An intelligent tea bud detection system and management method

Abstract

The present invention relates to the field of precise monitoring and management of grain and cash crops, and in particular to an intelligent tea bud detection system and management method, including: collecting tea tree tea bud image data by using equipment in a tea garden; preprocessing the collected image data; using LabelImg to label tea buds at different growth stages, and constructing a data set for a tea bud detection model; pre-training, formal training and optimization by YOLOv10, and saving the optimal weight file; fixing multiple cameras in the tea garden, regularly taking tea bud images and measuring tea bud length, and using a linear regression model to construct a tea bud length estimation model; and realizing intelligent detection and management of tea buds in tea gardens through a cloud platform. The method of the present invention can improve the speed and accuracy of tea bud quantity detection, realize yield estimation, give farming operation suggestions and coercion warnings, enhance the efficiency of tea garden production management, and inject strong impetus into the intelligent transformation and upgrading of the tea industry.

Description

Technical Field

[0001] The present invention relates to the field of precise monitoring and management of grain and cash crops, and in particular to an intelligent tea bud detection system and management method. Background Art

[0002] Tea is an important economic crop in my country. Indicators such as the growth rate, length, weight, and number and density of tea buds per unit area are closely related to agricultural operations such as judging the time of tea harvesting, estimating yield, and determining the time of pruning and fertilizing. However, the current methods for obtaining or judging these indicators mostly rely on manual measurement or experience, which are time-consuming, labor-intensive, inefficient, and lack of unified standards, making it difficult to meet the needs of large-scale tea garden management. Therefore, there is an urgent need for an intelligent and efficient tea bud detection and management method to improve the efficiency of tea data collection, standardize picking standards, and enhance the effectiveness of tea garden management.

[0003] Although many researchers at home and abroad have chosen to use deep learning models to identify tea buds, the focus of existing research is mostly on the improvement of tea bud recognition algorithms or the development of automatic tea picking equipment, and there are few practical applications involving standardized management of tea gardens. In addition, in the research on tea bud recognition algorithms, most researchers generally classify tea buds of different growth periods and different forms into one category, or only divide them into single buds, one bud and one leaf, one bud and two leaves, etc. Most of these studies focus on tea buds in the harvesting period and can only count the number of tea buds or confirm the picking points, but cannot detect tea buds in other growth stages or forms. Therefore, existing technologies cannot provide richer tea growth and yield information, nor can they guide a wider range of agricultural operations.

[0004] Based on the above background, the present invention creatively proposes an intelligent tea bud detection system and management method. The method constructs an efficient and accurate tea bud recognition model and tea bud length estimation model by extensively collecting and collating tea bud image data from different varieties, different seasons and different growth stages in the same season in the early stage, and deploys these models to the cloud platform. With the help of the management system of the cloud platform, the present invention can realize real-time monitoring of the growth status of tea buds, yield estimation and intelligent decision-making, providing strong data support and technical support for the scientific management of tea gardens. Summary of the invention

[0005] The purpose of the present invention is to solve the problems of low efficiency, lack of unified standards and inability to provide comprehensive growth information in the prior art manual measurement and to propose an intelligent tea bud detection system and management method.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme: an intelligent tea bud detection system and management method, comprising the following steps:

[0007] Step S1, using different visual devices in a tea garden to collect image data of tea buds of different varieties, different seasons, and different growth stages;

[0008] Step S2, pre-processing the collected image data, using an image processing algorithm to delete blurred images and crop images with obvious tea bud features;

[0009] Step S3, using LabelImg to label tea buds at different growth stages and construct a data set for a tea bud detection model;

[0010] Step S4, based on the labeled data set, pre-training, formal training and optimization are performed through YOLOv10 to save the optimal weight file;

[0011] Step S5, fixing multiple cameras in the tea garden, regularly taking images of tea buds and measuring the length of the tea buds, and using a linear regression model to construct a tea bud length estimation model;

[0012] Step S6, integrating the tea bud detection model, tea bud length estimation model, tea bud growth data and yield data through the cloud platform to realize intelligent detection and management of tea buds in the tea garden.

[0013] Furthermore, in step S1, the following sub-steps are also included:

[0014] S1-1, using a fixed camera, a mobile device and an IoT device to collect images in a tea garden, wherein the fixed camera is used for regular automatic photography, the mobile device is used for manual photography, and the IoT device is used for real-time monitoring of environmental data of the tea garden;

[0015] S1-2, setting the shooting angle and frequency of the camera, the shooting angle is that the camera tilts downward to shoot, and the frequency is that the tea garden image is automatically captured and saved every 1 hour;

[0016] S1-3, use a mobile device to take close-up images of tea buds once a week, with random angles and distances, and use an IoT device to record environmental data and time periods when the images are collected. The environmental data include temperature, soil moisture, rainfall, wind speed and direction, light intensity, radiation, and air pressure. The time periods are different time periods in a day from 7:00 to 18:00;

[0017] S1-4, determine the variety, season and growth stage when collecting tea buds from the tea tree, the varieties include Fuding Dabai and Sanhua 1951, the seasons include spring, summer and autumn, and the growth stages include germination period, single bud, one bud and one leaf, one bud and two leaves, and bud stationary period.

[0018] Furthermore, in step S2, the following sub-steps are also included:

[0019] S2-1, using image processing algorithms to delete blurry images without obvious features of tea buds, and retaining images where tea buds can be identified by naked eyes;

[0020] S2-2, use the image processing algorithm to crop out a part of the clear images with obvious tea bud features, name this part of the dataset A, and name the remaining dataset B.

