A tea seedling high temperature state recognition method and system based on chlorophyll fluorescence imaging

Abstract

This invention discloses a method for identifying the high-temperature state of tea seedlings based on chlorophyll fluorescence imaging, comprising the following steps: Step S1, acquiring chlorophyll fluorescence parameters and images of tea seedlings under different high-temperature stress conditions, performing Spearman rank correlation analysis on each parameter, and determining that the parameter significantly correlated with the stress level is the maximum photosynthetic efficiency Fv / Fm; Step S2, preprocessing the chlorophyll fluorescence images corresponding to Fv / Fm to construct a high-temperature stress state dataset; Step S3, constructing an improved YOLOv11-MEIP model, wherein the improved YOLOv11-MEIP model replaces the backbone network of the original network with a MobileNetV4 network and introduces EUCB, iRMB, and PConv modules; Step S4, training, validating, and testing the improved YOLOv11-MEIP model; Step S5, deploying the tested improved YOLOv11-MEIP model for identifying the high-temperature stress state of tea seedlings. This study aims to provide a rapid and accurate monitoring method for the high-temperature stress state of tea seedlings, meeting the real-time monitoring needs of large-scale tea gardens.

Description

Technical Field

[0001] This invention belongs to the field of agricultural intelligent monitoring technology, specifically relating to a method and system for identifying the high-temperature status of tea seedlings based on chlorophyll fluorescence imaging. Background Technology

[0002] Tea (Camellia sinensis), an important economic crop in my country, belongs to the genus Camellia in the family Theaceae. It prefers a warm and humid growing environment of 20-25℃ and can tolerate temperatures up to 35-40℃. When the ambient temperature exceeds 40℃, the tea plant's photosynthetic system is damaged and growth is inhibited; the critical lethal temperature is 45℃. Against the backdrop of global warming and frequent extreme heat events, the sustainable development of the tea industry faces severe challenges.

[0003] Photosynthesis is the physiological process in tea plants most sensitive to high-temperature stress, and chlorophyll fluorescence parameters can directly reflect the functional state of photosynthetic system II (PSII), providing core evidence for plant stress monitoring. Chlorophyll fluorescence imaging technology, as a non-destructive detection method, has the advantages of being rapid, sensitive, and non-invasive, and has been applied to plant stress physiology research. However, traditional monitoring methods based on this technology have the following drawbacks:

[0004] Relying on manual analysis of fluorescence parameters is subjective and inefficient, making it difficult to meet the real-time monitoring needs of large-scale tea gardens. Existing models mostly focus on adverse conditions such as drought and diseases, and there is a lack of dedicated identification models for high-temperature stress on tea seedlings. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for identifying the high-temperature state of tea seedlings based on chlorophyll fluorescence imaging, providing a rapid and accurate monitoring method for the high-temperature stress state of tea seedlings.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for identifying the high-temperature status of tea seedlings based on chlorophyll fluorescence imaging, comprising the following steps:

[0007] Step S1: Obtain chlorophyll fluorescence parameters and images of tea seedlings under different high temperature stress conditions, perform Spearman rank correlation analysis on each parameter, and determine the parameter that is significantly correlated with the stress level as the maximum photosynthetic efficiency Fv / Fm;

[0008] Step S2: Preprocess the chlorophyll fluorescence images corresponding to Fv / Fm to construct a high-temperature stress state dataset;

[0009] Step S3: Construct an improved YOLOv11-MEIP model. The improved YOLOv11-MEIP model replaces the backbone network of the original network with the MobileNetV4 network and introduces the EUCB module, iRMB module and PConv module.

[0010] Step S4: Divide the high temperature stress state dataset into a training set, a validation set, and a test set according to a preset ratio, and train, validate, and test the improved YOLOv11-MEIP model.

[0011] Step S5: Deploy the tested and approved improved YOLOv11-MEIP model for the identification of high temperature stress in tea seedlings.

[0012] Furthermore, step S1 specifically includes:

[0013] Thirteen chlorophyll fluorescence parameters of tea seedlings under different high-temperature stress conditions were obtained using a multifunctional plant photosynthetic phenotyping system.

[0014] The Spearman rank correlation analysis method was used to calculate the correlation coefficient between each chlorophyll fluorescence parameter and the high temperature stress level, and Fv / Fm was determined as the key characteristic parameter.

[0015] Furthermore, the preprocessing in step S2 includes:

[0016] The chlorophyll fluorescence images corresponding to Fv / Fm were converted to grayscale.

[0017] Histogram equalization is used to enhance image contrast;

[0018] Perform image normalization and size standardization;

[0019] Data augmentation techniques, including rotation, translation, flipping, brightness and contrast adjustments, are used to expand the dataset.

