Greenhouse environment adaptive regulation and control system and method based on vegetable growth stage recognition

Abstract

The invention relates to the technical field of facility agriculture intelligent regulation and control, in particular to a greenhouse environment self-adaptive regulation and control system and method based on vegetable growth stage recognition, and the system comprises a growth stage recognition module, an environment parameter monitoring module, a regulation and control strategy generation module and a regulation and control execution and feedback correction module. Static visual features and dynamic time sequence features are fused, a ResNet50 and Transform fusion model and a confidence verification mechanism are adopted, and the growth stage from the seedling stage to the mature stage is accurately recognized; and calling the dynamically updated parameter reference library, generating a regulation and control strategy by combining deviation grading and priority dynamic regulation logic, controlling multiple types of equipment to execute stepless regulation, and correcting and optimizing a regulation and control effect through closed-loop feedback. According to the invention, the problems of misjudgment, fixed reference and rigid response in the traditional regulation and control stage are solved, the regulation and control accuracy and robustness are improved, the resource waste is reduced, and the improvement of the yield and quality of vegetables is facilitated.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology in facility agriculture, and more specifically, to an adaptive control system and method for greenhouse environment based on vegetable growth stage identification. Background Technology

[0002] In large-scale facility agriculture, the precise control of greenhouse environmental parameters (temperature, light, humidity, etc.) directly affects the growth efficiency and quality of vegetables. Existing greenhouse environmental control technologies largely rely on fixed preset parameters or human experience to formulate control strategies, lacking precise perception of the real-time growth stages of vegetables. This leads to a mismatch between control actions and the growth needs of vegetables at different stages, often resulting in problems such as over-irrigation during the seedling stage and insufficient light during the fruiting stage, causing waste of water and energy resources, and making it difficult to guarantee vegetable yield and quality. Although some technologies have attempted to introduce image recognition to assist control, they generally suffer from insufficient recognition accuracy, relying only on single visual features without integrating dynamic temporal features, and lacking confidence verification mechanisms, making it prone to misjudging growth stages. Furthermore, existing control benchmark libraries are mostly fixed parameters, unable to be dynamically updated according to the actual growth status of vegetables, and the control priorities are rigid, making it difficult to respond quickly to emergencies when temperature or light exceeds safe thresholds, further exacerbating the risk of control mismatch. Therefore, there is an urgent need for a greenhouse environmental adaptive control system based on precise growth stage identification and with dynamic adaptability to solve the core pain points of existing technologies. Summary of the Invention

[0003] In view of this, the present invention addresses the shortcomings of the prior art by proposing a greenhouse environment adaptive control system and method based on vegetable growth stage identification, aiming to solve at least one of the problems mentioned in the background art.

[0004] In a first aspect, the present invention provides a greenhouse environment adaptive control system based on vegetable growth stage identification, comprising: a growth stage identification module configured to acquire a target greenhouse vegetable canopy image, perform preprocessing operations such as denoising, segmentation and target region extraction on the target greenhouse vegetable canopy image, extract static visual features and dynamic temporal features of the vegetable, and use a multi-feature fusion identification model combined with confidence verification to lock the current vegetable growth stage, wherein the multi-feature fusion identification model realizes the stage determination by weighted fusion calculation of static features and dynamic features, and the confidence verification determines the stage locking method according to the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets a preset threshold; An environmental parameter monitoring module is configured to collect environmental data of the target greenhouse, including temperature, air humidity, light intensity, soil-related parameters, and CO2 concentration. The regulation strategy generation module is configured to call a preset growth stage environmental parameter benchmark library, match the monitored environmental data with the optimal parameter range corresponding to the current growth stage, calculate the parameter deviation level, and generate an adaptive regulation strategy according to a preset regulation priority. The regulation execution and feedback correction module is configured to control the greenhouse regulation equipment to perform corresponding operations according to the adaptive regulation strategy, verify the regulation effect by re-collecting environmental data and vegetable canopy images, and dynamically correct the regulation strategy.

[0005] In some embodiments, the growth stage identification module is configured to acquire images of the canopy of a target greenhouse vegetable, perform preprocessing operations such as denoising, segmentation, and target region extraction on the canopy images, extract static visual features and dynamic temporal features of the vegetable, and use a multi-feature fusion identification model combined with confidence verification to determine the current growth stage of the vegetable. The multi-feature fusion identification model determines the stage by weighted fusion calculation of static and dynamic features. The confidence verification determines the stage locking method based on the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets a preset threshold, including: The vegetable growth stages include the seedling stage, seedling stage, flowering stage, fruiting stage, and maturity stage; When the number of target vegetable leaves is less than or equal to a preset first quantity threshold, and the growth rate of the target vegetable leaf area is greater than or equal to a preset first growth rate threshold, it is determined to be the seedling stage; When the number of target vegetable leaves is greater than the first quantity threshold and less than the preset second quantity threshold, and the growth rate of the target vegetable plant height is greater than or equal to the preset second growth rate, it is determined to be the seedling stage; When the number of target vegetable flowers detected is greater than or equal to a preset third quantity threshold, and the confidence level is greater than or equal to a preset first threshold, it is determined to be the flowering period; When the diameter of the target vegetable fruit is detected to be greater than or equal to a preset second threshold, it is determined to be the result period; When the percentage of colored pixels in the target vegetable fruit is greater than or equal to a preset third threshold, and the leaf area growth rate is less than or equal to a preset third growth rate threshold, it is determined to be the maturity period.

[0006] In some embodiments, the growth stage identification module is configured to acquire images of the canopy of a target greenhouse vegetable, perform preprocessing operations such as denoising, segmentation, and target region extraction on the canopy images, extract static visual features and dynamic temporal features of the vegetable, and use a multi-feature fusion identification model combined with confidence verification to determine the current growth stage of the vegetable. The multi-feature fusion identification model determines the stage by weighted fusion calculation of static and dynamic features. The confidence verification determines the stage locking method based on the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets a preset threshold. The module further includes: Images of the greenhouse vegetable canopy are captured using a high-definition RGB camera, with the acquisition frequency dynamically adjusted according to the growth stage, including: When the target vegetable is in the seedling stage or the mature seedling stage, the sampling frequency is a preset first frequency value; When the target vegetable is in its flowering, fruiting, or ripening stage, the sampling frequency is a preset second frequency value; After the data collection is complete, a three-layer filtering process is performed, including: When the standard deviation of the image grayscale value is greater than or equal to the preset first standard deviation threshold, the sharpness meets the standard. When the proportion of vegetable canopy pixels in the image is greater than or equal to a preset first proportion threshold, the target area meets the standard. When the percentage of occluded pixels in an image is less than or equal to a preset second percentage threshold, the occlusion is considered acceptable. When the sharpness, target area, and occlusion of the acquired image all meet the standards, it is a valid image; otherwise, the camera is triggered to re-acquire the image. If the number of resampling attempts equals the preset third threshold and no valid image is obtained, the previous recognition result is retained.

[0007] In some embodiments, the growth stage identification module is configured to acquire images of the canopy of a target greenhouse vegetable, perform preprocessing operations such as denoising, segmentation, and target region extraction on the canopy images, extract static visual features and dynamic temporal features of the vegetable, and use a multi-feature fusion identification model combined with confidence verification to determine the current growth stage of the vegetable. The multi-feature fusion identification model determines the stage by weighted fusion calculation of static and dynamic features. The confidence verification determines the stage locking method based on the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets a preset threshold. The module further includes: The static visual features include the number of leaves, leaf area, plant height, leaf color, and number of flowers and fruits. The number of leaves is extracted by connected component counting and only leaves with an area greater than or equal to a preset fourth threshold are counted. The leaf area is calculated by combining the pixel ratio with the camera calibration coefficient. The dynamic time-series features include the daily growth rate of leaf area and the daily growth rate of plant height calculated based on image sequences within a preset first time threshold, and abnormal data exceeding a preset multiple standard deviation must be removed when calculating the growth rate.