[0021] Furthermore, in step S3, the following sub-steps are also included:

[0022] S3-1, use LabelImg to label tea buds of different categories. According to different growth stages, tea buds are divided into five categories: germination stage, single bud, one bud and one leaf, one bud and two leaves, and stationary bud stage;

[0023] S3-2, when annotating images of the germination stage, single bud, and stationary bud stage, the annotation frame just frames the bottom end of the bud growth to the top end of the bud;

[0024] S3-3, when annotating an image of one bud and one leaf / one bud and two leaves, the annotation frame just frames the lowest growth point of the latest / two unfolded leaves to the top of the bud;

[0025] S3-4, the labeled files and original images of dataset A are saved as a pre-training dataset, and the labeled files and original images of dataset B are divided into a training set, a validation set, and a test set in a ratio of 7:2:1.

[0026] Furthermore, in step S4, the following sub-steps are also included:

[0027] S4-1, based on the preprocessed dataset A, uses four different variant models of the YOLOv10 series model s, m, l, and x for pre-training, and uses mAP50 (mean average precision) and frame rate per second FPS as indicators for evaluating model performance;

[0028] S4-2, after pre-training, select the YOLOv10m model variant with the best performance according to the mAP50 and FPS indicators, and save its weight file (PTA1);

[0029] S4-3, using PTA1 as the initial weight file, formally train the YOLOv10m model based on dataset B, and optimize the model parameters by adjusting the hyperparameters to obtain the optimal weight file (PTB1);

[0030] S4-4, optimize the trained YOLOv10m model and save the optimized optimal weight file (PTBn). The optimization includes lightweight design of the backbone network, improvement of the loss function, and pruning of the model.

[0031] Furthermore, in step S5, the following sub-steps are also included:

[0032] S5-1, fixing multiple cameras in the tea garden, wherein the lowest point height of the camera lens is on the same horizontal line as the initial canopy height of the tea garden, the direction of the camera lens is perpendicular to the growth direction of the tea branches, and the camera only photographs one tea bud;

[0033] S5-2, set each camera to continuously take 100 tea bud images at two different time points every day and save them to form a data set C(i). At the same time, use a ruler to measure the length of the tea buds within the range of the corresponding camera at these time points, and take the average of the three readings as the actual length of the tea buds Y(i);

[0034] S5-3, using the optimized YOLOv10m network model to detect the tea bud image in the data set C(i), obtain the coordinate value of the prediction box, and obtain the pixel height value X(i) of the tea bud;

[0035] S5-4, performing correlation analysis on the actual length Y(i) of the tea buds and the pixel height value X(i) at different distances between the camera and the tea buds, and selecting the distance dj with the greatest correlation;

[0036] S5-5, at the distance dj, a linear regression model is constructed between the actual length Y(i) of the tea bud and the pixel height value X(i), and the formula is as follows:

[0037] Where k and b are the parameters of the linear regression model, which are used to estimate the true length Y(i) of the tea bud based on its pixel height X(i).

[0038] Furthermore, in step S6, the following sub-steps are also included:

[0039] S6-1, integrate tea bud detection model, tea bud length estimation model, tea bud growth data and yield data into the cloud platform to achieve centralized data management and analysis;

[0040] S6-2, performing real-time data processing on the tea bud detection model and the tea bud length estimation model to generate a real-time monitoring report on the growth status of the tea buds;

[0041] S6-3, based on the tea bud growth data and yield data, estimate the fresh weight yield of tea leaves and generate a yield estimation report. The calculation method of the estimated fresh weight yield of tea leaves is as follows:

[0042] Where: P is the estimated yield value, in kg; a is the canopy area of ​​the tea tree captured by the fixed surveillance camera, in m²; Q is the actual planting area of ​​a certain variety of tea trees, in m²; N is the number of tea buds detected in area a, in pieces; W is the average fresh weight of a single bud measured, in kg; for the same variety of tea buds, the W value is different in different seasons;

[0043] S6-4, based on real-time monitoring reports and yield estimation reports, provides suggestions for agricultural operations and stress warnings and forecasts, including monitoring effective accumulated temperature in spring to warn of possible cold damage; monitoring humidity and precipitation in summer to warn of drought; indicating the picking period based on the number of tea buds; and predicting the picking time based on the growth rate of tea buds.

[0044] Furthermore, an intelligent tea bud detection system includes:

[0045] Image acquisition module: including fixed cameras, mobile devices and IoT devices, used to collect image data of tea buds of different varieties, seasons and growth stages in the tea garden;

[0046] Data preprocessing module: used to preprocess the collected image data, including deleting blurred images and cutting out images with obvious tea bud features;

[0047] Labeling module: including LabelImg, which is used to label the preprocessed image data and classify the tea buds into categories at different growth stages;

[0048] Model training module: Based on the annotated image dataset, use the YOLOv10 series model for pre-training, formal training and optimization, and save the optimal weight file;

[0049] Tea bud length estimation module: used to estimate the actual length of tea buds according to the pixel height of tea buds through a linear regression model;

[0050] Cloud platform: used to integrate tea bud detection model, tea bud length estimation model, IoT device data, tea bud growth data and yield data, to achieve real-time monitoring of tea bud growth status in tea gardens, estimation of tea fresh weight yield and intelligent management.

