Garden plant growth intelligent monitoring system based on visual identification

Abstract

The invention relates to the technical field of image recognition and garden management, in particular to a garden plant growth intelligent monitoring system based on visual recognition. By integrating visible light, multispectral and depth imaging equipment and combining an environment sensor, omnibearing data acquisition of plants is realized. After the collected data is subjected to registration and enhancement preprocessing, macro, medium and micro features are extracted by an improved deep learning model, and growth stages and diseases are identified. The decision-making unit fuses the visual features and the environmental parameters, constructs a dynamic association model and generates a personalized regulation and control strategy based on a self-learning mechanism, and the personalized regulation and control strategy is implemented by the execution unit. And the analysis unit judges the growth state by calculating weighted health parameters of indexes such as form, canopy and leaf color, and adaptively adjusts a construction period of a correlation model and an updating period of a dynamic threshold according to the growth state, so that a continuously optimized, accurate and efficient closed-loop intelligent maintenance system is formed. According to the invention, the identification precision of garden monitoring is improved.

Description

Technical Field

[0001] This invention relates to the fields of image recognition and garden management technology, and in particular to an intelligent monitoring system for garden plant growth based on visual recognition. Background Technology

[0002] Monitoring the growth of garden plants is a core aspect of garden maintenance and management. Traditional monitoring methods mainly rely on manual inspections, judging growth status by visually observing plant morphology and leaf condition. This approach suffers from low efficiency, high subjectivity, high rates of missed or false diagnoses, and the inability to monitor in real time. Existing technologies often employ single visual modalities or simple image processing algorithms, only capable of basic functions such as plant height measurement and obvious disease identification, failing to simultaneously monitor multiple growth parameters. Furthermore, most systems lack linkage analysis with environmental data, failing to form a closed loop of "monitoring-decision-execution," resulting in insufficiently targeted control strategies and an inability to meet the needs of refined garden management.

[0003] Furthermore, existing visual recognition models mostly employ general CNN architectures, which are not optimized for the complex growth environment of garden plants, resulting in shortcomings such as insufficient feature extraction and recognition accuracy being greatly affected by environmental interference. Meanwhile, monitoring equipment is mostly fixedly installed, with limited coverage, failing to meet the comprehensive monitoring needs of large-scale gardens. Therefore, there is an urgent need for an intelligent monitoring system that integrates multimodal visual perception, high-precision recognition models, and closed-loop control to address the deficiencies of existing technologies.

[0004] Chinese Patent Publication No. CN117496196A discloses a method, system, equipment, and medium for monitoring the maintenance of garden plants. This invention acquires image monitoring data and environmental parameter monitoring data of a designated monitoring area; then, based on the image monitoring data, it obtains plant area image data and corresponding plant category information; subsequently, based on the plant category information, it performs growth status identification processing on the plant area image data to obtain plant growth status information corresponding to the plant area image data; finally, it acquires target environmental indicator data corresponding to the plant category information and the plant growth status information, and obtains environmental indicator adjustment data based on the environmental parameter monitoring data and the target environmental indicator data.

[0005] It is evident that existing technologies have the following problems: traditional garden monitoring has insufficient accuracy in identifying different types and growth stages of plants in complex garden environments. Summary of the Invention

[0006] To address this issue, the present invention provides a visual recognition-based intelligent monitoring system for the growth of garden plants, which overcomes the problem of insufficient recognition accuracy in traditional garden monitoring for different types and growth stages of plants in complex garden environments.