[0020] Furthermore, in step S3, the improved YOLOv11 model consists of an input layer, a backbone network layer, a neck network layer, and a detection head, wherein:

[0021] The input layer is used to receive preprocessed chlorophyll fluorescence images;

[0022] The backbone network layer replaces the original network with MobileNetV4 network to extract multi-scale features of chlorophyll fluorescence images.

[0023] The neck network layer includes an EUCB module, an iRMB module, and a PConv module, which are used for feature upsampling and fusion;

[0024] The detection head is used to output the identification results of the high temperature stress state of tea seedlings.

[0025] Furthermore, the EUCB module includes upsampling layers and convolutional layers to maintain the spatial resolution of features while reducing computational cost; the iRMB module adopts an inverse residual structure to improve feature representation through an expansion-compression strategy; and the PConv module adopts a partial convolution strategy to reduce computational redundancy and memory access.

[0026] Furthermore, step S4 specifically includes:

[0027] The high-temperature stress state dataset was divided into a training set, a validation set, and a test set in a ratio of 8:1:1.

[0028] The improved YOLOv11 model is trained using the training set until the loss function value is less than a preset threshold.

[0029] The trained model is validated using the validation set. If the validation accuracy of the model is lower than the preset accuracy threshold, the training set is expanded and training continues.

[0030] The validated model is tested using the test set, and the model's precision, recall, and average accuracy are calculated. If the test metrics do not meet the preset requirements, the training set is expanded and training continues; if the requirements are met, the model training is complete.

[0031] Furthermore, it includes the following modules:

[0032] The chlorophyll fluorescence parameter acquisition module is used to acquire chlorophyll fluorescence parameters and images of tea seedlings under different high temperature stress conditions;

[0033] The feature parameter determination module is used to perform Spearman rank correlation analysis on various chlorophyll fluorescence parameters to determine Fv / Fm as the key feature parameter;

[0034] The dataset construction module is used to preprocess the chlorophyll fluorescence images corresponding to Fv / Fm to construct a high-temperature stress state dataset;

[0035] An improved YOLOv11 model building module is used to build deep learning models that include the MobileNetV4 backbone network, EUCB module, iRMB module and PConv module;

[0036] The model training and testing module is used to divide the high-temperature stress state dataset into training, validation, and testing sets to train, validate, and test the model.

[0037] The high-temperature state recognition module is used to deploy the trained model and intelligently identify the high-temperature stress state of tea seedlings.

[0038] Furthermore, the high-temperature stress state includes four levels: normal state, mild stress, moderate stress, and severe stress, with corresponding temperature ranges of 25-30℃, 31-35℃, 36-40℃, and 41-45℃, respectively.

[0039] Furthermore, the improved YOLOv11 model includes an input layer, a backbone network layer, a neck network layer, and a detection head, wherein the backbone network layer adopts the MobileNetV4 network, and the neck network layer includes an EUCB module, an iRMB module, and a PConv module.

[0040] Furthermore, in the improved YOLOv11 model, the EUCB module is used to maintain the spatial resolution of features while reducing computational load, the iRMB module is used to improve feature representation through inverse residual structure, and the PConv module is used to reduce computational redundancy and memory access.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] (1) By analyzing the fluorescence parameters and images of tea seedling leaves under different high temperature stresses obtained by chlorophyll fluorescence imaging technology, and combining them with the proposed improved YOLOv11 model, we can achieve efficient, non-destructive and intelligent identification of the state of tea seedlings under different high temperature stresses, and provide a fast and accurate monitoring method for the high temperature stress state of tea seedlings to meet the real-time monitoring needs of large-scale tea gardens.

[0043] (2) In order to reduce the number of parameters in the model network, improve the efficiency and accuracy of feature extraction, reduce computational redundancy and memory access, and improve model accuracy and efficiency while maintaining low computational cost, MobileNetV4 network, EUCB module, iRMB module and PConv are introduced into the YOLOv11 model respectively; compared with YOLOv11, YOLOv10, YOLOv8, SSD, Faster RCNN and other models, the efficiency and accuracy of hierarchical identification of high temperature stress are significantly improved, and a technical solution that can be promoted for intelligent phenotypic analysis in the agricultural field is provided.

[0044] (3) By combining chlorophyll fluorescence imaging technology with the lightweight YOLOv11-MEIP model, a tea tree stress state identification system was established, realizing non-destructive and high-precision dynamic monitoring, providing a new technical path for plant stress physiology research. Attached Figure Description

[0045] Figure 1 A bar chart showing the correlation between chlorophyll fluorescence parameters;

[0046] Figure 2Image of chlorophyll fluorescence and image grayscale histogram;

[0047] Figure 3 To improve the YOLOv11 architecture. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Please see Figures 1-3 This invention provides a method and system technical solution for identifying the high-temperature status of tea seedlings based on chlorophyll fluorescence imaging.

