Intelligent control system for breeding environment of forest frog

Abstract

The application provides a forest frog breeding environment intelligent regulation and control system, and relates to the technical field of Internet of Things control systems related to agricultural products.The technical solution points are as follows: a multi-dimensional perception module, an AI decision processing module, an execution module, and a remote interaction module are included.The multi-dimensional perception module is in communication connection with the AI decision processing module.The AI decision processing module is connected with the execution module and the remote interaction module respectively.The multi-dimensional perception module collects environmental parameter data, forest frog state data, external interference data, and micro-ecological data.The AI decision processing module generates specific regulation and control instructions based on historical data and real-time data.The regulation and control instructions at least include emergency relief instructions, temperature and humidity regulation and control instructions, light intensity regulation and control instructions, and feeding instructions.The forest frog breeding environment intelligent regulation and control system has the advantages of multi-dimensional perception, intelligent prediction of forest frog breeding environments, and real-time accurate and refined regulation and control.

Description

Technical Field

[0001] This application relates to the field of program control system technology, and more specifically, to an Internet of Things (IoT) control system related to agricultural products. Background Technology

[0002] As an economically valuable species with medicinal, edible, and ecological significance, the forest frog exhibits extremely high sensitivity to environmental parameters such as temperature, humidity, water quality, and light during its growth and development. The requirements differ significantly at different growth stages (tadpole, juvenile, and adult stages), and the precision of environmental control directly determines the survival rate and product quality. While current forest frog farming environmental control technology has achieved initial mechanization, many bottlenecks remain to be addressed, hindering its ability to meet the demands of large-scale, refined farming.

[0003] Existing technologies generally suffer from the problem of singular parameter control, focusing primarily on basic temperature and humidity regulation while neglecting the coordinated control of key parameters such as water quality (e.g., dissolved oxygen, ammonia nitrogen) and light intensity. This leads to an imbalance in the breeding environment, making frogs susceptible to diseases. Furthermore, the control response is lagging, relying on periodic manual inspections to collect data, failing to capture environmental fluctuations in real time. By the time abnormal parameters are detected, irreversible damage has already been done to frog growth.

[0004] Furthermore, existing systems lack a linkage mechanism with the physiological state of forest frogs, fail to adjust control strategies based on biological characteristics such as frog activity levels and population density, and cannot predict external disturbances such as predator invasions and extreme weather, resulting in insufficient targeted and forward-looking control. These problems lead to low survival rates, unstable growth cycles, and high breeding costs in forest frog farming, hindering the high-quality development of the forest frog farming industry. Therefore, there is an urgent need for an environmental control system with multi-dimensional perception, intelligent prediction, and precise control capabilities.

[0005] The above problems urgently need to be addressed. Summary of the Invention

[0006] The purpose of this application is to provide an intelligent control system for the breeding environment of forest frogs, which has the advantages of multi-dimensional perception, intelligent prediction and real-time precise control of the breeding environment of forest frogs.

[0007] This application provides an intelligent control system for the breeding environment of forest frogs, the technical solution of which is as follows: It includes a multi-dimensional perception module, an AI decision processing module, an execution module, and a remote interaction module. The multi-dimensional perception module is communicatively connected to the AI ​​decision processing module, and the AI ​​decision processing module is connected to both the execution module and the remote interaction module. The multi-dimensional sensing module collects environmental parameter data, frog status data, external disturbance data, and micro-ecological data. The AI ​​decision processing module generates specific control instructions based on historical and real-time data. The control instructions include at least one of the following: emergency relief instructions, temperature and humidity control instructions, light intensity control instructions, and feeding instructions. The execution module executes specific control commands; The remote interaction module provides data visualization and manual intervention control commands based on the established cloud platform.

[0008] Furthermore, in this application, the multi-dimensional sensing module includes an environmental parameter sensing submodule, a frog state sensing submodule, an external disturbance sensing submodule, and a micro-ecological sensing submodule; The environmental parameter sensing submodule collects environmental parameter data; The frog status perception submodule collects frog status data; The external interference sensing submodule collects data on extreme weather and natural enemy interference. The microecological sensing submodule collects bacterial colony data in the aquaculture pond water.