[0008] In some embodiments, the growth stage identification module is configured to acquire images of the canopy of a target greenhouse vegetable, perform preprocessing operations such as denoising, segmentation, and target region extraction on the canopy images, extract static visual features and dynamic temporal features of the vegetable, and use a multi-feature fusion identification model combined with confidence verification to determine the current growth stage of the vegetable. The multi-feature fusion identification model determines the stage by weighted fusion calculation of static and dynamic features. The confidence verification determines the stage locking method based on the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets a preset threshold. The module further includes: The multi-feature fusion recognition model is a fusion model of ResNet50 and Transformer. The model training dataset contains a preset number of labeled images of different vegetables and different growth stages, and the recognition accuracy is greater than or equal to a preset fifth threshold. When the model output confidence level is greater than or equal to the preset sixth threshold, the current growth stage is directly locked. When the model output confidence is less than the sixth threshold and greater than or equal to the preset seventh threshold, the stage with the highest frequency among the last three recognition results is selected. When the model output confidence level is less than the seventh threshold, the previous valid identification result is retained and a manual review reminder is triggered.

[0009] In some embodiments, the regulation strategy generation module is configured to call a preset growth stage environmental parameter benchmark library, match the monitored environmental data with the optimal parameter range corresponding to the current growth stage, calculate the parameter deviation level, and generate an adaptive regulation strategy according to a preset regulation priority, including: The preset growth stage environmental parameter benchmark library contains parameters for each growth stage of tomatoes, cucumbers, and lettuce, and each parameter is divided into three levels: optimal range, warning range, and safety threshold. If the same vegetable variety still shows abnormal growth after three consecutive adjustments at the same growth stage, the upper limit of the optimal range of the parameter will be automatically lowered or the lower limit will be adjusted upwards.

[0010] In some embodiments, the regulation strategy generation module is configured to call a preset growth stage environmental parameter benchmark library, match the monitored environmental data with the optimal parameter range corresponding to the current growth stage, calculate the parameter deviation level, and generate an adaptive regulation strategy according to a preset regulation priority, including: The deviation level is determined by deviation rate calculation. When the deviation rate is less than or equal to a preset first deviation threshold, the deviation level is a minor deviation. When the deviation rate is greater than the first deviation threshold and less than or equal to the preset second deviation threshold, the deviation level is small deviation; When the deviation rate is greater than the second deviation threshold and less than or equal to the preset third deviation threshold, the deviation level is medium deviation; When the deviation rate is greater than the third deviation threshold, the deviation level is large deviation; The control priority is sorted by light intensity, temperature, CO2 concentration, water and fertilizer parameters, and air humidity by default. When the temperature or light intensity exceeds the safety threshold, its control priority automatically rises to first.

[0011] In some embodiments, the regulation execution and feedback correction module is configured to control the greenhouse regulation equipment to perform corresponding operations according to the adaptive regulation strategy, verify the regulation effect by re-acquiring environmental data and vegetable canopy images, and dynamically correct the regulation strategy, including: The control equipment includes LED supplemental lighting, shade netting, ventilation fans, heating pipes, wet curtain cooling system, CO2 generator, and integrated water and fertilizer device; Before executing a control operation, the equipment operating status must be checked. If there is no fault, the control strategy will be executed. If there is a fault, the backup equipment will be switched and an alarm will be triggered. If there is no backup equipment, the control level will be reduced and the control duration will be extended.

[0012] In some embodiments, the regulation execution and feedback correction module is configured to control the greenhouse regulation equipment to perform corresponding operations according to the adaptive regulation strategy, verify the regulation effect by re-acquiring environmental data and vegetable canopy images, and dynamically correct the regulation strategy, and further includes: When the deviation level is micro-deviation, data is re-collected within the preset first time threshold after adjustment; When the deviation level is small or medium, data is re-acquired within the preset second time threshold after adjustment. When the deviation level is large, data will be re-collected within the preset third time threshold after adjustment. The current strategy is maintained when the parameters revert to the optimal range and the vegetable canopy image shows no abnormalities. When parameters fail to regress, the control intensity is increased according to the deviation level. If the image displays abnormalities, immediately stop the current adjustment, lower the relevant parameters, and update the corresponding parameter range in the benchmark library.

[0013] Secondly, this invention provides a method for adaptive control of greenhouse environment based on vegetable growth stage identification, comprising the following steps: Images of the canopy of target greenhouse vegetables are acquired. After preprocessing the images by denoising, segmentation, and target region extraction, static visual features and dynamic temporal features of the vegetables are extracted. A multi-feature fusion recognition model combined with confidence verification is used to determine the current growth stage of the vegetables. The multi-feature fusion recognition model determines the stage by weighted fusion calculation of static and dynamic features. The confidence verification determines the stage locking method based on the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets a preset threshold. Collect environmental data of the target greenhouse, including temperature, air humidity, light intensity, soil-related parameters, and CO2 concentration; The system calls a preset growth stage environmental parameter benchmark library, matches the monitored environmental data with the optimal parameter range corresponding to the current growth stage, calculates the parameter deviation level, and generates an adaptive control strategy according to the preset control priority. The greenhouse control equipment is controlled to perform corresponding operations according to the adaptive control strategy. The control effect is verified by re-collecting environmental data and vegetable canopy images, and the control strategy is dynamically corrected.

[0014] Compared with existing technologies, the beneficial effects of this invention are as follows: By combining a multi-feature fusion recognition model with a confidence verification mechanism, and integrating static visual features with dynamic temporal features, the accuracy and reliability of vegetable growth stage identification are significantly improved, avoiding the control mismatch problem caused by stage misjudgment from the source; the environmental parameter benchmark library for growth stages supports dynamic updates and can adaptively adjust the parameter range according to the actual growth status of vegetables, adapting to the growth needs of various vegetables such as tomatoes, cucumbers, and lettuce, breaking through the limitations of traditional fixed benchmark libraries; the dynamic adjustment logic of control priority ensures that emergency control is prioritized when key parameters such as temperature and light exceed the safety threshold, maximizing the protection of vegetable growth safety; through the "identification-monitoring-control-feedback" closed-loop mechanism, combined with deviation-level control and equipment failure response strategies, the accuracy and robustness of environmental control are significantly improved, effectively reducing water and energy waste, while increasing vegetable yield and quality, providing strong support for large-scale and intelligent planting in facility agriculture.

[0015] The above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure.

[0016] Other features and aspects of this disclosure will become clearer from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 Functional block diagram of a greenhouse environment adaptive control system based on vegetable growth stage identification provided in an embodiment of the present invention; Figure 2 A flowchart of a greenhouse environment adaptive control method based on vegetable growth stage identification provided in an embodiment of the present invention. Detailed Implementation

[0019] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] First embodiment, see reference Figure 1-2 As shown, a greenhouse environment adaptive control system based on vegetable growth stage identification according to an embodiment of this application includes: The growth stage identification module is configured to acquire images of the canopy of target greenhouse vegetables, perform preprocessing operations such as denoising, segmentation, and target region extraction on the images, extract static visual features and dynamic temporal features of the vegetables, and use a multi-feature fusion identification model combined with confidence verification to determine the current growth stage of the vegetables. The multi-feature fusion identification model determines the stage by weighted fusion calculation of static and dynamic features, and the confidence verification determines the stage locking method based on the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets a preset threshold. An environmental parameter monitoring module is configured to collect environmental data of the target greenhouse, including temperature, air humidity, light intensity, soil-related parameters, and CO2 concentration. The regulation strategy generation module is configured to call a preset growth stage environmental parameter benchmark library, match the monitored environmental data with the optimal parameter range corresponding to the current growth stage, calculate the parameter deviation level, and generate an adaptive regulation strategy according to a preset regulation priority. The regulation execution and feedback correction module is configured to control the greenhouse regulation equipment to perform corresponding operations according to the adaptive regulation strategy, verify the regulation effect by re-collecting environmental data and vegetable canopy images, and dynamically correct the regulation strategy.