[0050] A method for identifying the high-temperature status of tea seedlings based on chlorophyll fluorescence imaging includes the following steps:

[0051] Step S1: Obtain chlorophyll fluorescence parameters and images of tea seedlings under different high temperature stress conditions, perform Spearman rank correlation analysis on each parameter, and determine the parameter that is significantly correlated with the stress level as the maximum photosynthetic efficiency Fv / Fm;

[0052] Step S2: Preprocess the chlorophyll fluorescence images corresponding to Fv / Fm to construct a high-temperature stress state dataset;

[0053] Step S3: Construct an improved YOLOv11-MEIP model. The improved YOLOv11-MEIP model replaces the original network backbone with the MobileNetV4 network and introduces the EUCB module, iRMB module and PConv module.

[0054] Step S4: Divide the high temperature stress state dataset into training set, validation set and test set according to a preset ratio, and train, validate and test the improved YOLOv11-MEIP model;

[0055] Step S5: Deploy the tested and approved improved YOLOv11-MEIP model for the identification of high temperature stress in tea seedlings.

[0056] By analyzing the fluorescence parameters and images of tea seedling leaves under different high-temperature stresses obtained through chlorophyll fluorescence imaging technology, and combining them with the proposed improved YOLOv11 model, we can achieve efficient, non-destructive, and intelligent identification of the state of tea seedlings under different high-temperature stresses, providing a rapid and accurate monitoring method for the high-temperature stress state of tea seedlings.

[0057] To reduce the number of network parameters, improve the efficiency and accuracy of feature extraction, reduce computational redundancy and memory access, and maintain low computational cost while improving model accuracy and efficiency, the YOLOv11 model incorporates the MobileNetV4 network, the EUCB (Efficient Up-Convolution Block) module, the iRMB (Inverted Residual Mobile Block) module, and the PConv (Partial Convolution) module. Compared with models such as YOLOv11, YOLOv10, YOLOv8, SSD, and Faster RCNN, this significantly improves the efficiency and accuracy of hierarchical identification of high-temperature stress levels, providing a scalable technical solution for intelligent phenotypic analysis in agriculture.

[0058] By combining chlorophyll fluorescence imaging technology with the lightweight YOLOv11-MEIP model, a stress state identification system for tea trees was established, realizing non-destructive and high-precision dynamic monitoring, and providing a new technical approach for plant stress physiology research.

[0059] Step S1 is as follows:

[0060] Thirteen chlorophyll fluorescence parameters of tea seedlings under different high-temperature stress conditions were obtained using a multifunctional plant photosynthetic phenotyping system.

[0061] Specifically, one-year-old Yunnan large-leaf tea seedlings of variety "Yun Kang 10" with consistent growth were selected and planted in plastic pots with a diameter of 16cm and a height of 17cm. The seedlings were then nurtured at a constant temperature of 20℃ for 14 days to ensure they adapted to the new growing environment. During this period, 30 healthy seedlings with similar stem diameter and good growth were selected and divided into experimental and control groups. The seedlings were subjected to high-temperature stress in a temperature- and humidity-controlled incubator. Four different temperature gradients (25℃, 35℃, 40℃, and 45℃) were set for high-temperature stress treatment, with three replicates for each gradient. 25℃ served as the control temperature, while 30℃, 36℃, and 42℃ were the high-temperature stress temperatures. The temperature and humidity controlled incubator used was model RQX-300H, with a temperature control range of 5–50℃, a humidity control range of 50–90%RH, a light intensity of 12000 Lx, a temperature fluctuation range of ±1.5℃, and a temperature uniformity range of ±2℃. The environmental settings are shown in Table 1. Based on the different temperature levels set in the incubator, the high-temperature stress on tea seedlings was divided into four levels: CK, LV1, LV2, and LV3. A special high-brightness fluorescent lamp in the incubator provided a light intensity of 200 μmol / m². -2 s -1White light illumination simulated natural daytime lighting, while fluorescent lights were turned off during simulated nighttime periods. After 24 hours of treatment, 13 chlorophyll fluorescence parameters of tea seedlings were obtained using a multifunctional plant photosynthetic phenotyping system, including Fv / Fm, ΦPSII, NPQ, qP, ETR, F0, Fm, Fv, Fv' / Fm', qN, Y(NO), Y(NPQ), and Y(II).