[0007] To achieve the above objectives, the present invention provides a visual recognition-based intelligent monitoring system for the growth of garden plants, comprising: The multimodal data acquisition unit includes at least a visible light camera and a multispectral imager capable of periodically acquiring data, used to acquire visible light images, multispectral images, and depth images of garden plants; A data preprocessing unit, connected to the multimodal data acquisition unit, is used to perform image registration and data enhancement on the acquired multimodal images. A growth status identification unit, which is connected to the data preprocessing unit, is used to extract macroscopic, mesoscopic and microscopic features of plants from the preprocessed image based on image recognition technology and an improved deep learning model, as well as to identify the plant growth stage, disease type and growth rate. An environmental sensing unit is used to acquire ambient temperature and humidity, soil moisture, and light intensity based on multiple sensors. The decision-making unit is connected to the growth state recognition unit and the environment sensing unit respectively, and is used to periodically construct a visual feature-environment parameter association model and generate personalized control strategies based on a dynamic threshold self-learning mechanism. An execution unit, connected to the decision-making unit, is used to execute the generated personalized control strategy; An analysis unit, connected to the execution unit, is used to calculate visual health parameters, determine the monitoring status of garden plant growth based on these parameters, adjust the construction cycle of the visual feature-environment parameter correlation model based on the monitoring status, and adjust the adjustment cycle of the threshold based on the adjusted monitoring status. Among them, the visual health parameter is the weighted result of morphological qualification rate, canopy coverage rate and leaf color index.

[0008] Furthermore, the improved deep learning model in the growth state recognition unit includes an input layer for multimodal image channel fusion, a feature extraction layer with convolutional blocks and residual connections, an attention mechanism layer with parallel weighting of spatial attention and channel attention, a feature fusion layer that splices macro-meso-micro features, and an output layer that outputs at least the disease type, plant height and crown width, and growth rate.

[0009] Furthermore, the dynamic threshold self-learning mechanism in the decision-making unit includes constructing a sample library based on historical monitoring data, training a threshold prediction model based on the sample library using a gradient boosting tree algorithm, and periodically adjusting the threshold according to the information identified by the growth state identification unit.

[0010] Furthermore, the analysis unit is also used to calculate the difference between a preset value and the visual performance health parameter when the monitoring status is unqualified; the analysis unit is also used to adjust the construction cycle of the visual feature-environment parameter association model based on the ratio of the difference to the preset difference when the difference is greater than the preset difference; wherein, the monitoring status is determined to be unqualified when the preset value is greater than the visual performance health parameter.

[0011] Furthermore, the analysis unit is also used to reduce the construction cycle of the visual feature-environment parameter association model based on the ratio of the difference to a preset difference, and the reduction in the construction cycle of the visual feature-environment parameter association model is proportional to the ratio.

[0012] Furthermore, the analysis unit is also used to calculate the variance of the visual health parameters of all individuals of the same species and under the same care conditions within the monitoring area when the monitoring status is unqualified after adjusting the construction period of the visual feature-environment parameter association model; the analysis unit is also used to adjust the adjustment period of the threshold based on the difference between the variance and the preset variance when the variance is greater than the preset variance.

[0013] Furthermore, the analysis unit is also used to reduce the adjustment period of the threshold based on the difference between the variance and the preset variance, and the reduction in the adjustment period of the threshold is proportional to the difference.

[0014] Furthermore, the analysis unit is also used to, if the monitoring status is unqualified after adjusting the adjustment period of the threshold, repeatedly adjust the adjustment period of the threshold at least once until the number of adjustments is less than a preset number and the monitoring status is qualified, or the number of adjustments is equal to the preset number and the adjustment stops; the analysis unit is also used to, if the monitoring status is unqualified after the adjustment stops, plot a time-visual performance health parameter curve based on the acquired historical visual performance health parameters; the analysis unit is also used to calculate the integral of the time-visual performance health parameter curve, and if the integral is less than a preset integral, adjust the period of sensor data acquisition based on the ratio of the integral to the preset integral.

[0015] Furthermore, the analysis unit is also used to reduce the period of sensor data acquisition based on the ratio of the integral to a preset integral, and the reduction in the period of sensor data acquisition is inversely proportional to the ratio.

[0016] Furthermore, the analysis unit is also used to calculate the average pixel offset of matching feature point pairs between two images at the same timestamp when the monitoring status is unqualified after adjusting the period of sensor data acquisition; the analysis unit is also used to issue a notification of image registration failure when the average pixel offset is greater than a preset tolerance threshold.