[0021] It should be understood that adaptive control of the greenhouse environment is achieved through a closed-loop logic of "growth stage identification - environmental parameter monitoring - control strategy generation - control execution and feedback correction". Its core working logic is as follows: based on the accurate identification of vegetable growth stages, control strategies are dynamically generated and executed by monitoring the matching degree between environmental parameters and the optimal parameters for that stage in real time, and then continuously corrected through feedback data to ensure that the control actions always adapt to the needs of vegetable growth. This architecture breaks through the limitations of traditional fixed-parameter control, deeply integrating computer vision technology and automatic control technology. Its core lies in the full-link collaboration of "accurate identification - real-time monitoring - dynamic strategy - closed-loop feedback".

[0022] Image preprocessing employs a combination of Gaussian filtering for noise reduction, adaptive threshold segmentation, and morphological opening / closing operations, effectively separating the vegetable canopy from the soil, supports, and other background elements. The weighted fusion of static visual features and dynamic temporal features is determined through model training optimization, specifically: leaf number 0.25, leaf area 0.2, flower and fruit status 0.3, leaf color 0.15, and dynamic growth rate 0.1. This weight allocation is based on the contribution analysis of different features to the growth stage determination. The preset confidence threshold is divided into two levels: the lowest confidence threshold for triggering effective output is 0.8, meaning that the final growth stage result will only be output when the model outputs a confidence level ≥ 0.8; if it is below this threshold, the stage determination will not be updated to avoid misjudgment affecting regulation. The environmental data acquisition frequency is set to once per minute, with the temperature sensor having an accuracy of ±0.5℃, the air humidity sensor having an accuracy of ±2%RH, and the light intensity sensor having a range of 0-2000μmol / m². 2 • s (accuracy ±5%), CO2 sensor range 0-5000ppm (accuracy ±10ppm), soil-related parameters including soil moisture (range 0-100%, accuracy ±2%), soil EC value (range 0-5ms / cm, accuracy ±0.01ms / cm) and soil pH value (range 4.0-8.0, accuracy ±0.1), this accuracy configuration can meet the needs of precise control of greenhouse environment; The initial data for the growth stage-environmental parameter benchmark library comes from agricultural experimental data and authoritative literature, covering three core vegetables: tomatoes, cucumbers, and lettuce. More varieties can be adapted through a dynamic update mechanism in the future. The deviation level calculation is based on the "median value of the optimal interval" to ensure the objectivity of deviation judgment. The regulation priority setting is based on the physiological needs of vegetable growth. Light and temperature directly affect photosynthesis and enzyme activity, so they are listed as core regulation objects. All control devices support stepless adjustment, such as LED supplementary lighting power adjustment range of 0-100%, ventilation fan speed adjustment range of 0-3000r / min, and water-fertilizer ratio adjustment accuracy of 0.1ms / cm for integrated water and fertilizer device, which can avoid parameter fluctuations caused by "one-size-fits-all" control; the time interval between feedback data acquisition and control execution is dynamically adjusted according to the deviation level to ensure timely correction.

[0023] In some specific embodiments, the growth stage identification module is configured to acquire images of the canopy of the target greenhouse vegetable, perform preprocessing operations such as denoising, segmentation, and target region extraction on the canopy images, extract the static visual features and dynamic temporal features of the vegetable, and use a multi-feature fusion identification model combined with confidence verification to determine the current growth stage of the vegetable. The multi-feature fusion identification model determines the stage by weighted fusion calculation of static and dynamic features. The confidence verification determines the stage locking method based on the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets a preset threshold. This includes: The vegetable growth stages include the seedling stage, seedling stage, flowering stage, fruiting stage, and maturity stage; When the number of target vegetable leaves is less than or equal to a preset first quantity threshold, and the growth rate of the target vegetable leaf area is greater than or equal to a preset first growth rate threshold, it is determined to be the seedling stage; When the number of target vegetable leaves is greater than the first quantity threshold and less than the preset second quantity threshold, and the growth rate of the target vegetable plant height is greater than or equal to the preset second growth rate, it is determined to be the seedling stage; When the number of target vegetable flowers detected is greater than or equal to a preset third quantity threshold, and the confidence level is greater than or equal to a preset first threshold, it is determined to be the flowering period; When the diameter of the target vegetable fruit is detected to be greater than or equal to a preset second threshold, it is determined to be the result period; When the percentage of colored pixels in the target vegetable fruit is greater than or equal to a preset third threshold, and the leaf area growth rate is less than or equal to a preset third growth rate threshold, it is determined to be the maturity period.

[0024] It should be understood that precise segmentation of each stage is achieved through a combination of "static features + dynamic features," avoiding misjudgments caused by relying on a single feature. The working principle is as follows: For the typical morphological and growth characteristic differences at each stage of the vegetable growth cycle, corresponding feature threshold combinations are set. Only when two or more features simultaneously meet the threshold requirements is the corresponding stage determined, ensuring the accuracy and stability of stage segmentation. For example, in the seedling stage, not only is the number of leaves required to meet the standard, but the leaf area growth rate threshold must also be met to avoid misjudging slow-growing, weak seedlings as seedlings. In the maturity stage, both fruit coloring and growth rate are combined to ensure that the determination result is consistent with the physiological maturity of the vegetable.

[0025] The division of vegetable growth stages strictly follows botanical principles. The seedling stage is the initial stage of vegetative growth, characterized by a small number of leaves and a rapid growth rate. The mature seedling stage is a stage of vigorous vegetative growth, with the number of leaves and plant height increasing rapidly in tandem. The flowering, fruiting, and ripening stages are reproductive growth stages, marked by the appearance of flowers, the formation of fruits, and the ripening of fruits, respectively. For leafy vegetables such as lettuce, the fruiting stage refers to the reproductive growth stage after flowering and seed production. Although not the stage intended for consumption, it is still included in the classification to ensure the integrity of the growth cycle and to accommodate special planting needs such as seed saving.

[0026] The preset thresholds are as follows: First threshold (number of leaves in the seedling stage) is 4 leaves, and first growth rate threshold (leaf area growth rate in the seedling stage) is 10% / day; second threshold (number of leaves in the mature seedling stage) is 8 leaves, and second growth rate threshold (plant height growth rate in the mature seedling stage) is 8% / day; third threshold (number of flowers in the flowering stage) is 1 flower, and first threshold (confidence level in the flowering stage) is 0.9; second threshold (fruit diameter in the fruiting stage) is 1 cm; third threshold (percentage of colored pixels in the mature fruit stage) is 60%, and third growth rate threshold (leaf area growth rate in the mature fruit stage) is 3% / day. These thresholds were derived from statistical analysis of field trial data for tomatoes, cucumbers, and lettuce, covering the differences in growth characteristics among different varieties. For example, the plant height growth rate of cucumbers in the mature seedling stage is slightly higher than that of tomatoes, which can be fine-tuned using variety adaptation parameters in the benchmark library.

[0027] The system employs a multi-feature combination judgment logic. For example, during the seedling stage, both the leaf quantity range and the plant height growth rate threshold must be met simultaneously. This can effectively distinguish between abnormal plants that "meet the leaf quantity standard but grow slowly" and normal seedlings. The confidence threshold constraint is introduced during the flowering stage because flower morphology is easily affected by image angle and lighting. The model confidence verification is required to ensure the judgment is reliable.