[0009] Furthermore, in this application, the environmental parameter sensing submodule includes an infrared temperature sensor unit, a humidity sensor unit, a light sensor unit, and a water quality multi-parameter sensor unit. The infrared temperature sensor unit collects temperature data, the humidity sensor unit collects humidity data, the light sensor unit collects light intensity data, and the water quality multi-parameter sensor unit collects water quality data including pH and dissolved oxygen in the aquaculture pond. The frog state perception submodule includes a high-definition infrared camera unit, a biometric recognition unit, and a non-invasive bioelectric sensor unit. The high-definition infrared camera unit collects behavioral characteristic image data of the frog, and the non-invasive bioelectric sensor unit collects skin potential signal data of the frog. The external interference sensing submodule includes an acoustic vibration sensor unit, which collects characteristic vibration sound waves of airflow and air pressure as precursors to extreme weather and characteristic vibration sound wave data of predator activities. The microecological sensing submodule includes a microbial sensor unit, which collects data on the community ratio of beneficial and harmful bacteria in the aquaculture pond water.

[0010] Furthermore, in this application, the biometric identification unit operates on the image data acquired by the high-definition infrared camera unit: Target detection: The biometric recognition unit accurately locates individual forest frogs from the image using a biometric recognition algorithm, as shown in the formula: Confidence=P obj ×IOU; Where Confidence represents the target confidence level, used to determine the probability of whether it is a forest frog, and its value ranges from [0, 1]; P obj The probability of the target being predicted by the model is represented by IOU, which represents the intersection-union ratio between the predicted bounding box and the actual frog area. IOU is used to measure the localization accuracy and has a value range of [0, 1]. Feature extraction: Activity level is calculated based on the changes in the center coordinates of the frog in different frames. The activity level is used to determine whether the frog is in a stressed or healthy state. The formula is as follows: ; Where Activity represents the average activity level of the frog group, with a value range of [0, 1], where 0 represents stillness and 1 represents vigorous movement; N represents the number of valid frogs identified in a single frame image; (x i，t y i，t (x) represents the center coordinates (in pixels) of the i-th frog in frame t; i，t-1 y i，t-1 ) represents the center coordinates of the i-th frog in frame t-1, and T represents the calculation time window (set to 5s, corresponding to a camera frame rate of 10fps, for a total of 50 frames); D scale This represents the coordinate scaling factor (converting pixel distance to actual distance, e.g., 1 pixel = 0.5 cm, D). scale =50 pixels / m); Feature extraction: The cluster density reflects the degree of adaptation of forest frogs to the environment (e.g., clustering for warmth in low temperatures, dispersing to escape when water quality is abnormal). The formula is as follows: ; Where Density represents the density of frog colonies (unit: frogs / m², value range [0, 20], suitable for the carrying capacity of the breeding pond); N represents the effective number of frogs; R represents the radius of the monitoring area of ​​the breeding pond; M represents the number of pairwise frogs (M=N×(N-1) / 2); d i，j L represents the actual distance (m) between the i-th and j-th frogs. th represents the cluster distance threshold; I(·) represents the exponential function (takes 1 if the condition in parentheses is met, otherwise takes 0); Output frog status data: Combine activity level and cluster density to generate a frog status index, which is then used by the AI ​​decision-making module. The formula is as follows: State=α×Activity+β×(Density / D max ); Where State represents the overall state index of the forest frog (range [0, 1]); α and β represent weighting coefficients (α = 0.6, β = 0.4, determined through training with historical data, prioritizing the immediate state of activity response); D maxThis indicates the maximum carrying capacity of the aquaculture pond (animals / m²).