[0062] Table 1 Environmental settings for temperature and humidity controlled incubator

[0063]

[0064] Fluorescence image acquisition equipment: The PlantExplorer series multifunctional plant photosynthetic phenotyping system was used to acquire chlorophyll fluorescence parameters and fluorescence images of tea seedlings. The multifunctional plant photosynthetic phenotyping system includes... Figure 1 As shown, this system employs innovative multispectral chlorophyll fluorescence and visible light imaging technology, combined with the latest LED, CCD, and communication technologies, to achieve innovative measurements of plant phenotypes. The system can acquire chlorophyll fluorescence imaging simultaneously with RGB, chlorophyll, and anthocyanin imaging, with an imaging area of ​​40cm × 53cm. The device's built-in photochemical lamp is a 4-channel multispectral LED system, including two channels for white light, red light, blue light, and far-red light. This multispectral LED system allows for precise spectral control and can be used to study plant responses to different wavelengths of light. Its photochemical light intensity is 100-600 μmol / m² at 60cm. -2 s -1 The intensity is adjustable. To ensure that the Photosystem II (PSII) reaction center is fully open, the tea seedlings to be tested need to undergo a dark adaptation process of about 20 minutes before collecting chlorophyll fluorescence data. After the dark adaptation process, the fluorescence parameters and images of the tea seedlings are measured in the instrument, with each sample measured for 11-12 minutes.

[0065] Correlation analysis of fluorescence parameters: Chlorophyll fluorescence parameters can be directly measured using a multifunctional plant photosynthetic phenotyping system and presented in a numerically visualized form. In addition to the system's built-in values, 13 chlorophyll fluorescence parameters can be obtained through measurement. These parameters can serve as a basis for determining whether tea seedlings are affected by high-temperature stress; their specific meanings are shown in Table 2.

[0066] Table 2 Chlorophyll fluorescence parameters and their meanings

[0067]

[0068]

[0069]

[0070]

[0071]

[0072]

[0073] The formula for calculating photosynthetic parameters is as shown above, where F is the fluorescence intensity recorded before the saturation pulse is turned on under light adaptation.

[0074] Spearman's rank correlation analysis was used to calculate the correlation coefficients between various chlorophyll fluorescence parameters and the level of high-temperature stress, identifying Fv / Fm as the key characteristic parameter. Specifically, Spearman's rank correlation coefficient analysis was performed on 13 groups of chlorophyll fluorescence parameters under high-temperature stress conditions in tea seedlings. When calculating the Spearman's rank correlation coefficients, the quantitative values ​​of the high-temperature stress level were based on different high-temperature stress levels set in the experimental design, and used to perform correlation analysis with chlorophyll fluorescence parameters to identify parameters significantly correlated with high-temperature stress. The results are shown in Table 3. The analysis results showed that five characteristic parameters had a high correlation with the stress level, among which the fluorescence parameter Fv / Fm had the highest correlation with the stress level (-0.886). Therefore, the fluorescence parameter Fv / Fm was selected as the key indicator for judging the level of high-temperature stress. Figure 1 The results of Spearman correlation analysis are presented, and the degree of correlation among the various parameters can be intuitively observed through the visualization of bar charts.

[0075] Table 3. Correlation between chlorophyll fluorescence parameters and high temperature stress level

[0076]

[0077] Note: * indicates p < 0.05, ** indicates p < 0.01, *** indicates p < 0.001

[0078] Fluorescence Image Analysis: Gray-level histograms can intuitively reflect the distribution of gray levels in a grayscale image and are an important tool for analyzing image differences. They reflect the frequency of each gray level by counting the number of pixels at each gray level. In a gray-level histogram, the horizontal axis represents the gray level, and the vertical axis represents the frequency of that gray level. As a function of gray levels, the vertical axis of a gray-level histogram is usually normalized to the interval [0, 1] to more clearly show the distribution of each gray level. The specific calculation formula is shown below.

[0079]

[0080] In the formula, It refers to the grayscale level of a pixel. It has grayscale The number of pixels, and These are the number of rows and columns of the image, which represent the total number of pixels in the image.

[0081] Figure 2 The fluorescence images of the chlorophyll fluorescence parameter Fv / Fm and their corresponding grayscale histograms are shown. According to the results shown in the figure, when the tea seedling leaves are healthy, the gray value distribution range of the Fv / Fm parameter is [0.7, 0.9], and the fluorescence image appears green. When subjected to LV1 high temperature stress, the gray value distribution range changes to [0.5, 0.9], with some gray values ​​rising to the range of [0.5, 0.7], the leaf edges begin to turn yellow, and some leaves turn completely yellow. When subjected to LV2 high temperature stress, the gray value distribution range further changes to [0.3, 0.9], with about one-third of the gray values ​​located in the range of [0.3, 0.7], most of the leaves appear yellow, and some areas of the leaves begin to turn red. When subjected to LV3 high temperature stress, the gray value distribution range is [0.1, 0.9], with the gray values ​​transitioning from the range of [0.7, 0.9] to the range of [0.1, 0.3], with some gray values ​​rising to the range of [0.1, 0.3], most of the leaf edges are reddish-brown, and only a small number of leaf veins remain yellowish-green.

