, 9.206 and 0.041 respectively; D In terms of estimation, these four indicators are 0.911, 0.083, 4.330, and 0.051, respectively. The excellent performance of the model on different samples in the dataset also shows that the model can adapt to different plant samples. Therefore, the model has good generalization and robustness.

[0121] OJIP D The kinetic curve reflects the important biochemical reaction relationships across multiple scales in the photosynthesis process. The present invention uses the constructed deep neural network model to realize OJIP L Kinetic curve to OJIP D Kinetic curves and ChlFParms D The complex mapping shows that deep neural networks can fully describe the important biochemical reactions in the photosynthetic reaction of plant leaves. Since photosynthesis involves too many complex chemical reactions, and these reaction coefficients and reaction processes are not completely clear, it is almost impossible to completely solve the photosynthesis modeling problem using mechanism modeling. This invention provides a good case for using artificial intelligence algorithms to solve a series of modeling studies related to photosynthetic reactions in the future, which has important theoretical significance and also provides a reference for OJIP. D The measurement of kinetic curves provides a new method to accurately estimate the OJIP of plant leaves without dark adaptation. D Kinetic curves and ChlFParms D It saves a lot of time for related tests, and also provides a mitigation solution for the inconsistency between the natural environment and the plant physiological state when chlorophyll fluorescence is collected due to the waiting time during dark adaptation. Therefore, it has important practical value.

[0122] like Fig.10As shown, the third embodiment of the present invention further provides a dark-adapted chlorophyll fluorescence signal acquisition system under light-adapted conditions, comprising:

[0123] A model construction module 10 is used to construct an InceptionTime-LSTM model, wherein the InceptionTime-LSTM model includes: an InceptionTime module, an LSTM module, a first global average pooling layer and a fully connected layer;

[0124] The feature extraction module 20 is used to input the chlorophyll fluorescence kinetic curve under light adaptation into the InceptionTime module and the LSTM module in parallel, and obtain the InceptionTime output feature and the global OJIP time-dependent feature respectively, wherein the InceptionTime module includes a plurality of sequentially connected Inception modules, and at every preset step, the output of the corresponding Inception module is residually connected with the chlorophyll fluorescence kinetic curve under light adaptation;

[0125] The output module 30 is used to output the InceptionTime feature through the first global average pooling layer, splice it with the global OJIP time-dependent feature along the channel dimension, and then input it into the fully connected layer for feature processing and mapping, and output the chlorophyll fluorescence kinetic curve under dark adaptation and multiple chlorophyll fluorescence parameters.

[0126] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0127] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0128] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0129] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0130] Obviously, the above embodiments are merely examples for the purpose of clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.

Claims

1. A method for obtaining dark-adapted chlorophyll fluorescence signals under light-adapted conditions, characterized in that: The following steps are involved: Constructing an InceptionTime-LSTM model, the InceptionTime-LSTM model includes: an InceptionTime module, an LSTM module, a first global average pooling layer and a fully connected layer; The chlorophyll fluorescence kinetics curve under light adaptation is input into the InceptionTime module and the LSTM module in parallel, multi-scale features are extracted and long-term and short-term dependencies are processed, and the InceptionTime output features and the global OJIP time dependency features are obtained respectively, wherein the InceptionTime module includes a plurality of sequentially connected Inception modules, and at every preset step, the output of the corresponding Inception module is residually connected with the chlorophyll fluorescence kinetics curve under light adaptation; After the InceptionTime output feature passes through the first global average pooling layer, it is spliced ​​with the global OJIP time-dependent feature along the channel dimension and input into the fully connected layer for feature processing and mapping, and outputs the chlorophyll fluorescence kinetic curve under dark adaptation and multiple chlorophyll fluorescence parameters.

2. The method for obtaining dark-adapted chlorophyll fluorescence signals under light-adapted conditions according to claim 1, characterized in that: Each Inception module includes: a first branch, a second branch, a second global average pooling layer, and a SE attention module; The first branch includes: a bottleneck layer, a one-dimensional convolution block with multiple convolution kernels of different sizes, and the second branch includes: a maximum pooling layer and a one-dimensional convolution block.

3. The method for obtaining dark-adapted chlorophyll fluorescence signals under light-adapted conditions according to claim 2, characterized in that: Input the input feature sequence into the Inception module and extract the target feature sequence, including the following steps: The input feature sequence is input into two branches in parallel. The input feature sequence passes through the first branch to extract features from the input feature sequence, and the first branch feature sequence corresponding to the one-dimensional convolution block of convolution kernels of different sizes is output. The input feature sequence passes through the maximum pooling layer of the second branch, and performs downsampling operation to obtain the feature sequence after maximum pooling processing. The feature sequence after maximum pooling processing passes through the one-dimensional convolution block, and the second branch feature sequence is output; The second branch feature sequence is concatenated with the first branch feature sequence corresponding to the one-dimensional convolution blocks with different sizes of convolution kernels along the channel dimension to obtain an initial feature sequence; After the initial feature sequence passes through the second global average pooling layer, the SE attention module adjusts the weight of each channel to obtain the SE attention feature sequence; The initial feature sequence and the SE attention feature sequence are concatenated along the channel dimension to obtain the target feature sequence.