[0011] Furthermore, in this application, the AI ​​decision processing module includes an improved LSTM fusion prediction model submodule and a stress assessment submodule, and the execution module includes a stress relief submodule; The improved LSTM fusion prediction model submodule uses a three-level mathematical model of feature standardization, weighted fusion, and stress level quantization to unify and quantify heterogeneous data of skin potential signals in the behavioral dimension (State) and physiological dimension, and outputs a comprehensive emergency feature value that can be directly used by the stress assessment submodule. Feature standardization: Core features are extracted and standardized from the acquired skin potential signals to ensure consistency with the dimensional range [0, 1] of the State in the behavior dimension. The formula is as follows: ; Where V RMS The root mean square value of the skin potential signal (unit: mV) represents the average energy of the potential fluctuation; T represents the number of sampling points within the calculation window (sampling frequency 1kHz, window 5s, therefore T=5000); v k This represents the original value (mV) of the skin point at the kth sampling point. V RMS Mapping to the interval [0, 1] eliminates dimensional differences; the formula is: V norm =(V RMS -V min ) / (V max -V min ); Where V norm V represents the standardized skin potential characteristic value (range [0, 1]); min V represents the minimum RMS skin potential of a frog under normal conditions; max This represents the maximum RMS skin potential of a forest frog under severe stress. Weighted fusion: A dynamic weighted summation model is used to fuse physiological features V. norm The weights of the behavioral characteristic State are dynamically adjusted according to the growth stage of the forest frog, and the formula is as follows: S fusion =ω1×V norm +ω2×State; ω1 + ω2 = 1; Where S fusion ω1 represents the comprehensive stress characteristic value (range [0, 1]), which is directly input into the stress assessment submodule; ω2 represents the weight coefficient of skin potential (physiological index weight, reflecting the immediacy of stress); ω3 represents the weight coefficient of behavioral index (behavioral index weight, reflecting the persistence of stress). Stress level quantification: The stress assessment submodule will quantify S fusion The mapping is divided into three stress levels, providing clear instructions for the precise execution layer. The formula is as follows: ; Where Stress_Level represents the stress level (0=normal, 1=mild, 2=moderate / severe); θ1 represents the mild stress threshold; θ2 represents the moderate / severe stress threshold; The stress assessment submodule sends stress relief instructions to the stress relief submodule based on the stress level, and the stress relief submodule executes the stress relief instructions.

[0012] Furthermore, in this application, the stress relief submodule includes a low-frequency sound wave generator unit (20~30Hz) and a plant essential oil slow-release device unit; The stress relief submodule outputs instructions based on the stress level: when Stress_Level=1, it outputs a "basic relief instruction", which is activated by the low-frequency sound wave generator unit to release low-frequency sound waves; when Stress_Level=2, it outputs an "enhanced relief instruction", which simultaneously activates the low-frequency sound wave generator unit and the plant essential oil slow-release device unit to release low-frequency sound waves and plant essential oils, and simultaneously triggers the temperature and humidity control unit to slow down the rate of change of environmental parameters.

[0013] Furthermore, in this application, the improved LSTM fusion prediction model submodule uses historical data of aquaculture environment (over 3 years), frog growth cycle parameters, and real-time sensing data as the training set. It dynamically adjusts the weights according to the frog's regional species to generate standardized control command values ​​and destandardized actual control command values. The formula is as follows: Y=W o H att +b o ; ; Where Y represents the standardized control command value ([0, 1]); W o / b o These represent the weights / biases of the fully connected layer, respectively; H att Y represents the attention-weighted hidden layer output (integrating temporal and growth cycle features); reg This represents the actual control command value after destandardization (e.g., temperature 20℃, humidity 65%RH). This represents the inverse standard function (reduced to the actual physical quantity).

[0014] Furthermore, in this application, the execution module further includes a security protection submodule, and the external interference sensing submodule is linked to the security protection submodule; When the external interference sensing submodule detects an approaching predator, the safety protection submodule triggers the ultrasonic repelling device to release ultrasonic waves to repel the predator; when the external interference sensing submodule predicts extreme weather, the safety protection submodule controls the automatic closing device of the windproof and heat-insulating shed to close the windproof and heat-insulating shed before the extreme weather.

[0015] Furthermore, in this application, the AI ​​decision processing module further includes a micro-ecological balance sub-module, and the execution module further includes a water quality control sub-module; The microecological sensing submodule detects the community ratio of beneficial and harmful bacteria in the water body through a microbial sensor unit; the microecological balance submodule predicts the risk of water quality deterioration based on the community ratio data; and the water quality regulation submodule adds corresponding probiotics based on the prediction results of the microecological balance submodule.

[0016] Furthermore, in this application, the execution module further includes a temperature and humidity control submodule, a shading and supplemental lighting submodule, and a feeding linkage submodule. The temperature and humidity control submodule executes the temperature and humidity control instructions issued by the AI ​​decision processing module, the shading and supplemental lighting submodule executes the light intensity control instructions issued by the AI ​​decision processing module, and the feeding linkage submodule executes the feeding instructions issued by the AI ​​decision processing module. The remote interaction module automatically generates growth reports and equipment maintenance reminders through the cloud platform and pushes them to the user's terminal via the APP.