[0051] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0052] The invention adopts different visual devices in a tea garden to collect image data of tea buds of tea trees of different varieties, different seasons and different growth stages; pre-processes the collected image data, uses an image processing algorithm to delete blurred images, and cuts out images with obvious tea bud features; uses LabelImg to label tea buds of different growth stages, and constructs a data set of a tea bud detection model; based on the labeled data set, pre-training, formal training and optimization are performed through YOLOv10 to save the optimal weight file; multiple cameras are fixed in the tea garden, tea bud images are taken regularly and the length of the tea buds is measured, and a tea bud length estimation model is constructed by using a linear regression model; the tea bud detection model, the tea bud length estimation model, the tea bud growth data and the yield data are integrated through a cloud platform to realize intelligent detection and management of tea buds in the tea garden.

[0053] The present invention combines a cloud platform with deep learning technology to achieve rapid and accurate detection of tea buds. It can process a large amount of image data in a short time, automatically identify the number and growth status of tea buds, output the number of tea buds per unit area, and predict the fresh weight yield of tea, providing a scientific basis for tea garden yield estimation.

[0054] The present invention combines deep learning models with Internet of Things technology to monitor the growth dynamics and environmental changes of tea leaves in real time, analyze the growth status of tea buds in real time, and provide intelligent farming operation suggestions, which can help tea garden managers take timely measures to optimize tea garden management and improve tea yield and quality.

[0055] The present invention realizes data sharing and remote management through a cloud platform, which greatly improves the convenience of tea garden management. Real-time data of the tea garden can be accessed anytime and anywhere through the cloud platform. The cloud platform also supports long-term storage and analysis of data. These data can be used for historical comparison and trend analysis, so as to make more accurate management decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0057] Figure 1 A schematic diagram of the structure of an intelligent tea bud detection system and management method provided by an embodiment of the present invention;

[0058] Figure 2 A flow chart of a management method provided by an embodiment of the present invention;

[0059] Figure 3A schematic diagram of tea bud labeling is provided for an embodiment of the present invention;

[0060] Figure 4 A diagram of the system module architecture provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0061] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the intelligent tea bud detection system and management method proposed by the present invention, its specific implementation, structure, features and effects, in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0062] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0063] The following examples are for illustrative purposes only and are not intended to limit the scope of the present invention.

[0064] The specific scheme of an intelligent tea bud detection system and management method provided by the present invention is described in detail below with reference to the accompanying drawings. Example

[0065] See also Figure 1 and Figure 2 , which shows a structural schematic diagram of an intelligent tea bud detection system and management method provided by an embodiment of the present invention and a flow chart of the management method, the method comprising the following steps:

[0066] Step S1, using different visual devices in a tea garden to collect image data of tea buds of different varieties, different seasons, and different growth stages;

[0067] Wherein step S1 also includes the following sub-steps:

[0068] S1-1, using a fixed camera, a mobile device and an IoT device to collect images in a tea garden, wherein the fixed camera is used for regular automatic photography, the mobile device is used for manual photography, and the IoT device is used for real-time monitoring of environmental data of the tea garden;

[0069] S1-2, setting the shooting angle and frequency of the camera, the shooting angle is that the camera tilts downward to shoot, and the frequency is that the tea garden image is automatically captured and saved every 1 hour;

[0070] S1-3, use a mobile device to take close-up images of tea buds once a week, with random angles and distances, and use an IoT device to record environmental data and time periods when the images are collected. The environmental data include temperature, soil moisture, rainfall, wind speed and direction, light intensity, radiation, and air pressure. The time periods are different time periods in a day from 7:00 to 18:00;

[0071] S1-4, determine the variety, season and growth stage when collecting tea buds from the tea tree, the varieties include Fuding Dabai and Sanhua 1951, the seasons include spring, summer and autumn, and the growth stages include germination period, single bud, one bud and one leaf, one bud and two leaves, and bud stationary period.

[0072] See also Figure 3 A schematic diagram of tea bud labeling is provided for an embodiment of the present invention.

[0073] It should be noted that the data collection site is located in a tea garden in a certain place, and the collection time is from March to October 2024. The visual equipment includes fixed cameras (Changhong CH001 dome camera) and mobile devices (different models of mobile phones and tablets of iPhone, Xiaomi, OPPO, and Huawei).

[0074] The specific collection method is as follows: the camera is tilted downward to shoot, and the actual area of ​​the tea garden captured includes different scales of 1-2m² and 5-10m². The tea garden image is automatically captured and saved every 1 hour; close-up images of tea leaves and tea buds are manually taken with a mobile phone every 1-2 weeks, with random angles and distances.

[0075] The two varieties include Fuding Dabai and Sanhua 1951. Fuding Dabai is one of the famous white tea varieties in China, with high economic value and market recognition. Its tea buds are plump, hairy and tender, and it is an important raw material for making high-quality white tea. Choosing Fuding Dabai as the research object can well represent the growth characteristics of white tea trees; Sanhua 1951 is an excellent variety bred through systematic selection, with the characteristics of high yield, high quality and wide adaptability. It is an important achievement in the field of tea breeding and cultivation, suitable for making green tea, black tea, white tea, jasmine tea and other types of tea.

[0076] The three seasons include spring, summer and autumn. The picking time for spring tea is from early March to late April, the picking time for summer tea is from late May to late July, and the picking time for autumn tea is from early September to early October.