[0082] Furthermore, the preprocessing in step S2 includes:

[0083] The chlorophyll fluorescence images corresponding to Fv / Fm were converted to grayscale.

[0084] Histogram equalization is used to enhance image contrast;

[0085] Perform image normalization and size standardization;

[0086] Data augmentation techniques, including rotation, translation, flipping, brightness and contrast adjustment, were used to expand the dataset, ultimately constructing a dataset containing four levels of high temperature stress (normal state, mild stress, moderate stress and severe stress).

[0087] In step S3, the improved YOLOv11 model consists of an input layer, a backbone network layer, a neck network layer, and a detection head, wherein:

[0088] The input layer is used to receive the preprocessed chlorophyll fluorescence image;

[0089] The backbone network layer replaces the original network with MobileNetV4, which is lightweight and can efficiently extract multi-scale features of chlorophyll fluorescence images.

[0090] The neck network layer contains EUCB, iRMB, and PConv modules, which are used for feature upsampling and fusion.

[0091] EUCB module: includes upsampling layers and convolutional layers, which reduce computational cost while maintaining the spatial resolution of features;

[0092] iRMB module: adopts inverted residual structure and improves feature representation capability through expansion-compression strategy;

[0093] PConv module: adopts a partial convolution strategy, performing convolution calculations only on a portion of the input channels to reduce computational redundancy and memory access;

[0094] The detection head is used to output the identification results of the high temperature stress status of tea seedlings, including the stress level category and confidence level.

[0095] The EUCB module includes upsampling and convolutional layers to maintain the spatial resolution of features while reducing computation; the iRMB module uses an inverse residual structure to improve feature representation through an expansion-compression strategy; and the PConv module uses a partial convolution strategy to reduce computational redundancy and memory access.

[0096] This invention provides an improved, lightweight YOLOv11 network model for recognizing fluorescence images of tea seedlings under different temperature levels. The structure of the improved network model is as follows: Figure 3 As shown:

[0097] (1) In order to reduce the number of network parameters, improve model accuracy and efficiency while maintaining low computational cost, and lay the foundation for subsequent mobile device deployment, the YOLOv11 backbone network is improved by using MobileNetV4 network.

[0098] (2) In order to achieve efficient feature upsampling and fusion, and improve the computational efficiency and feature representation ability of the model, the EfficientUp-ConvolutionBlock (EUCB) module is introduced into the YOLOv11 network.

[0099] (3) In order to maintain the lightweight nature of the model while significantly improving the efficiency and accuracy of feature extraction, the InvertedResidualMobileBlock (iRMB) module was introduced into YOLOv11.

[0100] (4) PartialConvolution (PConv) was introduced into the model to reduce computational redundancy and memory access.

[0101] Furthermore, step S4 specifically involves:

[0102] The high-temperature stress dataset was divided into training, validation, and test sets in a ratio of 8:1:1.

[0103] The improved YOLOv11 model is trained using the training set until the loss function value is less than a preset threshold.

[0104] The trained model is validated using a validation set. If the validation accuracy of the model is lower than the preset accuracy threshold, the training set is expanded and training continues.

[0105] The validated model is tested using a test set, and the model's precision, recall, and average accuracy are calculated. If the test metrics do not meet the preset requirements, the training set is expanded and training continues; if the requirements are met, the model training is complete.

[0106] Specifically, the constructed high-temperature stress state dataset is divided into a training set, a validation set, and a test set in an 8:1:1 ratio:

[0107] The improved YOLOv11-MEIP model was trained using the training set with a batch size of 16, an initial learning rate of 0.01, and the Adam optimizer. The training lasted for 100 epochs until the loss function value was less than 0.001.

[0108] The trained model is validated using a validation set, and the validation accuracy is calculated. If the validation accuracy is below 95%, the training set is expanded and training continues.

[0109] The validated model is tested using a test set, and its precision, recall, and mean AP50 are calculated. If the test metrics do not meet the preset requirements (precision < 98%, recall < 98%, or mAP50 < 98.5%), the training set is expanded and training continues; if the requirements are met, the model training is complete.

[0110] Step S5: Model Deployment and Application. The trained improved YOLOv11-MEIP model is converted to ONNX format, optimized using TensorRT, and deployed to an edge computing device. In practical applications, Fv / Fm fluorescence images of tea seedlings are acquired using a portable chlorophyll fluorescence imager and input into the deployed model. The model outputs the high-temperature stress level of the tea seedlings (normal state: 25-30℃; mild stress: 31-35℃; moderate stress: 36-40℃; severe stress: 41-45℃) and confidence level in real time.