[0028] In some specific embodiments, the growth stage identification module is configured to acquire images of the canopy of target greenhouse vegetables, perform preprocessing operations such as denoising, segmentation, and target region extraction on the images, extract static visual features and dynamic temporal features of the vegetables, and use a multi-feature fusion identification model combined with confidence verification to determine the current growth stage of the vegetables. The multi-feature fusion identification model determines the stage by weighted fusion calculation of static and dynamic features. The confidence verification determines the stage locking method based on the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets a preset threshold. The module also includes: Images of the greenhouse vegetable canopy are captured using a high-definition RGB camera, with the acquisition frequency dynamically adjusted according to the growth stage, including: When the target vegetable is in the seedling stage or the mature seedling stage, the sampling frequency is a preset first frequency value; When the target vegetable is in its flowering, fruiting, or ripening stage, the sampling frequency is a preset second frequency value; After the data collection is complete, a three-layer filtering process is performed, including: When the standard deviation of the image grayscale value is greater than or equal to the preset first standard deviation threshold, the sharpness meets the standard. When the proportion of vegetable canopy pixels in the image is greater than or equal to a preset first proportion threshold, the target area meets the standard. When the percentage of occluded pixels in an image is less than or equal to a preset second percentage threshold, the occlusion is considered acceptable. When the sharpness, target area, and occlusion of the acquired image all meet the standards, it is a valid image; otherwise, the camera is triggered to re-acquire the image. If the number of resampling attempts equals the preset third threshold and no valid image is obtained, the previous recognition result is retained.

[0029] It should be understood that the "dynamic acquisition frequency + three-layer screening mechanism" ensures the validity of input images, providing a high-quality data source for subsequent feature extraction and stage recognition. Its working principle is as follows: the acquisition frequency is adjusted according to the growth rate of the vegetables at different growth stages. High-frequency acquisition is used during the rapid growth stages of seedlings and mature seedlings to ensure the capture of growth dynamics; low-frequency acquisition is used during the relatively stable growth stages of flowering, fruiting, and maturity to reduce system energy consumption. After acquisition, invalid images with blurriness, significant background interference, or severe occlusion are eliminated through three layers of screening based on clarity, target area proportion, and occlusion degree. Simultaneously, a re-acquisition mechanism and an anomaly fallback strategy are set up to ensure the continuity and validity of image data.

[0030] The preset first frequency (sampling frequency during seedling and mature seedling stages) is 3 minutes per sampling, which can promptly capture rapid changes in leaf quantity and leaf area during the seedling stage. The preset second frequency (sampling frequency during flowering, fruiting, and maturity stages) is 5 minutes per sampling, ensuring that no key growth stages are missed while avoiding data redundancy caused by high-frequency sampling. The dynamic adjustment of the sampling frequency is triggered automatically by the system after recognizing the current growth stage, without the need for manual intervention.

[0031] The first standard deviation threshold (clarity threshold) is preset to 120. A grayscale standard deviation ≥120 indicates rich image details and meets the clarity standard. This threshold is based on a large number of greenhouse images and can effectively distinguish clear images from blurry images in environments such as fog or backlight. The first proportion threshold (target area proportion) is preset to 85% to ensure that the core content of the image is the vegetable canopy and reduce the interference of background such as soil and supports on feature extraction. The second proportion threshold (occlusion proportion) is preset to 8%. An occlusion pixel proportion ≤8% indicates that the image has no serious occlusion and key features can be effectively extracted. Exceeding this threshold may lead to misjudgment of features such as leaf quantity and flower and fruit status. The third threshold (number of re-sampling times) is preset to 3 times. If no valid image is obtained after 3 consecutive samplings, the previous recognition result is retained to avoid system control interruption due to abnormal image acquisition. At the same time, an equipment alarm is triggered to remind staff to check the camera position or greenhouse environment.

[0032] The three-layer screening mechanism ensures data source reliability by considering image quality, core content, and interference levels. The combination of re-sampling mechanism and fallback strategy guarantees data quality while maintaining system stability, preventing system paralysis caused by a single failure.

[0033] In some specific embodiments, the growth stage identification module is configured to acquire images of the canopy of target greenhouse vegetables, perform preprocessing operations such as denoising, segmentation, and target region extraction on the images, extract static visual features and dynamic temporal features of the vegetables, and use a multi-feature fusion identification model combined with confidence verification to determine the current growth stage of the vegetables. The multi-feature fusion identification model determines the stage by weighted fusion calculation of static and dynamic features. The confidence verification determines the stage locking method based on the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets a preset threshold. The module also includes: The static visual features include the number of leaves, leaf area, plant height, leaf color, and number of flowers and fruits. The number of leaves is extracted by connected component counting and only leaves with an area greater than or equal to a preset fourth threshold are counted. The leaf area is calculated by combining the pixel ratio with the camera calibration coefficient. The dynamic time-series features include the daily growth rate of leaf area and the daily growth rate of plant height calculated based on image sequences within a preset first time threshold, and abnormal data exceeding a preset multiple standard deviation must be removed when calculating the growth rate.

[0034] It should be understood that the reliability of extracted static visual features and dynamic temporal features is ensured through "precise feature definition + outlier removal," providing high-quality feature input for subsequent model recognition. Its working principle is as follows: static visual feature extraction focuses on typical morphological indicators of vegetable growth stages, ensuring accuracy through specific algorithms; dynamic temporal feature extraction is based on continuous image sequences, reflecting the growth trend of vegetables and avoiding the accidental influence of instantaneous morphological features; simultaneously, an outlier removal mechanism is introduced to eliminate outliers caused by environmental interference or equipment errors, ensuring the authenticity and stability of the feature data.

[0035] The number of leaves was extracted using connected component counting. The specific process was as follows: the preprocessed image was binarized, connected components were marked, and only those with an area ≥ a preset fourth threshold (0.5cm) were counted. 2 The connected regions are considered as effective leaves, eliminating small noise areas caused by dust, withered leaves, etc.; leaf area is calculated by "pixel percentage × actual area conversion factor", which is obtained through camera calibration. The calibration method is as follows: place a standard board of known area in the greenhouse, take an image, calculate the pixel percentage of the standard board, and then obtain the conversion relationship between pixels and actual area to ensure the accuracy of leaf area calculation; plant height is converted by the pixel distance between the highest point of the vegetable canopy and the soil surface in the image, and the soil surface is accurately located by semantic segmentation algorithm; leaf color is represented by the mean of the H channel of the HSV color space, which is more robust to changes in light intensity and can reduce the impact of greenhouse light fluctuations on color features; the number of flowers and fruits is extracted by the YOLOv8 object detection algorithm. The model has been specifically trained for vegetable flowers and fruits in a greenhouse environment, with a recognition accuracy of ≥97%.

[0036] The preset first time threshold (time-series feature calculation period) is 48 hours, which means that the daily growth rate of leaf area and the daily growth rate of plant height are calculated based on the image sequence of the past 48 hours. This period can reflect the growth trend and avoid the impact of short-term fluctuations. The growth rate is calculated using the formula "(current day's feature value - previous day's feature value) / previous day's feature value × 100%" to ensure the objectivity of the calculation results. The preset multiple threshold (abnormal data removal standard) is 3 times, which means that when the feature value at a certain moment exceeds 3 times the standard deviation of the feature value of the past 48 hours, it is judged as abnormal data and removed. This standard conforms to the statistical outlier judgment principle and can effectively exclude feature anomalies caused by sudden environmental factors such as rainstorms and extreme temperatures.

[0037] The combination of static and dynamic features reflects both the current morphological state and the growth trend, providing comprehensive feature input for the multi-feature fusion recognition model. Targeted optimization mechanisms are set for the extraction of each feature, such as the area threshold for leaf number statistics, color space selection for color features, and outlier removal for dynamic features, to ensure the accuracy and reliability of the feature data.

[0038] In some specific embodiments, the growth stage identification module is configured to acquire images of the canopy of target greenhouse vegetables, perform preprocessing operations such as denoising, segmentation, and target region extraction on the images, extract static visual features and dynamic temporal features of the vegetables, and use a multi-feature fusion identification model combined with confidence verification to determine the current growth stage of the vegetables. The multi-feature fusion identification model determines the stage by weighted fusion calculation of static and dynamic features. The confidence verification determines the stage locking method based on the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets a preset threshold. The module also includes: The multi-feature fusion recognition model is a fusion model of ResNet50 and Transformer. The model training dataset contains a preset number of labeled images of different vegetables and different growth stages, and the recognition accuracy is greater than or equal to a preset fifth threshold. When the model output confidence level is greater than or equal to the preset sixth threshold, the current growth stage is directly locked. When the model output confidence is less than the sixth threshold and greater than or equal to the preset seventh threshold, the stage with the highest frequency among the last three recognition results is selected. When the model output confidence level is less than the seventh threshold, the previous valid identification result is retained and a manual review reminder is triggered.