[0017] Compared with existing technologies, the advantages of this invention lie in its ability to achieve comprehensive data acquisition of plants by integrating visible light, multispectral, and depth imaging devices with environmental sensors. After registration and enhancement preprocessing, the acquired data is used by an improved deep learning model to extract macro, meso, and micro features and identify growth stages and diseases. The decision-making unit integrates visual features and environmental parameters to construct a dynamic correlation model and generates personalized control strategies based on a self-learning mechanism, which are then implemented by the execution unit. The analysis unit determines the growth status by calculating weighted health parameters of indicators such as morphology, canopy, and leaf color, and adaptively adjusts the construction cycle of the correlation model and the update cycle of the dynamic threshold accordingly, thus forming a continuously optimized, precise, and efficient closed-loop intelligent maintenance system. This invention improves the identification accuracy of garden monitoring.

[0018] Furthermore, this invention extracts comprehensive plant features based on an improved deep learning model, thereby extracting plant features more accurately and further improving the identification accuracy of garden monitoring.

[0019] Furthermore, the present invention, based on a dynamic threshold self-learning mechanism, can more accurately determine the threshold, thereby better adapting to the differences in different plant varieties and growth stages, and further improving the identification accuracy of garden monitoring.

[0020] Furthermore, the present invention determines the cause of the monitoring status failure based on the difference between the preset value and the visual health parameter, which can more accurately determine the cause, thereby enabling more precise adjustments in the future, and further improving the recognition accuracy of garden monitoring.

[0021] Furthermore, this invention reduces the construction cycle of the visual feature-environment parameter association model based on the ratio of the difference to the preset difference, which makes the original model association rules more effective when plant varieties are changed or climate zones change significantly, thereby further improving the identification accuracy of garden monitoring.

[0022] Furthermore, this invention determines the reasons for unqualified monitoring status based on the variance of the visual health parameters of all individuals of the same species under the same maintenance conditions within the monitoring area. This allows for more accurate identification of the causes, enabling more effective adjustments to be made subsequently, thereby further improving the identification accuracy of garden monitoring.

[0023] Furthermore, the present invention reduces the threshold adjustment period based on the difference between the variance and the preset variance, which makes the threshold more accurate, thereby making the generation of personalized control strategies more accurate, and further improving the identification accuracy of garden monitoring.

[0024] Furthermore, this invention determines the cause of unqualified monitoring status based on the integral of the plotted time-visual performance health parameter curve, which can more accurately determine the cause of unqualified monitoring status, thereby enabling more effective adjustments to be made subsequently based on the cause of unqualified status, and further improving the identification accuracy of garden monitoring.

[0025] Furthermore, this invention reduces the data acquisition cycle of the sensor based on the ratio of the integral to the preset integral, enabling the acquisition of more accurate data based on more effective acquisition timing of the sensor, thereby further improving the identification accuracy of garden monitoring.

[0026] Furthermore, this invention determines the cause of unqualified monitoring status based on the average pixel offset of matching feature point pairs between two images at the same timestamp, which can more accurately determine the cause and make subsequent adjustments more effective, thereby further improving the recognition accuracy of garden monitoring. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the structure of the intelligent monitoring system for garden plant growth based on visual recognition, according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the steps of the intelligent monitoring method for garden plant growth based on visual recognition, as described in an embodiment of the present invention. Figure 3 This is a flowchart illustrating the steps of determining health based on the comparison results of visual performance parameters and preset values ​​in an embodiment of the present invention. Figure 4 This is a flowchart illustrating the steps of determining the monitoring status based on the adjustment period of the adjusted threshold, according to an embodiment of the present invention. Detailed Implementation

[0028] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0029] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0030] It should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0031] Please see Figure 1 As shown, it is a structural schematic diagram of the intelligent monitoring system for garden plant growth based on visual recognition according to an embodiment of the present invention.