[0077] The five growth stages include: budding stage, single bud, one bud and one leaf, one bud and two leaves, and bud station stage. The budding stage refers to the state where the bud is wrapped by the fish leaf, or the bud is slightly longer than the fish leaf; the single bud refers to the state from the fish leaf turning outward to the first leaf wrapping the bud and not unfolding; one bud and one leaf refers to a basically unfolded true leaf separated from the bud and a full bud; one bud and two leaves refers to two basically unfolded true leaves separated from the bud and a full bud; the bud station stage refers to the bud being very small, the bud head is almost invisible, and the true leaf next to the bud is relatively hard.

[0078] The collected tea bud data include the growth stages in each season: spring tea includes images of four stages: germination period, single bud, one bud and one leaf, one bud and two leaves; summer tea includes images of three stages: single bud, one bud and one leaf, one bud and two leaves; autumn tea includes images of four stages: single bud, one bud and one leaf, one bud and two leaves, and bud stage.

[0079] Step S2, pre-processing the collected image data, deleting blurred images, and cutting out images with obvious tea bud features;

[0080] Wherein step S2 also includes the following sub-steps:

[0081] S2-1, using image processing algorithms to delete blurry images without obvious features of tea buds, and retaining images where tea buds can be identified by naked eyes;

[0082] S2-2, use the image processing algorithm to crop out a part of the clear images with obvious tea bud features, name this part of the dataset A, and name the remaining dataset B.

[0083] It should be noted that the preprocessing of image data includes deleting blurred images without obvious features of tea buds to ensure that the tea buds in the image can be identified by the naked eye during labeling. Deleting blurred images can reduce the interference of noise data on model training. Blurred images are caused by poor shooting conditions (insufficient light, camera shake, etc.). These images are difficult to provide clear tea bud features, which is not conducive to model learning. Deleting images without obvious features of tea buds can improve the purity of the data set and ensure that the model can focus on learning the features of tea buds rather than other irrelevant background information.

[0084] The image processing algorithm automatically screens as follows:

[0085] import cv2

[0086] import numpy as np

[0087] def is_blurry(image, threshold=50):

[0088] gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

[0089] laplacian = cv2.Laplacian(gray, cv2.CV_64F)

[0090] variance = laplacian.var()

[0091] return variance <threshold

[0092] # Example: Filtering Blurry Images

[0093] images = [...] # List of images

[0094] clear_images = [img for img in images if not is_blurry(img)]

[0095] Cropping out clear images with obvious tea bud features can further improve the quality of the dataset. By cropping, irrelevant background information in the image can be removed, allowing the model to focus more on the features of the tea buds.

[0096] The cropped images are named Dataset A and used for model pre-training to improve the initialization effect of the model. The remaining images are named Dataset B and used for formal training and testing to ensure that the model is trained and verified on a wider range of data.

[0097] The image processing algorithm automatically crops as follows:

[0098] import cv2

[0099] import numpy as np

[0100] def crop_tea_buds(image, model):

[0101] results = model(image)

[0102] tea_buds = results.xyxy[0] # Assume the model returns the bounding box of the tea buds

[0103] cropped_images = []

[0104] for bud in tea_buds:

[0105] x1, y1, x2, y2 = map(int, bud[:4])

[0106] cropped_image = image[y1:y2, x1:x2]

[0107] cropped_images.append(cropped_image)

[0108] return cropped_images

[0109] # Example: Cropping tea bud image

[0110] model = load_model('yolov10_tea_buds.pth') # Load pre-trained model

[0111] clear_images = [...] # Clear image list

[0112] cropped_images = []

[0113] for img in clear_images:

[0114] cropped_images.extend(crop_tea_buds(img, model))

[0115] Step S3, using LabelImg to label tea buds at different growth stages and construct a data set for a tea bud detection model;

[0116] Wherein step S3 also includes the following sub-steps:

[0117] S3-1, use LabelImg to label tea buds of different categories. According to different growth stages, tea buds are divided into five categories: germination stage, single bud, one bud and one leaf, one bud and two leaves, and stationary bud stage;

[0118] S3-2, when annotating images of the germination stage, single bud, and stationary bud stage, the annotation frame just frames the bottom end of the bud growth to the top end of the bud;

[0119] S3-3, when annotating an image of one bud and one leaf / one bud and two leaves, the annotation frame just frames the lowest growth point of the latest / two unfolded leaves to the top of the bud;

[0120] S3-4, the labeled files and original images of dataset A are saved as a pre-training dataset, and the labeled files and original images of dataset B are divided into a training set, a validation set, and a test set in a ratio of 7:2:1.

[0121] It should be noted that LabelImg is an open source graphical interface tool based on Python and Qt, which is used to annotate target objects in images. Through LabelImg, objects in images can be easily annotated and annotation files can be generated (PASCAL VOC format, YOLO format). The annotation files can be used to train deep learning models.

[0122] Specific operations of using LabelImg to label different types of tea buds:

[0123] 1. Open the LabelImg tool and load the image to be labeled.

[0124] 2. Use a rectangular bounding box to frame the tea buds.

[0125] 3. Select the corresponding category (germination stage, single bud, one bud and one leaf, one bud and two leaves, bud station stage).

[0126] 4. Save the annotation results and generate an annotation file (XML file).