[0039] It should be understood that by clarifying the specific types of multi-feature fusion recognition models and the detailed logic of confidence verification, the core is to ensure high accuracy and stability of recognition during the growth stage through a high-performance model architecture and hierarchical confidence verification. Its working principle is as follows: A fusion model of ResNet50 and Transformer is adopted, fully leveraging the advantages of ResNet50 in static image feature extraction and the capabilities of Transformer in temporal feature modeling to achieve deep fusion of static visual features and dynamic temporal features. Through a hierarchical confidence verification mechanism, the reliability of the model output results is evaluated. High-confidence results are directly adopted, medium-confidence results are optimized through temporal fusion, and low-confidence results trigger manual review, forming a dual guarantee of "automatic recognition + manual fallback" to avoid control mismatch caused by model misjudgment.

[0040] ResNet50 serves as the backbone network for extracting static visual features from images. This network addresses the vanishing gradient problem in deep networks through residual connections, effectively extracting fine-grained features such as leaf texture and flower / fruit morphology. The Transformer encoder models the dependencies of dynamic temporal features, capturing long-term correlations in temporal data such as leaf area growth rate and plant height growth rate through a self-attention mechanism, adapting to the continuous characteristics of vegetable growth. The model training dataset contains 80,000 labeled images (preset quantity 80,000), covering five growth stages of three vegetables: tomatoes, cucumbers, and lettuce. It also includes image samples under different light, temperature, and humidity conditions to ensure the model's generalization ability. The training process employs the cross-entropy loss function, with AdamW as the optimizer and an initial learning rate of 1e-4. A learning rate decay strategy is used to improve training stability. The final model recognition accuracy is ≥98.2% (preset fifth threshold is 98.2%), meeting the needs of precise greenhouse control.

[0041] The preset sixth threshold (high confidence threshold) is 0.9. When the model output confidence is ≥0.9, it indicates that the model's recognition results are highly reliable, and the current growth stage is directly locked. The preset seventh threshold (medium confidence threshold) is 0.8. When the confidence is between 0.8 and 0.9, it indicates that the model's recognition results have some uncertainty. In this case, the stage with the highest frequency among the last three recognition results is selected, and the results are optimized using temporal continuity to reduce the impact of single recognition errors. When the confidence is <0.8, it indicates that the model's recognition results are not reliable enough, possibly due to poor image quality, abnormal vegetable growth, or other factors. In this case, the previous valid recognition result is retained, and a manual review reminder is triggered through system pop-ups, SMS messages, etc. Before the manual review is responded to, the original control strategy is maintained to avoid control disorder caused by stage misjudgment.

[0042] The fusion architecture of ResNet50 and Transformer adapts to the needs of multi-feature input, improving the recognition accuracy by more than 5% compared with a single model; the hierarchical confidence verification mechanism balances recognition efficiency and accuracy, ensuring the reliability of automatic recognition in most scenarios, while solving the recognition problem in extreme scenarios through manual review, and adapting to the complex planting environment of greenhouses.

[0043] In some specific embodiments, the regulation strategy generation module is configured to call a preset growth stage environmental parameter benchmark library, match the monitored environmental data with the optimal parameter range corresponding to the current growth stage, calculate the parameter deviation level, and generate an adaptive regulation strategy according to a preset regulation priority, including: The preset growth stage environmental parameter benchmark library contains parameters for each growth stage of tomatoes, cucumbers, and lettuce, and each parameter is divided into three levels: optimal range, warning range, and safety threshold. If the same vegetable variety still shows abnormal growth after three consecutive adjustments at the same growth stage, the upper limit of the optimal range of the parameter will be automatically lowered or the lower limit will be adjusted upwards.

[0044] It should be understood that the adaptability and timeliness of the benchmark database are ensured through a "three-level parameter range + dynamic optimization mechanism." Its working principle is as follows: the benchmark database sets three levels of parameter ranges for each growth stage of each vegetable: optimal range, warning range, and safety threshold. The optimal range is the ideal environmental range for vegetable growth, the warning range is the critical range requiring intervention, and the safety threshold is the physiological limit that cannot be exceeded. Simultaneously, a dynamic update mechanism for the benchmark database is set. When abnormal growth still occurs in vegetables after multiple consecutive adjustments, it indicates that the current benchmark database parameters do not match the actual planting environment. The system automatically adjusts the optimal range, achieving self-optimization of the benchmark database and overcoming the limitations of traditional fixed benchmark databases.

[0045] The benchmark library contains parameters for each growth stage of tomatoes, cucumbers, and lettuce. Each parameter is divided into three levels of intervals, with the warning interval being ±10% of the optimal interval and the safety threshold being ±20% of the optimal interval. Taking tomatoes during the fruiting stage as an example, the optimal temperature range is 25-30℃ (daytime) / 22-24℃ (nighttime), the corresponding warning interval is 22.5-33℃ (daytime) / 19.8-26.4℃ (nighttime), and the safety threshold is 20-36℃ (daytime) / 17.6-28.8℃ (nighttime); the optimal light intensity is 1500-1800 μmol / m². 2 •s, the warning range is 1350-1980 μmol / m 2 •s, safety threshold is 1200-2160 μmol / m 2 The division of the three-level intervals is based on experimental data on the physiological characteristics of vegetables. The setting of the warning interval can trigger the control action in advance to avoid the parameters deviating too much from the optimal range; the setting of the safety threshold can prevent extreme environments from causing irreversible damage to vegetables.

[0046] The criteria for judging abnormal growth are that the image shows undesirable features such as wilting and scorching of leaves, and the corresponding percentage of abnormal pixels is ≥5%. The trigger condition for dynamic updates is that if abnormal growth is still detected after three consecutive control operations on the same vegetable variety at the same growth stage, it indicates that the current optimal range may not be suitable for the actual planting scenario due to factors such as regional climate, soil conditions, and variety differences. The update range is to automatically lower the upper limit of the optimal range for this parameter or raise the lower limit by 5%-10%, and the specific adjustment direction is determined according to the type of abnormality: if the abnormality is caused by excessively high parameters (such as leaf scorching due to excessive light), the upper limit of the optimal range is lowered; if the abnormality is caused by excessively low parameters (such as slow growth due to excessively low temperature), the lower limit of the optimal range is raised. For example, during the fruiting period of tomatoes, the optimal light range of 1500-1800 μmol / m² is used as the benchmark. 2If leaf scorching still occurs after three adjustments (abnormal pixel rate 6%), the upper limit of the optimal light range will be automatically lowered by 8%, adjusting it to 1500-1656 μmol / m². 2 •s, to ensure that subsequent regulatory actions are more in line with actual needs.

[0047] The three-level parameter range design enables graded response of regulation, avoiding parameter fluctuations caused by "emphasizing control at the slightest deviation"; the dynamic update mechanism makes the benchmark library adaptive, which can be adapted to different planting environments and variety characteristics, eliminating the need for frequent manual parameter adjustments and reducing operation and maintenance costs.

[0048] In some specific embodiments, the regulation strategy generation module is configured to call a preset growth stage environmental parameter benchmark library, match the monitored environmental data with the optimal parameter range corresponding to the current growth stage, calculate the parameter deviation level, and generate an adaptive regulation strategy according to a preset regulation priority, including: The deviation level is determined by deviation rate calculation. When the deviation rate is less than or equal to a preset first deviation threshold, the deviation level is a minor deviation. When the deviation rate is greater than the first deviation threshold and less than or equal to the preset second deviation threshold, the deviation level is small deviation; When the deviation rate is greater than the second deviation threshold and less than or equal to the preset third deviation threshold, the deviation level is medium deviation; When the deviation rate is greater than the third deviation threshold, the deviation level is large deviation; The control priority is sorted by light intensity, temperature, CO2 concentration, water and fertilizer parameters, and air humidity by default. When the temperature or light intensity exceeds the safety threshold, its control priority automatically rises to first.