[0111] MobileNetV4 is the latest generation of the MobileNet series models, featuring a general and efficient architecture. Its core innovations include the Universal Inverted Bottleneck (UIB) search block, Mobile Multi-Query Attention (Mobile MQA), and an optimized neural architecture search (NAS). UIB integrates the MobileNet Inverted Bottleneck (IB) block, ConvNext, Feed Forward Network (FFN), and a novel Extra Depthwise (ExtraDW) variant. The two optional depthwise convolutions in the UIB block have four possible instantiations. MobileNet Inverted Bottleneck performs spatial blending on the activations of the extended features to increase model capacity at a lower cost; ConvNext allows for spatial blending of larger kernel sizes at a lower cost by performing spatial blending before expansion; ExtraDW combines the advantages of ConvNext and IB, increasing network depth and receptive field at a lower cost; FFN is a stack of two 1x1 pointwise convolutions (PWs) with activation and normalization layers in between. This design allows for expansion of the receptive field at each network stage as needed, enhancing feature extraction capabilities and maximizing computational utilization.

[0112] Evaluation Metrics and Operating Environment: To verify the overall performance of the improved YOLOv11 model, this study selected P (Precision), R (Recall), Balance Score (F1), Average Precision (AP), and mean Average Precision (mAP) as evaluation metrics to assess the model's object detection performance. The calculation expressions for these evaluation metrics are as follows:

[0113]

[0114]

[0115]

[0116]

[0117] In the formula, TP represents the number of samples correctly identified by the model, FP represents the number of samples that actually exist but are incorrectly identified by the model, and FN represents the number of samples that the model does not detect.

[0118] This invention improves the YOLOv11 network model. To verify the effectiveness of the improved model, the effects of the improvements on each module were statistically analyzed, and the results are shown in Table 4.

[0119] Table 4 Comparison Results of Ablation Experiments

[0120]

[0121] Note: M indicates MobileNetV4 improvement; E indicates Efficient up-convolution block improvement; I indicates Inverted Residual Mobile Block improvement; P indicates Partial Convolution improvement.

[0122] As shown in Table 4, using MobileNetV4 as the backbone of the YOLOv11 model significantly reduced the number of network parameters while improving both accuracy and efficiency. Precision and mAP50 increased by 0.9% and 1.17%, respectively, while performance metrics such as parameters, gradients, GFLOPs, and weight decreased by 16.2%, 16.2%, 1.2%, and 1.6%, respectively. Improvements to the Efficient up-convolution block optimized the upsampling process, thereby improving the model's efficiency and feature representation capabilities when processing images, resulting in improvements of Precision and mAP50 of 0.94% and 1.5%, respectively. Improvements to the Inverted Residual Mobile Block, by optimizing the residual block structure, enhanced the model's learning and generalization abilities. Without significantly increasing model complexity or computational burden, Precision, Recall, and mAP50 improved by 1.94%, 6.57%, and 2.87%, respectively, effectively improving the model's performance on visual tasks. The Partial Convolution improvement reduces the model's parameters and computational burden by introducing its lightweight structure. Both parameters and gradients are reduced by 3.2% compared to the original model, while GFLOPs and weight are reduced by 0.6 and 0.4, respectively. Meanwhile, the precision, recall, and mAP50 metrics are improved by 0.93%, 6.26%, and 2.27%, respectively.

[0123] In summary, the YOLOv11-MEIP model achieves a significant performance improvement compared to the original YOLOv11 model. Although the improved model has an increased number of layers, the Precision, Recall, and mAP50 metrics are significantly improved, increasing by 4.05%, 7.86%, and 3.42%, respectively. Meanwhile, the Model's Parameters, Gradients, and GFLOPs have all decreased substantially, decreasing by 29.45%, 29.45%, and 30.76%, respectively. The Model Weight also decreased by 1.6 MB, further confirming the effectiveness of these module improvements in enhancing model performance. These improvements not only improve the model's detection accuracy but also enhance its applicability and reliability in future practical applications.

[0124] In summary, the YOLOv11-MEIP model achieves a significant performance improvement compared to the original YOLOv11 model. Although the improved model has an increased number of layers, the Precision, Recall, and mAP50 metrics are significantly improved, increasing by 4.05%, 7.86%, and 3.42%, respectively. Meanwhile, the Model's Parameters, Gradients, and GFLOPs have all decreased substantially, decreasing by 29.45%, 29.45%, and 30.76%, respectively. The Model Weight also decreased by 1.6 MB, further confirming the effectiveness of these module improvements in enhancing model performance. These improvements not only improve the model's detection accuracy but also enhance its applicability and reliability in future practical applications.