[0127] The labeled files and original images of dataset A are saved as pre-training datasets. The labeled files and original images of dataset B are divided into training set, validation set and test set in a one-to-one correspondence with a ratio of 7:2:1. The specific operations are as follows:

[0128] import os

[0129] import random

[0130] from sklearn.model_selection import train_test_split

[0131] def split_dataset(data_dir, train_ratio=0.7, val_ratio=0.2, test_ratio=0.1):

[0132] images = [os.path.join(data_dir, f) for f in os.listdir(data_dir)if f.endswith('.jpg

[0133] Step S4: Based on the labeled data set, pre-training, formal training and optimization are performed through YOLOv10 to save the optimal weight file;

[0134] Wherein step S4 also includes the following sub-steps:

[0135] S4-1, based on the preprocessed dataset A, uses four different variant models of the YOLOv10 series model s, m, l, and x for pre-training, and uses mAP50 (mean average precision) and frame rate per second FPS as indicators for evaluating model performance;

[0136] S4-2, after pre-training, select the YOLOv10m model variant with the best performance according to the mAP50 and FPS indicators, and save its weight file (PTA1);

[0137] S4-3, using PTA1 as the initial weight file, formally train the YOLOv10m model based on dataset B, and optimize the model parameters by adjusting the hyperparameters to obtain the optimal weight file (PTB1);

[0138] S4-4, optimize the trained YOLOv10m model and save the optimized optimal weight file (PTBn). The optimization includes lightweight design of the backbone network, improvement of the loss function, and pruning of the model.

[0139] It should be noted that the YOLO (You Only Look Once) series is a popular single-stage target detection algorithm, known for its high efficiency and accuracy. YOLOv10 is the latest version of the series, providing four variant models of different sizes to meet different application scenarios and performance requirements.

[0140] The four variant models s, m, l, and x are: YOLOv10-s (Small), YOLOv10-m (Medium), YOLOv10-l (Large), and YOLOv10-x (Extra Large).

[0141] YOLOv10-s (Small): The model is small and has low computational complexity. It is suitable for running on resource-constrained devices and for scenarios that require fast detection but do not require high accuracy. It usually has a high frame rate (FPS) but a relatively low detection accuracy (mAP).

[0142] YOLOv10-m (Medium): The model size is moderate, balancing computational complexity and detection accuracy. It is suitable for most practical applications and scenarios that require a balance between accuracy and speed. It achieves a good balance between detection accuracy (mAP) and frame rate (FPS).

[0143] YOLOv10-l (Large): The model is larger and the computational complexity is higher, but the detection accuracy is higher. It is suitable for scenarios with high requirements for detection accuracy. It is suitable for scenarios that require high-precision detection. The detection accuracy (mAP) is higher, but the frame rate (FPS) is relatively low.

[0144] YOLOv10-x (Extra Large): This model has the largest model, the highest computational complexity, and the highest detection accuracy, but also the highest hardware requirements. It is suitable for scenarios with extremely high detection accuracy requirements. It has the highest detection accuracy (mAP) but the lowest frame rate (FPS), and is suitable for running on high-performance computing devices.

[0145] mAP (mean Average Precision) is a commonly used performance evaluation indicator in target detection tasks. It is used to measure the detection accuracy of the model. mAP50 is a special case of mAP, which represents the average precision when the IoU (Intersection over Union) threshold is 0.5. The higher the mAP50, the better the detection accuracy of the model.

[0146] FPS (Frames Per Second) is an indicator to measure the real-time performance of the model. FPS = 1 / single-frame processing time, measured in frames per second. It indicates the number of image frames that the model can process per second. The higher the FPS, the better the real-time performance of the model.

[0147] The optimization includes lightweight design of the backbone network: adding a new deep convolution module SDCB (Spindle Depthwise Convolution Block) to the YOLOv10m backbone network to replace the CIB module in C2fCIB, improving the feature extraction capability of the backbone network, and reducing the amount of calculation and parameters of the entire network without reducing accuracy, thereby improving the efficiency of the YOLOv10m network.

[0148] Improved loss function: EMASlideLoss is used to replace the original category classification loss function of YOLOv10, which solves the problem of sample imbalance and makes the model pay more attention to difficult-to-classify and misclassified samples.

[0149] Prune the model: Use the layer-adaptive amplitude pruning (LAMP) strategy to optimize the efficiency and performance of the model, reduce unnecessary parameters and connections in the neural network, reduce the storage requirements and computational complexity of the YOLOv10m network, and do not reduce the detection accuracy.

[0150] Step S5, fixing multiple cameras in the tea garden, regularly taking images of tea buds and measuring the length of the tea buds, and using a linear regression model to construct a tea bud length estimation model;

[0151] Wherein, in step S5, the following sub-steps are also included:

[0152] S5-1, fixing multiple cameras in the tea garden, wherein the lowest point height of the camera lens is on the same horizontal line as the initial canopy height of the tea garden, the direction of the camera lens is perpendicular to the growth direction of the tea branches, and the camera only photographs one tea bud;

[0153] S5-2, set each camera to continuously take 100 tea bud images at two different time points every day and save them to form a data set C(i). At the same time, use a ruler to measure the length of the tea buds within the range of the corresponding camera at these time points, and take the average of the three readings as the actual length of the tea buds Y(i);

[0154] S5-3, using the optimized YOLOv10m network model to detect the tea bud image in the data set C(i), obtain the coordinate value of the prediction box, and obtain the pixel height value X(i) of the tea bud;

[0155] S5-4, performing correlation analysis on the actual length Y(i) of the tea buds and the pixel height value X(i) at different distances between the camera and the tea buds, and selecting the distance dj with the greatest correlation;

[0156] S5-5, at the distance dj, a linear regression model is constructed between the actual length Y(i) of the tea bud and the pixel height value X(i), and the formula is as follows:

[0157] Where k and b are the parameters of the linear regression model, which are used to estimate the true length Y(i) of the tea bud based on its pixel height X(i).