[0017] Other features and advantages of this application will be set forth in the following description and will be apparent in part from the description or may be learned by practicing embodiments of this application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings. Beneficial effects

[0018] ① Comprehensive perception coverage, solidifying the data foundation for intelligent regulation: Through four perception sub-modules—environment, frog status, external interference, and micro-ecology—and combined with dedicated hardware such as infrared thermometry, bioelectric sensors, and microbial sensors, it achieves comprehensive data collection of temperature and humidity, water quality physicochemical indicators, frog behavior / physiological characteristics, predator / extreme weather signals, and the proportion of microbial communities in the water. This breaks through the limitations of traditional systems that rely on "single and partial data collection," providing comprehensive and accurate data source support for AI decision-making. ② Precise quantification of frog behavior and intelligent stress relief to improve breeding adaptability: The activity level, cluster density and comprehensive status index of forest frogs are quantified by biometric recognition algorithms. Combined with a two-dimensional fusion model of skin potential signals, the stress state can be accurately determined. With the addition of a non-invasive relief solution using low-frequency sound waves and plant essential oils, a closed loop of "frog behavior perception - stress assessment - precise relief" is formed, which solves the problems of traditional regulation ignoring the physiological state of forest frogs and blind stress intervention, and ensures the stable growth of forest frogs. ③ AI-driven decision-making is precise and adaptable to the needs of multi-scenario aquaculture: The improved LSTM fusion prediction model sub-module uses more than 3 years of historical data, growth cycle parameters and real-time data as training sets. It can dynamically adjust the weights according to the regional species of forest frogs and generate targeted control instructions, breaking through the traditional "one-size-fits-all" control mode. At the same time, through feature standardization, weighted fusion and other algorithms, it realizes the unified quantification of heterogeneous data, improves the scientific nature and adaptability of decision-making, and adapts to the needs of forest frog aquaculture in different regions and at different growth stages. ④ Multi-dimensional proactive risk prevention and control to reduce aquaculture losses: By linking external interference perception with safety protection sub-modules, it can respond in advance to predator invasion and extreme weather; by using microecological perception and precise addition of probiotics, it can predict and control the risk of water quality deterioration; combined with real-time intervention in stress states, it can build a full-scenario risk prevention and control system of "predators-extreme weather-water quality-frog state" to solve the pain points of traditional aquaculture of "passive risk response and high loss rate"; ⑤ Precise zoning and convenient management enhance the practicality of aquaculture: The execution module covers multiple functions such as temperature and humidity control, shading and supplemental lighting, feeding, stress relief, and water quality control, enabling personalized zoning control of the aquaculture area to meet the differentiated needs of forest frogs at different growth stages; the remote interaction module generates growth reports and equipment maintenance reminders through the cloud platform and pushes them to the APP, breaking spatial limitations, supporting unattended operation and remote manual intervention, reducing aquaculture management costs, and improving operational convenience. Attached Figure Description

[0019] Figure 1 This application provides an architecture diagram of an intelligent control system for the breeding environment of forest frogs. Figure 2 This is an architecture diagram of the multi-dimensional perception module in this application; Figure 3 This is an architecture diagram of the environmental parameter sensing submodule in this application; Figure 4 This is an architecture diagram of the frog state perception submodule in this application; Figure 5 This is an architecture diagram of the external interference sensing submodule of this application; Figure 6 This is an architecture diagram of the micro-ecological sensing submodule of this application; Figure 7 This is an architecture diagram of the AI ​​decision processing module in this application; Figure 8 This is an architecture diagram of the execution module of this application; Figure 9 This is the architecture diagram of the stress relief submodule in this application. Detailed Implementation

[0020] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0021] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0022] Please refer to Figures 1 to 9 As shown, this application provides an intelligent control system for the breeding environment of forest frogs, including a multi-dimensional sensing module, an AI decision processing module, an execution module, and a remote interaction module. The multi-dimensional sensing module and the AI ​​decision processing module are communicatively connected, and the AI ​​decision processing module is connected to both the execution module and the remote interaction module. The multi-dimensional sensing module collects environmental parameter data, forest frog status data, external disturbance data, and microecological data. The AI ​​decision processing module generates specific control instructions based on historical and real-time data. These control instructions include at least one of the following: emergency relief instructions, temperature and humidity control instructions, light intensity control instructions, and feeding instructions. The execution module executes the specific control instructions.