[0032] The system includes a multimodal data acquisition unit, a data preprocessing unit, a growth status identification unit, an environmental sensing unit, a decision-making unit, an execution unit, and an analysis unit.

[0033] The multimodal data acquisition unit includes at least a visible light camera and a multispectral imager capable of periodically acquiring data, which are used to acquire visible light images, multispectral images, and depth images of garden plants; The data preprocessing unit is connected to the multimodal data acquisition unit and is used to perform image registration and data enhancement on the acquired multimodal images. The growth status identification unit is connected to the data preprocessing unit. It is used to extract macroscopic, mesoscopic and microscopic features of plants from the preprocessed image based on image recognition technology and an improved deep learning model, as well as to identify the plant growth stage, disease type and growth rate. The environmental sensing unit is used to acquire environmental temperature and humidity, soil moisture and light intensity based on multiple sensors; The decision-making unit is connected to the growth state recognition unit and the environment sensing unit respectively. It is used to periodically construct a visual feature-environment parameter association model and generate personalized control strategies based on a dynamic threshold self-learning mechanism. The execution unit is connected to the decision-making unit and is used to execute the generated personalized control strategy; The analysis unit is connected to the execution unit and is used to calculate visual performance health parameters, determine the monitoring status of garden plant growth based on the visual performance health parameters, adjust the construction cycle of the visual feature-environment parameter correlation model based on the monitoring status, and adjust the adjustment cycle of the threshold based on the adjusted monitoring status. Among them, the visual health parameter is the weighted result of morphological qualification rate, canopy coverage rate and leaf color index.

[0034] Specifically, the multimodal data acquisition unit includes a mobile inspection robot equipped with an RGB camera, an infrared thermal imager, and a multispectral imager, and a three-axis gimbal to achieve 360° shooting without blind spots. It also includes fixed monitoring equipment, which is embedded in key locations in the garden. The acquisition cycle can be set remotely, complementing the mobile robot to ensure full monitoring coverage.

[0035] Specifically, the macroscopic aspects of plants include plant height and crown width, the mesoscopic aspects include leaf cluster morphology, and the microscopic aspects include texture and pigment distribution.

[0036] Specifically, the data preprocessing unit includes using a wavelet threshold denoising algorithm to remove illumination noise and environmental noise from the image; achieving spatial alignment of multimodal images through SIFT feature point matching to complete image registration; expanding the sample set by random cropping, flipping, and contrast adjustment to address the problem of scarce garden plant image samples; and normalizing the pixel values ​​of the multimodal images to the [0,1] interval.

[0037] Specifically, the visual feature-environmental parameter association model is the core algorithm of the linkage decision-making module, connecting the intelligent growth status recognition module and the environmental sensing unit. By establishing a nonlinear mapping of dual-modal data, it achieves a three-dimensional assessment of plant growth status. The model input includes three types of data: first, 128-dimensional visual features, such as 32-dimensional disease features, including the proportion of lesion area and spread rate; 48-dimensional nutritional features, including chlorophyll content; and 48-dimensional growth features, including the daily increase in plant height; second, 8-dimensional environmental parameters, such as soil moisture content and meteorological parameters; and third, auxiliary information such as plant variety (One-Hot encoding) and growth stage (binary encoding), which are standardized and filtered for outliers before input.

[0038] The core of this model adopts an architecture of "cross-modal interaction, feature selection, and improved LightGBM prediction": First, 30 sets of interactive features are generated through static interactions, such as lesion proportion and air humidity, and dynamic interactions, such as lesion spread rate and humidity change rate. An attention mechanism is then used to assign high weights (0.3-0.5) to key associations, such as "lesion and humidity" in fungal disease scenarios. Next, 80 core features are selected through mutual information entropy and RFE (Reference-Free Evaluator) to reduce computational complexity. The model is trained using improved LightGBM, with offline parameter optimization using 5-fold cross-validation and online incremental updates via federated learning, balancing efficiency and data security.