[0158] It should be noted that the camera is installed at the same level as the initial canopy height of the tea garden, ensuring that the captured image can cover the entire picture of the tea buds.

[0159] Camera direction: The camera lens direction is perpendicular to the growth direction of the tea branches, ensuring that the shape of the tea buds is clearly visible in the captured image.

[0160] Camera coverage: Each camera contains only one tea bud, avoiding mutual interference between multiple tea buds and improving detection accuracy.

[0161] Shooting time: Select two different time points every day (morning and afternoon) and take 100 consecutive images of tea buds at each time.

[0162] Data saving: Save the captured images as data set C(i), where i represents the camera number.

[0163] Real length measurement: While taking images, use a ruler to measure the length of the tea buds within the range of the corresponding camera, and take the average of three readings as the real length Y(i) of the tea buds; the bud length in the germination stage, single bud, and stationary bud stage is the length from the bottom of the bud to the tip of the bud, and the length of a bud and a leaf bud is the length from the top of the growth part of the second unfolded leaf adjacent to the bud to the tip of the bud.

[0164] The implementation method of detecting the tea bud images in the dataset C(i) using the optimized YOLOv10m network model includes:

[0165] Model loading: Load the optimized YOLOv10m model weight file (PTBn).

[0166] Image detection: Detect each tea bud image in the dataset C(i) and obtain the coordinate value of the prediction box.

[0167] Pixel height calculation: Calculate the pixel height value X(i) of the tea bud according to the coordinate value of the prediction box.

[0168] Correlation analysis: Calculate the correlation coefficient between the actual length Y(i) of the tea bud and the pixel height value X(i) at different distances between the camera and the tea bud. dj is the distance with the largest correlation.

[0169] The parameters k and b are estimated by the least squares method, so that the model can accurately estimate the true length Y(i) based on the pixel height X(i).

[0170] Step S6: Integrate the tea bud detection model, tea bud length estimation model, tea bud growth data and yield data through the cloud platform to realize intelligent detection and management of tea buds in the tea garden;

[0171] Wherein, in step S6, the following sub-steps are also included:

[0172] S6-1, integrate tea bud detection model, tea bud length estimation model, tea bud growth data and yield data into the cloud platform to achieve centralized data management and analysis;

[0173] S6-2, performing real-time data processing on the tea bud detection model and the tea bud length estimation model to generate a real-time monitoring report on the growth status of the tea buds;

[0174] S6-3, based on the tea bud growth data and yield data, estimate the fresh weight yield of tea leaves and generate a yield estimation report. The calculation method of the estimated fresh weight yield of tea leaves is as follows:

[0175] Where: P is the estimated yield value, in kg; a is the canopy area of ​​the tea tree captured by the fixed surveillance camera, in m²; Q is the actual planting area of ​​a certain variety of tea trees, in m²; N is the number of tea buds detected in area a, in pieces; W is the average fresh weight of a single bud measured, in kg; for the same variety of tea buds, the W value is different in different seasons;

[0176] S6-4, based on real-time monitoring reports and yield estimation reports, provides suggestions for agricultural operations and stress warnings and forecasts, including monitoring effective accumulated temperature in spring to warn of possible cold damage; monitoring humidity and precipitation in summer to warn of drought; indicating the picking period based on the number of tea buds; and predicting the picking time based on the growth rate of tea buds.

[0177] It should be noted that the agricultural operation guidance suggestions and early warning forecasts include:

[0178] 1. In spring, an effective accumulated temperature of >10℃, about 45-55℃, is required from the budding of tea buds to the formation of one bud and one leaf. The tea bud detection model is used to determine the beginning of the budding period and whether it has entered the one bud and one leaf period. If the monitored effective accumulated temperature is low, the system can issue a possible cold damage warning.

[0179] 2. If the humidity is lower than 50% in summer, the growth of tea trees will be inhibited; if the monthly precipitation is lower than 50 mm for several consecutive months, the system will issue a warning of possible drought and recommend irrigation; if the maximum temperature is above 35°C and the humidity is below 60% for 8-10 consecutive days, a warning of possible heat and drought damage will be issued.

[0180] 3. Through the tea bud detection model, according to the previous measurement area, the number of buds within one square meter is calculated. If it contains 10-15 buds, it means that the harvesting period has begun. The system prompts that no other farming operations should be performed during the picking period.

[0181] 4. Based on the tea bud length estimation model, calculate the tea bud growth rate, predict when the tea bud will enter a certain growth stage, and give suggestions on when to pick the tea with higher yield. You can also manually enter the tea bud length.

[0182] 5. If the tea bud detection system detects a certain number of tea buds in the budding stage, the system will prompt that the tea trees need to be pruned and fertilized.

[0183] 6. After the autumn tea enters the bud-keeping period, the tea trees stop growing, that is, they no longer produce new buds. The system prompts that shallow pruning, application of base fertilizer, and application of garden-closing medicine are required.

[0184] See also Figure 4 , which shows a system module architecture diagram of an intelligent tea bud detection system provided by an embodiment of the present invention, the system comprising:

[0185] Image acquisition module: including fixed cameras, mobile devices and IoT devices, used to collect image data of tea buds of different varieties, seasons and growth stages in the tea garden;

[0186] It should be noted that the fixed camera is used to regularly capture images of tea buds at a fixed position in the tea garden;

[0187] Mobile devices: including mobile phones and tablets of different brands and models, were used to manually capture close-up images of tea buds.