[0049] It should be understood that by quantifying the degree of deviation and dynamically adjusting priorities, the precision of the control strategy and the emergency response capability are achieved. Its working principle is as follows: First, the deviation rate is used to quantify the degree of deviation between the current environmental parameters and the optimal range, classifying different deviation levels and formulating differentiated control intensities for each level; second, a default control priority is set, and an emergency priority adjustment mechanism is also set. When core parameters such as temperature or light exceed the safety threshold, their control priority is automatically raised to first place, prioritizing emergency control to ensure the safety of vegetable growth and achieving the dual goals of "conventional control for optimization and emergency control for safety."

[0050] The deviation rate is calculated using the formula: "Deviation Rate = (Current Environmental Parameter Value - Median of Optimal Interval) / Median of Optimal Interval × 100%". This formula uses the median of the optimal interval as a benchmark and objectively reflects the degree to which the parameter deviates from the ideal value. A first deviation threshold of 3% is preset. When |deviation Rate| ≤ 3%, it is considered a minor deviation, indicating a small deviation and minimal impact on vegetable growth, requiring no active adjustment but only continuous monitoring. A second deviation threshold of 8% is preset. When 3% < |deviation Rate| ≤ 8%, it is considered a small deviation, requiring mild adjustment to prevent further deviation. A third deviation threshold of 15% is preset. When 8% < |deviation Rate| ≤ 15%, it is considered a medium deviation, requiring moderate adjustment to quickly bring the parameter back to the optimal range. When |deviation Rate| > 15%, it is considered a large deviation, requiring severe adjustment to prevent continued parameter deviation from seriously affecting vegetable growth. For example, the median optimal temperature for tomato seedlings is 24℃, and the current temperature is 26.4℃, with a deviation rate of 10%, which is considered a medium deviation, requiring moderate cooling adjustment.

[0051] The default control priority is ordered as follows: "Light > Temperature > CO2 Concentration > Water and Fertilizer Parameters > Air Humidity." This order is based on the physiological needs of vegetable growth: Light is the energy source for photosynthesis; temperature affects enzyme activity and metabolic rate, both of which directly determine vegetable growth efficiency and survival status, hence they are listed as core control targets. CO2 concentration affects photosynthetic efficiency, water and fertilizer parameters affect nutrient absorption, and air humidity mainly affects the risk of disease occurrence, so their priority decreases in that order. The dynamic adjustment logic is triggered when temperature or light exceeds a safe threshold. At this time, the control priority of that parameter automatically rises to first, pausing or postponing the control of other non-urgent parameters, and prioritizing the concentration of equipment resources to resolve the core parameter anomaly. For example, if the system is performing CO2 supplementation control and detects that the temperature has risen to 37℃ (exceeding the safe threshold of 36℃ for tomato fruiting), CO2 supplementation will be stopped immediately, and a powerful cooling control will be initiated with the ventilation fan running at full speed and the evaporative cooling pads fully open, until the temperature returns to within the safe threshold, and then CO2 supplementation will resume.

[0052] The default priority ensures the optimal allocation of control resources in normal scenarios, focusing on core needs; the dynamic adjustment mechanism ensures emergency response capabilities in extreme scenarios, avoiding irreversible damage to vegetables due to prioritizing the control of non-core parameters, thus balancing the optimization and safety of control.

[0053] In some specific embodiments, the regulation execution and feedback correction module is configured to control the greenhouse regulation equipment to perform corresponding operations according to the adaptive regulation strategy, verify the regulation effect by re-collecting environmental data and vegetable canopy images, and dynamically correct the regulation strategy, including: The control equipment includes LED supplemental lighting, shade netting, ventilation fans, heating pipes, wet curtain cooling system, CO2 generator, and integrated water and fertilizer device; Before executing a control operation, the equipment operating status must be checked. If there is no fault, the control strategy will be executed. If there is a fault, the backup equipment will be switched and an alarm will be triggered. If there is no backup equipment, the control level will be reduced and the control duration will be extended.

[0054] It should be understood that the accuracy and reliability of control execution are ensured through "stepless adjustment equipment + fault redundancy mechanism". Its working principle is as follows: Selecting control equipment that supports stepless adjustment allows for continuous control actions output according to the deviation level, avoiding parameter fluctuations caused by discrete adjustment; before executing the control operation, the equipment's operating status is self-checked. If no fault is found, the strategy is executed; if a fault exists, backup equipment is activated or the control scheme is adjusted to ensure uninterrupted control action. Simultaneously, an alarm is triggered to remind staff to perform maintenance, improving the system's robustness.

[0055] The control equipment covers all aspects of greenhouse environment control needs, with the following specific configuration: LED supplemental lighting power adjustment range 0-100%, single lamp power 300W, light intensity adjustment accuracy ±50μmol / m². 2 • The supplementary lighting intensity can be precisely adjusted according to the light deviation level; the shade net is electrically controlled, with an opening and closing range of 0-100%, adjustment accuracy of ±5%, and response time ≤10 seconds; the ventilation fan speed adjustment range is 0-3000 r / min, with a maximum air volume of 15000 m³ / min. 3 The ventilation intensity can be controlled by adjusting the rotation speed; the heating tube power adjustment range is 0-5kW, with a heating accuracy of ±0.5℃, adapting to the control needs of different temperature deviations; the wet curtain cooling system opening degree adjustment range is 0-100%, with a cooling efficiency of ≥80%, and can be linked with the ventilation fan to achieve rapid cooling; the CO2 generator release rate adjustment range is 0-5L / min, with a purity of ≥99.5%, which can accurately replenish CO2; the integrated water and fertilizer device supports stepless adjustment of water and fertilizer ratio, with an EC value adjustment range of 0.8-2.5ms / cm and a pH value adjustment range of 5.5-6.5, with adjustment accuracies of ±0.1ms / cm and ±0.1 respectively, which can accurately supply water and fertilizer according to the growth stage and soil parameters.

[0056] Equipment status self-check is achieved through real-time querying of equipment operating parameters (such as current, voltage, and operating feedback signals) via industrial PLC. The self-check time is ≤1 second, ensuring that the timeliness of control is not affected. If no equipment fault is detected (operating parameters are within the normal range), the corresponding operation is executed according to the control strategy. If a fault is detected (such as abnormal fan speed or failure of supplemental lighting), the system prioritizes switching to the backup equipment (each core control device is equipped with one backup device). The backup equipment startup time is ≤30 seconds, ensuring continuous control action. If there is no backup equipment (such as some small greenhouses not equipped with backup water and fertilizer devices), the control level is reduced (such as changing from heavy control to medium control) and the control duration is extended (by 50%) to avoid parameters failing to return to the optimal range due to insufficient control intensity. At the same time, the system triggers equipment fault alarms through audible and visual alarms, remote notifications, etc. The alarm information includes the name, location, and type of faulty equipment, facilitating quick location and repair by staff.

[0057] The stepless adjustment device enables precise control actions, avoiding parameter over-adjustment or under-adjustment; the fault redundancy mechanism ensures that the system can still maintain basic control functions when the equipment fails, reducing planting losses caused by equipment failure and enhancing the practical application value of the system.

[0058] In some specific embodiments, the regulation execution and feedback correction module is configured to control the greenhouse regulation equipment to perform corresponding operations according to the adaptive regulation strategy, verify the regulation effect by re-collecting environmental data and vegetable canopy images, and dynamically correct the regulation strategy, and further includes: When the deviation level is micro-deviation, data is re-collected within the preset first time threshold after adjustment; When the deviation level is small or medium, data is re-acquired within the preset second time threshold after adjustment. When the deviation level is large, data will be re-collected within the preset third time threshold after adjustment. The current strategy is maintained when the parameters revert to the optimal range and the vegetable canopy image shows no abnormalities. When parameters fail to regress, the control intensity is increased according to the deviation level. If the image displays abnormalities, immediately stop the current adjustment, lower the relevant parameters, and update the corresponding parameter range in the benchmark library.