[0039] The output is corrected for variety and growth stage, such as seedling stage, by a correction factor of 0.9, and finally outputs a health score (0-100 points) and three risk levels (high / medium / low) for disease, nutrition, and environmental stress.

[0040] Please see Figure 2 The above is a flowchart of the steps of the intelligent monitoring method for garden plant growth based on visual recognition in an embodiment of the present invention.

[0041] The steps of the intelligent monitoring method for garden plant growth based on visual recognition include: S1, acquire visible light images, multispectral images and depth images of garden plants through a multimodal data acquisition unit that includes at least a visible light camera and a multispectral imager capable of periodically acquiring data; S2, at least image registration and data enhancement are performed on the acquired multimodal images through a data preprocessing unit connected to the multimodal data acquisition unit; S3, the growth state identification unit connected to the data preprocessing unit extracts macroscopic, mesoscopic and microscopic features of the plant from the preprocessed image using an improved deep learning model based on image recognition technology, and identifies the plant growth stage, disease type and growth rate. S4 acquires ambient temperature and humidity, soil moisture and light intensity based on multiple sensors through the environmental sensing unit; S5, a visual feature-environment parameter association model is periodically constructed by the decision unit connected to the growth state recognition unit and the environment sensing unit respectively, and a personalized control strategy is generated based on the dynamic threshold self-learning mechanism. S6, the generated personalized control strategy is executed by the execution unit connected to the decision-making unit; S7, the analysis unit connected to the execution unit calculates visual performance health parameters, determines the monitoring status of garden plant growth based on the visual performance health parameters, adjusts the construction cycle of the visual feature-environment parameter correlation model based on the monitoring status, and adjusts the adjustment cycle of the threshold based on the adjusted monitoring status. Among them, the visual health parameter is the weighted result of morphological qualification rate, canopy coverage rate and leaf color index.

[0042] Please see Figure 3 As shown, it is a flowchart of the steps for determining the health status based on the comparison results of visual performance parameters and preset values ​​in an embodiment of the present invention.

[0043] Specifically, taking the precise maintenance of garden plants as an example, the core parameter settings deeply combine the physical performance limits of the monitoring hardware with the actual fault tolerance requirements of complex natural environments. During long-term operation, the system continuously collects and analyzes historical growth data, environmental response records, and feedback on control effects, and uses these statistical patterns to set the values ​​of subsequent preset or critical parameters.

[0044] Specifically, the morphological qualification rate is the percentage of plants that meet the target plant type, i.e., the crown width / plant height ratio; crown coverage is the ratio of the vegetation projection area to the planned area, used to evaluate the greening effect; leaf color index is a green index calculated based on RGB or multispectral density, where the leaf color index uses NDVI, and its original range is approximately [-1, 1]. A piecewise linear stretching function is needed to map the typical range of healthy vegetation to [0, 1]. With a preset value L0 = 0.85, the comparison process between the visual health parameter L and the preset value L0 is as follows: If the visual health parameter L is greater than or equal to the preset value L0, then the monitoring status of the garden plants' growth is deemed qualified. If the visual health parameter L is less than the preset value L0, then the monitoring status of the garden plants' growth is determined to be unqualified.

[0045] Specifically, when the monitoring status is unqualified, the difference between the preset value and the visual health parameter is calculated. If the difference is greater than the preset difference, it indicates that the original model correlation law has failed when the plant variety is changed or the climate zone undergoes significant changes, such as entering an abnormal climate year. In this case, the construction cycle of the visual feature-environment parameter correlation model is adjusted based on the ratio of the difference to the preset difference. The preset ratio of the difference to the preset difference is P0 = 1.2. The comparison process based on the ratio P of the difference to the preset difference and the preset ratio P0 is as follows: If the ratio P of the difference to the preset difference is less than or equal to the preset ratio P0, the construction period of the visual feature-environment parameter association model will be adjusted to 0.92 times the original construction period, where the adjusted value will be rounded up. If the ratio P of the difference to the preset difference is greater than the preset ratio P0, the construction period of the visual feature-environment parameter association model will be adjusted to 0.76 times the original construction period, where the adjusted value will be rounded up.