[0188] IoT devices: used to monitor environmental data of the tea garden in real time, including temperature, soil moisture, rainfall, wind speed and direction, light intensity, radiation and air pressure.

[0189] Data preprocessing module: used to preprocess the collected image data, including deleting blurred images and cutting out images with obvious tea bud features;

[0190] It should be noted that the data preprocessing module includes a data set division unit, which is used to divide the preprocessed image data into a pre-training data set and a formal training data set;

[0191] Labeling module: including LabelImg, which is used to label the preprocessed image data and classify the tea buds into categories at different growth stages;

[0192] It should be noted that the labeling module includes a labeling unit: the LabelImg tool is used to label tea buds at different growth stages. The labeling categories include germination period, single bud, one bud and one leaf, one bud and two leaves, and bud period.

[0193] Model training module: Based on the annotated image dataset, use the YOLOv10 series model for pre-training, formal training and optimization, and save the optimal weight file;

[0194] It should be noted that the model training module includes a pre-training unit, a formal training unit and an optimization unit.

[0195] Pre-training unit: Based on the pre-training dataset, use the four variant models of YOLOv10, s, m, l, and x, for pre-training, select the YOLOv10m model with the best performance and save the pre-training weight file; Formal training unit: Based on the formal training dataset and pre-training weight file, use the YOLOv10m model for formal training to obtain the optimal weight file; Optimization unit: Optimize the YOLOv10m model, including lightweight design of the backbone network, improvement of the loss function, use of pruning strategies, and finally save the optimized optimal weight file.

[0196] Tea bud length estimation module: used to estimate the actual length of tea buds according to the pixel height of tea buds through a linear regression model;

[0197] It should be noted that the tea bud length estimation module includes an image capturing unit, a length measuring unit, and a linear regression modeling unit.

[0198] Image capture unit: multiple cameras are fixed in the tea garden to regularly capture and save images of tea buds; length measurement unit: a ruler is used to measure the actual length of the tea buds within the range of the corresponding camera; linear regression modeling unit: based on the correlation between the pixel height and the actual length of the tea buds, a linear regression model is constructed to estimate the length of the tea buds.

[0199] Cloud platform: used to integrate tea bud detection model, tea bud length estimation model, IoT device data, tea bud growth data and yield data, to achieve real-time monitoring of tea bud growth status in tea gardens, estimation of tea fresh weight yield and intelligent management.

[0200] It should be noted that the cloud platform includes a data storage unit, a data processing unit, a management unit, and a user interface.

[0201] Data storage unit: used to store collected image data, tea bud growth data, yield data and IoT device data; data processing unit: used to process and analyze the output results of the tea bud detection model, tea bud length estimation results and IoT device data; management unit: used to provide tea fresh weight yield estimation values, agricultural operation suggestions and stress warning forecasts; user interface: used to display tea bud growth status, yield estimation results, agricultural operation suggestions and warning information, and support users to perform remote management and data query.

[0202] In this way, an intelligent tea bud detection system and management method can realize real-time capture of tea bud growth images in tea gardens, accurate detection of growth status, estimation of fresh weight yield of tea leaves and intelligent management of tea gardens.

[0203] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. An intelligent tea bud detection and management method, characterized in that: The method includes: Step S1, using different visual devices in a tea garden to collect image data of tea buds of different varieties, different seasons, and different growth stages; Step S2, pre-processing the collected image data, using an image processing algorithm to delete blurred images and crop images with obvious tea bud features; Step S3, using Label Img to label tea buds at different growth stages and construct a data set for a tea bud detection model; Step S4, based on the labeled data set, pre-training, formal training and optimization are performed through YOLOv10 to save the optimal weight file; Step S5, fixing multiple cameras in the tea garden, regularly taking images of tea buds and measuring the length of the tea buds, and using a linear regression model to construct a tea bud length estimation model; Step S6, integrating the tea bud detection model, the tea bud length estimation model, the tea bud growth data and the yield data through the cloud platform to realize intelligent detection and management of tea buds in the tea garden; Wherein step S4 also includes the following sub-steps: S4-1, based on the preprocessed dataset A, uses four different variant models of the YOLOv10 series model s, m, l, and x for pre-training, and uses the mean average precision mAP50 and frame rate per second FPS as indicators for evaluating model performance; S4-2, after pre-training, select the YOLOv10m model variant with the best performance according to the mAP50 and FPS indicators, and save its weight file PTA1; S4-3, using PTA1 as the initial weight file, formally train the YOLOv10m model based on data set B, and optimize the model parameters by adjusting the hyperparameters to obtain the optimal weight file PTB1; S4-4, optimize the trained YOLOv10m model and save the optimized optimal weight file PTBn. The optimization includes lightweight design of the backbone network, improvement of the loss function, and pruning of the model.