[0059] It should be understood that the closed-loop optimization of the control effect is achieved through "dynamic feedback cycle + graded correction strategy". Its working principle is as follows: Differentiated feedback data collection cycles are set according to the deviation level; the greater the deviation, the shorter the feedback cycle, ensuring rapid tracking of the control effect. After collecting feedback data, the control effect is verified through "parameter regression judgment + image anomaly judgment". Three correction strategies—maintain, upgrade, and stop—are formulated for different verification results, forming a closed loop of "control-feedback-correction" to continuously optimize the control strategy and ensure that environmental parameters are always adapted to the vegetable growth stage.

[0060] The first preset time threshold (feedback period for minor deviations) is 30 minutes. Under minor deviations, parameter shifts are small, and parameter return after adjustment is fast. The 30-minute feedback period balances monitoring accuracy and system energy consumption. The second preset time threshold (feedback period for small and medium deviations) is 20 minutes. For these types of deviations, faster tracking of the adjustment effect is needed to avoid deviation rebound. The third preset time threshold (feedback period for large deviations) is 10 minutes. Under large deviations, parameter shifts are severe, requiring high-frequency monitoring to ensure effective adjustment and prevent continuous parameter deterioration. Feedback data collection includes two types of data: environmental parameters and vegetable canopy images. Environmental parameters are collected once per minute, and the image collection frequency is consistent with the collection frequency corresponding to the growth stage, ensuring the timeliness and completeness of the data.

[0061] The parameter regression criterion is "the current environmental parameters fall within the optimal range," and the image anomaly criterion is "there are no abnormal features such as wilting, scorching, or deformity in the vegetable canopy image, and the percentage of abnormal pixels is <5%." When the parameters return to the optimal range and the image is normal, it indicates that the control effect is satisfactory, and the current control strategy should be maintained, with subsequent monitoring at the original frequency. When the parameters do not return to the optimal range, it indicates that the control intensity is insufficient, and the control intensity should be upgraded according to the deviation level, with an upgrade range of 20%-30% of the original control intensity. For example, under small deviation, the original control of the supplemental lighting power was 50%, and after the upgrade, it should be adjusted to 60%-65%; under medium deviation, the original control of the ventilation fan speed was 2000 r / min, and after the upgrade, it should be adjusted to 2400-2600 r / min. When the image shows anomalies, it indicates that the control action is excessive (such as excessive supplemental lighting causing leaf scorching), and the current control should be stopped immediately. At the same time, the relevant parameters should be reduced by 30%, and the corresponding parameter range in the benchmark library should be updated to prevent similar anomalies from recurring. For example, if the percentage of leaf burn pixels is detected to be 7% after adjustment, the LED supplemental light is immediately turned off, restarted after 30 minutes and the power is reduced by 30%, while the upper limit of the optimal illumination range for this stage in the benchmark library is updated.

[0062] The dynamic feedback cycle ensures timely feedback at different deviation levels, preventing control failure due to slow feedback; the graded correction strategy achieves precise optimization of control effects, ensuring that parameters quickly return to the optimal range and timely correcting over-control, avoiding secondary damage to vegetables, thus forming a complete closed-loop control logic.

[0063] A second embodiment of the greenhouse environment adaptive control method based on vegetable growth stage identification, according to an embodiment of this application, includes the following steps: S100. Acquire images of the canopy of the target greenhouse vegetables. After preprocessing the images by denoising, segmentation, and target region extraction, extract the static visual features and dynamic temporal features of the vegetables. Use a multi-feature fusion recognition model combined with confidence verification to lock the current growth stage of the vegetables. The multi-feature fusion recognition model determines the stage by weighted fusion calculation of static and dynamic features. The confidence verification determines the stage locking method based on the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets the preset threshold. S200. Collect environmental data of the target greenhouse, including temperature, air humidity, light intensity, soil-related parameters, and CO2 concentration. S300: Call the preset growth stage environmental parameter benchmark library, match the monitored environmental data with the optimal parameter range corresponding to the current growth stage, calculate the parameter deviation level, and generate an adaptive control strategy according to the preset control priority. S400: Control the greenhouse control equipment to perform corresponding operations according to the adaptive control strategy, verify the control effect by re-collecting environmental data and vegetable canopy images, and dynamically correct the control strategy.

[0064] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0065] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0066] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0067] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A greenhouse environment adaptive control system based on vegetable growth stage identification, characterized in that, include: The growth stage identification module is configured to acquire images of the canopy of target greenhouse vegetables, perform preprocessing operations such as denoising, segmentation, and target region extraction on the images, extract static visual features and dynamic temporal features of the vegetables, and use a multi-feature fusion identification model combined with confidence verification to determine the current growth stage of the vegetables. The multi-feature fusion identification model determines the stage by weighted fusion calculation of static and dynamic features, and the confidence verification determines the stage locking method based on the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets a preset threshold. An environmental parameter monitoring module is configured to collect environmental data of the target greenhouse, including temperature, air humidity, light intensity, soil-related parameters, and CO2 concentration. The regulation strategy generation module is configured to call a preset growth stage environmental parameter benchmark library, match the monitored environmental data with the optimal parameter range corresponding to the current growth stage, calculate the parameter deviation level, and generate an adaptive regulation strategy according to a preset regulation priority. The regulation execution and feedback correction module is configured to control the greenhouse regulation equipment to perform corresponding operations according to the adaptive regulation strategy, verify the regulation effect by re-collecting environmental data and vegetable canopy images, and dynamically correct the regulation strategy.

2. The greenhouse environment adaptive control system based on vegetable growth stage identification according to claim 1, characterized in that, The growth stage identification module is configured to acquire images of the target greenhouse vegetable canopy, perform preprocessing operations such as denoising, segmentation, and target region extraction on the images, extract static visual features and dynamic temporal features of the vegetables, and use a multi-feature fusion identification model combined with confidence verification to determine the current growth stage of the vegetables. The multi-feature fusion identification model determines the stage by weighted fusion calculation of static and dynamic features. The confidence verification determines the stage locking method based on the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets a preset threshold, including: The vegetable growth stages include the seedling stage, seedling stage, flowering stage, fruiting stage, and maturity stage; When the number of target vegetable leaves is less than or equal to a preset first quantity threshold, and the growth rate of the target vegetable leaf area is greater than or equal to a preset first growth rate threshold, it is determined to be the seedling stage; When the number of target vegetable leaves is greater than the first quantity threshold and less than the preset second quantity threshold, and the growth rate of the target vegetable plant height is greater than or equal to the preset second growth rate, it is determined to be the seedling stage; When the number of target vegetable flowers detected is greater than or equal to a preset third quantity threshold, and the confidence level is greater than or equal to a preset first threshold, it is determined to be the flowering period; When the diameter of the target vegetable fruit is detected to be greater than or equal to a preset second threshold, it is determined to be the result period; When the percentage of colored pixels in the target vegetable fruit is greater than or equal to a preset third threshold, and the leaf area growth rate is less than or equal to a preset third growth rate threshold, it is determined to be the maturity period.