[0046] Specifically, if the monitoring status is unqualified after adjusting the construction cycle of the visual feature-environment parameter association model, the variance of the visual health parameters of all individuals of the same species under the same care conditions within the monitoring area is calculated. If the variance is greater than the preset variance, it indicates that there is a problem with the dynamic threshold self-learning mechanism, causing the threshold to drift to an unreasonable range. The adjustment cycle of the threshold is then adjusted based on the difference between the variance and the preset variance. The preset difference between the variance and the preset variance is Q0=0.1. The comparison process between the difference Q and the preset difference Q0 is as follows: If the difference Q between the variance and the preset variance is less than or equal to the preset difference Q0, the adjustment period of the threshold is adjusted to 0.89 times the original adjustment period, where the adjusted value is rounded up. If the difference Q between the variance and the preset variance is greater than the preset difference Q0, the adjustment period of the threshold is adjusted to 0.65 times the original adjustment period, where the adjusted value is rounded up.

[0047] Please see Figure 4 The diagram shown is a flowchart illustrating the steps of determining the monitoring status based on the adjustment period of the adjusted threshold in an embodiment of the present invention.

[0048] Specifically, if the monitoring status is unqualified after adjusting the threshold adjustment period, the threshold adjustment period is repeated at least once until the number of adjustments is less than the preset number and the monitoring status is qualified, or the number of adjustments is equal to the preset number, at which point the adjustment stops. If the monitoring status is unqualified after stopping the adjustment, a time-visual performance health parameter curve is plotted based on the acquired historical visual performance health parameters. The curve integral is calculated. If the integral is less than the preset integral, it indicates that the sensor acquisition timing is inappropriate, causing data deviation. The sensor acquisition data period is then adjusted based on the ratio of the integral to the preset integral. The preset ratio of the integral to the preset integral is R0 = 0.92. The comparison process between the ratio R of the integral to the preset integral and the preset ratio R0 is as follows: If the ratio R of the integral to the preset integral is less than or equal to the preset ratio R0, the period of the sensor data acquisition will be adjusted to 0.64 times the original period, and the adjusted value will be rounded up. If the ratio R of the integral to the preset integral is greater than the preset ratio R0, the period of the sensor data acquisition will be adjusted to 0.83 times the original period, and the adjusted value will be rounded up.

[0049] Specifically, if the monitoring status is unqualified after adjusting the period of sensor data acquisition, the average pixel offset of the matching feature point pair between two images at the same timestamp is calculated. If the average pixel offset is greater than the preset tolerance threshold, it indicates that the multimodal image registration has failed, that is, there is spatial inconsistency between different data sources or within the data, and a notification of image registration failure is issued.