[0125] In the field of deep learning, upsampling and convolution operations are key steps in tasks such as image reconstruction, super-resolution, and semantic segmentation. However, traditional upsampling methods often suffer from low computational efficiency, large parameter counts, and insufficient feature fusion. To address these issues, an efficient up-convolution block (EUCB) is introduced, aiming to achieve efficient feature upsampling and fusion through innovative structural design and optimization strategies, thereby improving the computational efficiency and feature representation capabilities of the model.

[0126] EUCB upsamples the feature map of the current stage stepwise to match its dimension and resolution with the feature map of the next skip connection. First, EUCB performs an upsampling operation, doubling the size of the input feature map using UpSampling with a scaling factor of 2. Then, it applies a 3x3 depthwise convolution (DWConv) followed by batch normalization (BN) and ReLU activation; these steps efficiently enhance the feature map without significantly increasing computational cost. Finally, a 1x1 convolution reduces the number of channels, ensuring the upsampled feature map matches the number of channels in the next stage. EUCB uses depthwise convolution instead of the standard 3x3 convolution for efficient computation, making it particularly suitable for applications with high computational efficiency requirements.

[0127] In the field of computer vision, with the continuous improvement of mobile device computing power and the increasing complexity of visual tasks, the demand for efficient and lightweight neural network modules is also growing. YOLOv11, as an advanced real-time object detection model, still faces the dual challenges of efficiency and accuracy in feature extraction. To address this issue, we introduced the iRMB (Inverted Residual Mobile Block) module into YOLOv11. iRMB combines the advantages of inverted residual structures and attention mechanisms, significantly improving the efficiency and accuracy of feature extraction while maintaining the model's lightweight nature.

[0128] Lightweight Partial Convolution (PConv): After model lightweighting, the overall floating-point operations (FLOPS) of the model are significantly reduced, but the floating-point operations per second are not significantly improved. Therefore, in order to achieve a dynamic balance between operator processing speed and overall model FLOPS, we introduced PConv into the model to reduce computational redundancy and memory access.

[0129] A high-temperature status identification system for tea seedlings based on chlorophyll fluorescence imaging was also disclosed, comprising the following modules:

[0130] The chlorophyll fluorescence parameter acquisition module is used to acquire chlorophyll fluorescence parameters and images of tea seedlings under different high temperature stress conditions;

[0131] The feature parameter determination module is used to perform Spearman rank correlation analysis on various chlorophyll fluorescence parameters to determine Fv / Fm as the key feature parameter;

[0132] The dataset construction module is used to preprocess the chlorophyll fluorescence images corresponding to Fv / Fm to construct a high-temperature stress state dataset;

[0133] An improved YOLOv11 model building module is used to build deep learning models that include the MobileNetV4 backbone network, EUCB module, iRMB module and PConv module;

[0134] The model training and testing module is used to divide the high-temperature stress state dataset into training, validation, and testing sets to train, validate, and test the model.

[0135] The high-temperature stress identification module is used to deploy the trained model and intelligently identify the high-temperature stress status of tea seedlings. The high-temperature stress status includes four levels: normal, mild stress, moderate stress, and severe stress, with corresponding temperature ranges of 25-30℃, 31-35℃, 36-40℃, and 41-45℃, respectively.

[0136] Furthermore, the improved YOLOv11 model includes an input layer, a backbone network layer, a neck network layer, and a detection head. The backbone network layer uses the MobileNetV4 network, and the neck network layer includes the EUCB module, the iRMB module, and the PConv module.

[0137] Furthermore, in the improved YOLOv11 model, the EUCB module is used to maintain the spatial resolution of features while reducing computational cost, the iRMB module is used to improve feature representation through inverse residual structure, and the PConv module is used to reduce computational redundancy and memory access.

[0138] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for identifying the high-temperature status of tea seedlings based on chlorophyll fluorescence imaging, characterized in that, Includes the following steps: Step S1: Obtain chlorophyll fluorescence parameters and images of tea seedlings under different high temperature stress conditions, perform Spearman rank correlation analysis on each parameter, and determine the parameter that is significantly correlated with the stress level as the maximum photosynthetic efficiency Fv / Fm; Step S2: Preprocess the chlorophyll fluorescence images corresponding to Fv / Fm to construct a high-temperature stress state dataset; Step S3: Construct an improved YOLOv11-MEIP model. The improved YOLOv11-MEIP model replaces the backbone network of the original network with the MobileNetV4 network and introduces the EUCB module, iRMB module and PConv module. Step S4: Divide the high temperature stress state dataset into a training set, a validation set, and a test set according to a preset ratio, and train, validate, and test the improved YOLOv11-MEIP model. Step S5: Deploy the tested and approved improved YOLOv11-MEIP model for the identification of high temperature stress in tea seedlings.