3. The greenhouse environment adaptive control system based on vegetable growth stage identification according to claim 2, characterized in that, The growth stage identification module is configured to acquire images of the canopy of target greenhouse vegetables, perform preprocessing operations such as denoising, segmentation, and target region extraction on the canopy images, extract static visual features and dynamic temporal features of the vegetables, and use a multi-feature fusion identification model combined with confidence verification to determine the current growth stage of the vegetables. The multi-feature fusion identification model determines the stage by weighted fusion calculation of static and dynamic features. The confidence verification determines the stage locking method based on the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets a preset threshold. The module also includes: Images of the greenhouse vegetable canopy are captured using a high-definition RGB camera, with the acquisition frequency dynamically adjusted according to the growth stage, including: When the target vegetable is in the seedling stage or the mature seedling stage, the sampling frequency is a preset first frequency value; When the target vegetable is in its flowering, fruiting, or ripening stage, the sampling frequency is a preset second frequency value; After the data collection is complete, a three-layer filtering process is performed, including: When the standard deviation of the image grayscale value is greater than or equal to the preset first standard deviation threshold, the sharpness meets the standard. When the proportion of vegetable canopy pixels in the image is greater than or equal to a preset first proportion threshold, the target area meets the standard. When the percentage of occluded pixels in an image is less than or equal to a preset second percentage threshold, the occlusion is considered acceptable. When the sharpness, target area, and occlusion of the acquired image all meet the standards, it is a valid image; otherwise, the camera is triggered to re-acquire the image. If the number of resampling attempts equals the preset third threshold and no valid image is obtained, the previous recognition result is retained.

4. The greenhouse environment adaptive control system based on vegetable growth stage identification according to claim 3, characterized in that, The growth stage identification module is configured to acquire images of the canopy of target greenhouse vegetables, perform preprocessing operations such as denoising, segmentation, and target region extraction on the canopy images, extract static visual features and dynamic temporal features of the vegetables, and use a multi-feature fusion identification model combined with confidence verification to determine the current growth stage of the vegetables. The multi-feature fusion identification model determines the stage by weighted fusion calculation of static and dynamic features. The confidence verification determines the stage locking method based on the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets a preset threshold. The module also includes: The static visual features include the number of leaves, leaf area, plant height, leaf color, and number of flowers and fruits. The number of leaves is extracted by connected component counting and only leaves with an area greater than or equal to a preset fourth threshold are counted. The leaf area is calculated by combining the pixel ratio with the camera calibration coefficient. The dynamic time-series features include the daily growth rate of leaf area and the daily growth rate of plant height calculated based on image sequences within a preset first time threshold, and abnormal data exceeding a preset multiple standard deviation must be removed when calculating the growth rate.

5. The greenhouse environment adaptive control system based on vegetable growth stage identification according to claim 4, characterized in that, The growth stage identification module is configured to acquire images of the canopy of target greenhouse vegetables, perform preprocessing operations such as denoising, segmentation, and target region extraction on the canopy images, extract static visual features and dynamic temporal features of the vegetables, and use a multi-feature fusion identification model combined with confidence verification to determine the current growth stage of the vegetables. The multi-feature fusion identification model determines the stage by weighted fusion calculation of static and dynamic features. The confidence verification determines the stage locking method based on the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets a preset threshold. The module also includes: The multi-feature fusion recognition model is a fusion model of ResNet50 and Transformer. The model training dataset contains a preset number of labeled images of different vegetables and different growth stages, and the recognition accuracy is greater than or equal to a preset fifth threshold. When the model output confidence level is greater than or equal to the preset sixth threshold, the current growth stage is directly locked. When the model output confidence is less than the sixth threshold and greater than or equal to the preset seventh threshold, the stage with the highest frequency among the last three recognition results is selected. When the model output confidence level is less than the seventh threshold, the previous valid identification result is retained and a manual review reminder is triggered.

6. The greenhouse environment adaptive control system based on vegetable growth stage identification according to claim 5, characterized in that, The regulation strategy generation module is configured to call a preset growth stage environmental parameter benchmark library, match the monitored environmental data with the optimal parameter range corresponding to the current growth stage, calculate the parameter deviation level, and generate an adaptive regulation strategy according to a preset regulation priority, including: The preset growth stage environmental parameter benchmark library contains parameters for each growth stage of tomatoes, cucumbers, and lettuce, and each parameter is divided into three levels: optimal range, warning range, and safety threshold. If the same vegetable variety still shows abnormal growth after three consecutive adjustments at the same growth stage, the upper limit of the optimal range of the parameter will be automatically lowered or the lower limit will be adjusted upwards.

7. The greenhouse environment adaptive control system based on vegetable growth stage identification according to claim 6, characterized in that, The regulation strategy generation module is configured to call a preset growth stage environmental parameter benchmark library, match the monitored environmental data with the optimal parameter range corresponding to the current growth stage, calculate the parameter deviation level, and generate an adaptive regulation strategy according to a preset regulation priority, including: The deviation level is determined by deviation rate calculation. When the deviation rate is less than or equal to a preset first deviation threshold, the deviation level is a minor deviation. When the deviation rate is greater than the first deviation threshold and less than or equal to the preset second deviation threshold, the deviation level is small deviation; When the deviation rate is greater than the second deviation threshold and less than or equal to the preset third deviation threshold, the deviation level is medium deviation; When the deviation rate is greater than the third deviation threshold, the deviation level is large deviation; The control priority is sorted by light intensity, temperature, CO2 concentration, water and fertilizer parameters, and air humidity by default. When the temperature or light intensity exceeds the safety threshold, its control priority automatically rises to first.

8. The greenhouse environment adaptive control system based on vegetable growth stage identification according to claim 7, characterized in that, The regulation execution and feedback correction module is configured to control the greenhouse regulation equipment to perform corresponding operations according to the adaptive regulation strategy, verify the regulation effect by re-collecting environmental data and vegetable canopy images, and dynamically correct the regulation strategy, including: The control equipment includes LED supplemental lighting, shade netting, ventilation fans, heating pipes, wet curtain cooling system, CO2 generator, and integrated water and fertilizer device; Before executing a control operation, the equipment operating status must be checked. If there is no fault, the control strategy will be executed. If there is a fault, the backup equipment will be switched and an alarm will be triggered. If there is no backup equipment, the control level will be reduced and the control duration will be extended.

9. A greenhouse environment adaptive control system based on vegetable growth stage identification according to claim 8, characterized in that, The regulation execution and feedback correction module is configured to control the greenhouse regulation equipment to perform corresponding operations according to the adaptive regulation strategy, verify the regulation effect by re-collecting environmental data and vegetable canopy images, and dynamically correct the regulation strategy. It also includes: When the deviation level is micro-deviation, data is re-collected within the preset first time threshold after adjustment; When the deviation level is small or medium, data is re-acquired within the preset second time threshold after adjustment. When the deviation level is large, data will be re-collected within the preset third time threshold after adjustment. The current strategy is maintained when the parameters revert to the optimal range and the vegetable canopy image shows no abnormalities. When parameters fail to regress, the control intensity is increased according to the deviation level. If the image displays abnormalities, immediately stop the current adjustment, lower the relevant parameters, and update the corresponding parameter range in the benchmark library.

10. A method for adaptive control of greenhouse environment based on vegetable growth stage identification, characterized in that, An adaptive greenhouse environment control system based on vegetable growth stage identification, as described in any one of claims 1 to 9, comprises the following steps: Images of the canopy of target greenhouse vegetables are acquired. After preprocessing the images by denoising, segmentation, and target region extraction, static visual features and dynamic temporal features of the vegetables are extracted. A multi-feature fusion recognition model combined with confidence verification is used to determine the current growth stage of the vegetables. The multi-feature fusion recognition model determines the stage by weighted fusion calculation of static and dynamic features. The confidence verification determines the stage locking method based on the confidence value range output by the model, and outputs the final growth stage result only when the confidence level meets a preset threshold. Collect environmental data of the target greenhouse, including temperature, air humidity, light intensity, soil-related parameters, and CO2 concentration; The system calls a preset growth stage environmental parameter benchmark library, matches the monitored environmental data with the optimal parameter range corresponding to the current growth stage, calculates the parameter deviation level, and generates an adaptive control strategy according to the preset control priority. The greenhouse control equipment is controlled to perform corresponding operations according to the adaptive control strategy. The control effect is verified by re-collecting environmental data and vegetable canopy images, and the control strategy is dynamically corrected.

Images