4. The method for acquiring dark-adapted chlorophyll fluorescence signals under light-adapted conditions according to claim 1, characterized in that: The InceptionTime module includes 9 Inception modules connected in sequence.

5. The method for acquiring dark-adapted chlorophyll fluorescence signals under light-adapted conditions according to claim 1, characterized in that: After the InceptionTime output feature passes through the first global average pooling layer, it is spliced ​​with the global OJIP time-dependent feature along the channel dimension and input into the fully connected layer for feature processing and mapping, and the chlorophyll fluorescence kinetic curve under dark adaptation and multiple chlorophyll fluorescence parameters are output. The specific formula is: Among them, F out represents the chlorophyll fluorescence kinetic curve and multiple chlorophyll fluorescence parameters under dark adaptation, g fc represents the fully connected layer, g avg represents the first global average pooling layer, F inceptiontime Indicates the InceptionTime output feature, Represents a concat operation.

6. The method for acquiring dark-adapted chlorophyll fluorescence signals under light-adapted conditions according to claim 1, characterized in that: The training process of the InceptionTime-LSTM model is: Acquire a chlorophyll fluorescence kinetic curve data set, wherein the chlorophyll fluorescence kinetic curve data set includes: a plurality of chlorophyll fluorescence kinetic curves under dark adaptation and chlorophyll fluorescence kinetic curves under light adaptation; After preprocessing the chlorophyll fluorescence dynamics curve dataset, it is divided into training set, validation set and test set in proportion; The chlorophyll fluorescence kinetic curve under light adaptation in the training set is used as the input of the InceptionTime-LSTM model, and the chlorophyll fluorescence kinetic curve under dark adaptation and multiple chlorophyll fluorescence parameters are output. The mean square error loss function and Adam optimizer are used to train the InceptionTime-LSTM model. The trained InceptionTime-LSTM model is verified by the validation set to obtain the target InceptionTime-LSTM model.

7. The method for acquiring dark-adapted chlorophyll fluorescence signals under light-adapted conditions according to claim 6, characterized in that: When obtaining the chlorophyll fluorescence kinetic curve data set, the data acquisition device is any one of the FluorPen handheld chlorophyll fluorometer and FluorX-FX001 handheld chlorophyll fluorometer.

8. The method for acquiring dark-adapted chlorophyll fluorescence signals under light-adapted conditions according to claim 6, characterized in that: The chlorophyll fluorescence kinetic curve data set is preprocessed, including Z-Score normalization processing of the chlorophyll fluorescence kinetic curve data set.

9. The method for acquiring dark-adapted chlorophyll fluorescence signals under light-adapted conditions according to claim 1, characterized in that: Multiple chlorophyll fluorescence parameters include: maximum variable fluorescence, relative variable fluorescence at 2ms, relative variable fluorescence at 30ms, maximum photochemical efficiency of PSII, maximum rate of reduction of primary electron acceptor of PSII, maximum photochemical efficiency of PSII, quantum yield for electron transfer, probability of captured excitons transferring electrons to other electron acceptors in the electron transfer chain that exceed primary electron acceptor of PSII, light energy absorbed by unit reaction center, light energy captured by unit reaction center, light energy used for electron transfer by unit reaction center, energy dissipated by unit reaction center, energy absorbed by unit cross-sectional area, energy captured by unit cross-sectional area, quantum yield of electron transfer by unit cross-sectional area, energy dissipated by unit cross-sectional area, number of reaction centers per unit area, quantum yield of electron transfer to the end of PSI, efficiency of electron transfer to the end of PSI, performance index based on absorbed light energy, and comprehensive performance index of chlorophyll fluorescence parameters.

10. A dark-adapted chlorophyll fluorescence signal acquisition system under light-adapted conditions, characterized in that: include: A model construction module, used to construct an InceptionTime-LSTM model, wherein the InceptionTime-LSTM model includes: an InceptionTime module, an LSTM module, a first global average pooling layer and a fully connected layer; A feature extraction module is used to input the chlorophyll fluorescence kinetics curve under light adaptation into the InceptionTime module and the LSTM module in parallel, and obtain the InceptionTime output feature and the global OJIP time-dependent feature respectively, wherein the InceptionTime module includes a plurality of sequentially connected Inception modules, and at every preset step, the output of the corresponding Inception module is residually connected with the chlorophyll fluorescence kinetics curve under light adaptation; The output module is used to output the InceptionTime feature after passing through the first global average pooling layer, splicing it with the global OJIP time-dependent feature along the channel dimension, and then input it into the fully connected layer for feature processing and mapping, and output the chlorophyll fluorescence kinetic curve under dark adaptation and multiple chlorophyll fluorescence parameters.

Images