2. The method for identifying the high-temperature status of tea seedlings based on chlorophyll fluorescence imaging according to claim 1, characterized in that, Step S1 specifically involves: Thirteen chlorophyll fluorescence parameters of tea seedlings under different high-temperature stress conditions were obtained using a multifunctional plant photosynthetic phenotyping system. The Spearman rank correlation analysis method was used to calculate the correlation coefficient between each chlorophyll fluorescence parameter and the high temperature stress level, and Fv / Fm was determined as the key characteristic parameter.

3. The method for identifying the high-temperature status of tea seedlings based on chlorophyll fluorescence imaging according to claim 1, characterized in that, The preprocessing in step S2 includes: The chlorophyll fluorescence images corresponding to Fv / Fm were converted to grayscale. Histogram equalization is used to enhance image contrast; Perform image normalization and size standardization; Data augmentation techniques, including rotation, translation, flipping, brightness and contrast adjustments, are used to expand the dataset.

4. The method for identifying the high-temperature status of tea seedlings based on chlorophyll fluorescence imaging according to claim 1, characterized in that, In step S3, the improved YOLOv11 model consists of an input layer, a backbone network layer, a neck network layer, and a detection head, wherein: The input layer is used to receive preprocessed chlorophyll fluorescence images; The backbone network layer replaces the original network with MobileNetV4 network to extract multi-scale features of chlorophyll fluorescence images. The neck network layer includes an EUCB module, an iRMB module, and a PConv module, which are used for feature upsampling and fusion; The detection head is used to output the identification results of the high temperature stress state of tea seedlings.

5. The method for identifying the high-temperature status of tea seedlings based on chlorophyll fluorescence imaging according to claim 4, characterized in that, The EUCB module includes upsampling layers and convolutional layers to maintain the spatial resolution of features while reducing computational cost; the iRMB module adopts an inverse residual structure to improve feature representation through an expansion-compression strategy; and the PConv module adopts a partial convolution strategy to reduce computational redundancy and memory access.

6. The method for identifying the high-temperature status of tea seedlings based on chlorophyll fluorescence imaging according to claim 1, characterized in that, Step S4 specifically involves: The high-temperature stress state dataset was divided into a training set, a validation set, and a test set in a ratio of 8:1:

1. The improved YOLOv11 model is trained using the training set until the loss function value is less than a preset threshold. The trained model is validated using the validation set. If the validation accuracy of the model is lower than the preset accuracy threshold, the training set is expanded and training continues. The validated model is tested using the test set, and the model's precision, recall, and average accuracy are calculated. If the test metrics do not meet the preset requirements, the training set is expanded and training continues; if the requirements are met, the model training is complete.

7. A system for identifying the high-temperature status of tea seedlings based on chlorophyll fluorescence imaging, characterized in that: Includes the following modules: The chlorophyll fluorescence parameter acquisition module is used to acquire chlorophyll fluorescence parameters and images of tea seedlings under different high temperature stress conditions; The feature parameter determination module is used to perform Spearman rank correlation analysis on various chlorophyll fluorescence parameters to determine Fv / Fm as the key feature parameter; The dataset construction module is used to preprocess the chlorophyll fluorescence images corresponding to Fv / Fm to construct a high-temperature stress state dataset; An improved YOLOv11 model building module is used to build deep learning models that include the MobileNetV4 backbone network, EUCB module, iRMB module and PConv module; The model training and testing module is used to divide the high-temperature stress state dataset into training, validation, and testing sets to train, validate, and test the model. The high-temperature state recognition module is used to deploy the trained model and intelligently identify the high-temperature stress state of tea seedlings.

8. The tea seedling high-temperature status identification system based on chlorophyll fluorescence imaging according to claim 7, characterized in that, The high-temperature stress state includes four levels: normal state, mild stress, moderate stress, and severe stress, with corresponding temperature ranges of 25-30℃, 31-35℃, 36-40℃, and 41-45℃, respectively.

9. A tea seedling high-temperature status identification system based on chlorophyll fluorescence imaging according to claim 7, characterized in that, The improved YOLOv11 model includes an input layer, a backbone network layer, a neck network layer, and a detection head. The backbone network layer uses the MobileNetV4 network, and the neck network layer includes an EUCB module, an iRMB module, and a PConv module.

10. A tea seedling high-temperature status identification system based on chlorophyll fluorescence imaging according to claim 7, characterized in that, In the improved YOLOv11 model, the EUCB module is used to maintain the spatial resolution of features while reducing computational cost, the iRMB module is used to improve feature representation through inverse residual structure, and the PConv module is used to reduce computational redundancy and memory access.

Images