2. An intelligent tea bud detection and management method as claimed in claim 1, characterized in that: Wherein step S1 also includes the following sub-steps: S1-1, using a fixed camera, a mobile device and an IoT device to collect images in a tea garden, wherein the fixed camera is used for regular automatic photography, the mobile device is used for manual photography, and the IoT device is used for real-time monitoring of environmental data of the tea garden; S1-2, setting the shooting angle and frequency of the camera, the shooting angle is that the camera tilts downward to shoot, and the frequency is that the tea garden image is automatically captured and saved every 1 hour; S1-3, use a mobile device to take close-up images of tea buds once a week, with random angles and distances, and use an IoT device to record environmental data and time periods when the images are collected. The environmental data include temperature, soil moisture, rainfall, wind speed and direction, light intensity, radiation, and air pressure. The time periods are different time periods in a day from "7:00 to 18:00"; S1-4, determine the variety, season and growth stage when collecting tea buds from the tea tree, the varieties include Fuding Dabai and Sanhua 1951, the seasons include spring, summer and autumn, and the growth stages include germination period, single bud, one bud and one leaf, one bud and two leaves, and bud stationary period.

3. The intelligent tea bud detection and management method according to claim 1, characterized in that: Wherein step S2 also includes the following sub-steps: S2-1, using image processing algorithms to delete blurry images without obvious features of tea buds, and retaining images where tea buds can be identified by naked eyes; S2-2, use the image processing algorithm to crop out a part of the clear images with obvious tea bud features, name this part of the dataset A, and name the remaining dataset B.

4. The intelligent tea bud detection and management method according to claim 1, characterized in that: Wherein step S3 also includes the following sub-steps: S3-1, use LabelImg to label tea buds of different categories. According to different growth stages, tea buds are divided into five categories: germination stage, single bud, one bud and one leaf, one bud and two leaves, and stationary bud stage; S3-2, when annotating images of the germination stage, single bud, and stationary bud stage, the annotation frame just frames the bottom end of the bud growth to the top end of the bud; S3-3, when annotating an image of one bud and one leaf / one bud and two leaves, the annotation frame just frames the lowest growth point of the latest / two unfolded leaves to the top of the bud; S3-4, the labeled files and original images of dataset A are saved as a pre-training dataset, and the labeled files and original images of dataset B are divided into a training set, a validation set, and a test set in a ratio of 7:2:

1.

5. The intelligent tea bud detection and management method according to claim 1, characterized in that: Wherein, in step S5, the following sub-steps are also included: S5-1, fixing multiple cameras in the tea garden, wherein the lowest point height of the camera lens is on the same horizontal line as the initial canopy height of the tea garden, the direction of the camera lens is perpendicular to the growth direction of the tea branches, and the camera only photographs one tea bud; S5-2, set each camera to continuously take 100 tea bud images at two different time points every day and save them to form a data set C(i). At the same time, use a ruler to measure the length of the tea buds within the range of the corresponding camera at these time points, and take the average of the three readings as the actual length of the tea buds Y(i); S5-3, using the optimized YOLOv10m network model to detect the tea bud image in the data set C(i), obtain the coordinate value of the prediction box, and obtain the pixel height value X(i) of the tea bud; S5-4, performing correlation analysis on the actual length Y(i) of the tea buds and the pixel height value X(i) at different distances between the camera and the tea buds, and selecting the distance dj with the greatest correlation; S5-5, at the distance dj, a linear regression model is constructed between the actual length Y(i) of the tea bud and the pixel height value X(i), and the formula is as follows: Y(i)=k*X(i)+b in, k and b are the parameters of the linear regression model, which are used to estimate the true length Y(i) of the tea bud based on its pixel height X(i).

6. An intelligent tea bud detection and management method as claimed in claim 1, characterized in that: Wherein, in step S6, the following sub-steps are also included: S6-1, integrate tea bud detection model, tea bud length estimation model, tea bud growth data and yield data into the cloud platform to achieve centralized data management and analysis; S6-2, performing real-time data processing on the tea bud detection model and the tea bud length estimation model to generate a real-time monitoring report on the growth status of the tea buds; S6-3, based on the tea bud growth data and yield data, estimate the fresh weight yield of tea leaves and generate a yield estimation report. The calculation method of the estimated fresh weight yield of tea leaves is as follows: Where: P is the estimated yield value, in kg; a is the area of ​​the tea tree canopy captured by the fixed surveillance camera, in m 2 ; Q is the actual planting area of ​​a certain variety of tea trees, in m 2 ; N is the number of tea buds detected in area a, in pieces; W is the average fresh weight of a single bud measured, in kg; for the same variety of tea buds, the W value is different in different seasons; S6-4, based on real-time monitoring reports and yield estimation reports, provides suggestions for agricultural operations and stress warnings and forecasts, including monitoring effective accumulated temperature in spring to warn of possible cold damage; monitoring humidity and precipitation in summer to warn of drought; indicating the picking period based on the number of tea buds; and predicting the picking time based on the growth rate of tea buds.

7. A system for implementing the intelligent tea bud detection and management method as claimed in claim 1, characterized in that: include: Image acquisition module: including fixed cameras, mobile devices and IoT devices, used to collect image data of tea buds of different varieties, seasons and growth stages in the tea garden; Data preprocessing module: used to preprocess the collected image data, including deleting blurred images and cutting out images with obvious tea bud features; Labeling module: including Label Img, which is used to label the preprocessed image data and classify the tea buds into categories at different growth stages; Model training module: Based on the annotated image dataset, use the YOLOv10 series model for pre-training, formal training and optimization, and save the optimal weight file; Tea bud length estimation module: used to estimate the actual length of tea buds according to the pixel height of tea buds through a linear regression model; Cloud platform: used to integrate tea bud detection model, tea bud length estimation model, IoT device data, tea bud growth data and yield data, to achieve real-time monitoring of tea bud growth status in tea gardens, estimation of tea fresh weight yield and intelligent management.

Images