[0050] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0051] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A visual recognition-based intelligent monitoring system for garden plant growth, characterized in that, The system comprises: a multi-modal data acquisition unit comprising at least a visible light camera and a multispectral imager capable of periodically acquiring data, for acquiring visible light images, multispectral images and depth images of the garden plants; a data preprocessing unit connected to the multi-modal data acquisition unit, for at least image registration and data enhancement of the acquired multi-modal images; a growth state recognition unit connected to the data preprocessing unit, for extracting plant macro, meso and micro features from the preprocessed images based on image recognition technology using an improved deep learning model, and identifying plant growth stages, disease types and growth rates; an environmental sensing unit for acquiring environmental temperature and humidity, soil moisture and light intensity based on multiple sensors; a decision unit connected to the growth state recognition unit and the environmental sensing unit, for periodically constructing a visual feature-environment parameter correlation model, and generating individualized regulation strategies based on a dynamic threshold self-learning mechanism; an execution unit connected to the decision unit, for executing the generated individualized regulation strategies; an analysis unit connected to the execution unit, for calculating visual performance health parameters, determining the monitoring state of the garden plant growth based on the visual performance health parameters, adjusting the construction period of the visual feature-environment parameter correlation model based on the monitoring state, and adjusting the adjustment period of the threshold based on the adjusted monitoring state, wherein the visual performance health parameters are the weighted results of the morphological qualification rate, the canopy coverage rate and the leaf color index. 2.The visual recognition-based intelligent monitoring system for garden plant growth according to claim 1, wherein, The improved deep learning model in the growth state recognition unit comprises an input layer for multi-modal image channel fusion, a convolutional block and a residual connection feature extraction layer, a spatial attention and channel attention parallel weighting attention mechanism layer, a macro-meso-micro feature splicing feature fusion layer, and an output layer outputting at least disease type, plant height and canopy, and growth rate. 3.The visual recognition based intelligent monitoring system for garden plant growth according to claim 1, wherein, The dynamic threshold self-learning mechanism in the decision unit comprises constructing a sample library based on historical monitoring data, and training a threshold prediction model based on the sample library using a gradient boosting tree algorithm, and periodically adjusting the threshold based on the information identified by the growth state recognition unit. 4.The visual recognition-based intelligent monitoring system for garden plant growth according to claim 1, wherein, The analysis unit is also used to calculate the difference between the preset value and the visual performance health parameters in the case of unqualified monitoring state; The analysis unit is also used to adjust the construction period of the visual feature-environment parameter correlation model based on the ratio of the difference to the preset difference in the case of the difference being greater than the preset difference; wherein the unqualified monitoring state is determined when the preset value is greater than the visual performance health parameters. 5.The visual recognition-based intelligent monitoring system for garden plant growth according to claim 4, characterized in that, The analysis unit is also used to reduce the construction period of the visual feature-environment parameter correlation model based on the ratio of the difference to the preset difference, and the reduction amplitude of the construction period of the visual feature-environment parameter correlation model is proportional to the ratio. 6.The visual recognition-based intelligent monitoring system for garden plant growth according to claim 5, wherein, The analysis unit is also used to calculate the variance of the visual performance health parameters of all individuals of the same species under the same maintenance condition in the monitoring area in the case of unqualified monitoring state after adjusting the construction period of the visual feature-environment parameter correlation model. The analysis unit is further configured to adjust the adjustment period of the threshold based on a difference between the variance and the preset variance when the variance is greater than the preset variance. 7.The visual recognition based intelligent monitoring system for garden plant growth according to claim 6, wherein, The analysis unit is further configured to decrease the adjustment period of the threshold based on the difference between the variance and the preset variance, and the decrease amplitude of the adjustment period of the threshold is proportional to the difference. 8.The visual recognition-based intelligent monitoring system for garden plant growth according to claim 7, wherein, The analysis unit is further configured to repeat the adjustment of the adjustment period of the threshold at least once when the monitoring state is unqualified after the adjustment of the adjustment period of the threshold, until the adjustment times is less than a preset number of times and the monitoring state is qualified or the adjustment times is equal to the preset number of times. The analysis unit is further configured to draw a time-vision performance health parameter curve based on the acquired historical vision performance health parameters when the monitoring state is unqualified after the adjustment. The analysis unit is further configured to calculate an integral of the time-vision performance health parameter curve, and adjust the data acquisition period of the sensor based on a ratio between the integral and a preset integral when the integral is less than the preset integral. 9.The visual recognition-based intelligent monitoring system for garden plant growth according to claim 8, wherein, The analysis unit is further configured to decrease the data acquisition period of the sensor based on the ratio between the integral and the preset integral, and the decrease amplitude of the data acquisition period of the sensor is inversely proportional to the ratio. 10.The visual recognition-based intelligent monitoring system for garden plant growth according to claim 9, wherein, The analysis unit is further configured to calculate an average pixel displacement of matching feature points between two images at the same time stamp when the monitoring state is unqualified after the adjustment of the data acquisition period of the sensor. The analysis unit is further configured to send a notification of image registration failure when the average pixel displacement is greater than a preset tolerance threshold.

Images