[0023] Specifically, the system includes a multi-dimensional perception module, an AI decision-making and processing module, an execution module, and a remote interaction module. The multi-dimensional perception module and the AI ​​decision-making and processing module are communicatively connected, and the AI ​​decision-making and processing module is connected to both the execution module and the remote interaction module. This design clarifies the core architecture of the system, addressing the fragmented structure of existing systems by breaking down functional modules (perception, decision-making, execution, etc.). Each module performs its specific function while also coordinating and linking together, providing hardware and logical support for "intelligent regulation." The multi-dimensional perception module collects environmental parameter data, frog status data, external disturbance data, and micro-ecological data. The AI ​​decision-making and processing module generates specific regulation commands based on historical and real-time data. These commands include at least one of the following: emergency relief commands, temperature and humidity control commands, light intensity control commands, and feeding commands. The execution module executes these specific regulation commands. The remote interaction module provides data visualization and manual intervention functions based on a cloud platform. The multi-dimensional perception module provides the data source for decision-making, while the AI ​​decision-making and processing module acts as the "central hub," issuing commands to ensure accurate transmission and avoid response delays caused by module disconnections in existing technologies. The multi-dimensional sensing module includes sub-modules for environmental parameter sensing, frog status sensing, external disturbance sensing, and microecological sensing. This module clarifies the detailed functions of the multi-dimensional sensing module, overcoming the limitation of existing technologies that only measure environmental parameters. By collecting data from multiple dimensions (environment, frog status, disturbance, and microecology), it provides a comprehensive basis for AI decision-making, solving the problem of low matching between regulation and frog needs. The AI ​​decision processing module generates regulation instructions based on historical and real-time data. This module clarifies the core technology and functions of the AI ​​decision processing module. Through quantitative analysis of historical and real-time data, it generates precise regulation instructions, overcoming the blindness of existing technologies' "experience-based regulation." The execution module translates AI instructions into specific actions, specifically regulating key elements such as temperature, humidity, water quality, and safety. Zonal regulation adapts to the needs of different aquaculture areas, avoiding environmental fluctuations caused by the "one-size-fits-all" regulation of existing technologies. The remote interaction module provides data visualization and manual intervention regulation instructions based on a dedicated cloud platform. This module clarifies the core carrier and function of human-computer interaction, improving the system's practicality. The cloud platform enables real-time data monitoring and report generation, and the manual intervention and control command function combines intelligence and human intervention to solve the problems of "inconvenient operation and difficulty in unattended operation" in the existing system.

[0024] In some preferred embodiments, the multi-dimensional sensing module includes an environmental parameter sensing submodule, a frog state sensing submodule, an external disturbance sensing submodule, and a microecological sensing submodule; the environmental parameter sensing submodule collects environmental parameter data; the frog state sensing submodule collects frog state data; the external disturbance sensing submodule collects extreme weather and predator interference data; and the microecological sensing submodule collects bacterial colony data in the aquaculture pond water.

[0025] Specifically, the multi-dimensional sensing module includes an environmental parameter sensing submodule, a frog state sensing submodule, an external disturbance sensing submodule, and a microecological sensing submodule. The environmental parameter sensing submodule collects environmental parameter data; the frog state sensing submodule collects frog state data; the external disturbance sensing submodule collects data on extreme weather and predator interference; and the microecological sensing submodule collects bacterial colony data from the aquaculture pond. This clarifies the subdivided functional dimensions of the module, transforming the abstract concept of "multi-dimensional sensing" into four concrete submodules: environmental parameters, frog state, external disturbance, and microecology. This lays the foundation for subsequent definition of each... The specific sensors and detection indicators of the sub-modules provide the logical foundation; breaking through the single limitation of existing technologies that "only focus on environmental parameter collection", by adding the perception dimensions of "frog status, external interference, and micro-ecology", it achieves full-element data coverage of the breeding scenario, solving the problem of low control accuracy caused by incomplete data in traditional systems; laying the data foundation for the system's "perception-decision-execution" closed loop, the four sub-modules respectively provide environmental adaptation basis, frog physiological needs basis, risk warning basis, and water health basis, ensuring that the AI ​​decision processing layer can generate control instructions that fit the growth and breeding safety of frogs.

[0026] In some preferred embodiments, the environmental parameter sensing submodule includes an infrared temperature sensor unit, a humidity sensor unit, a light sensor unit, and a water quality multi-parameter sensor unit. The infrared temperature sensor unit collects temperature data, the humidity sensor unit collects humidity data, the light sensor unit collects light intensity data, and the water quality multi-parameter sensor unit collects water quality data including pH and dissolved oxygen in the aquaculture pond. The frog status sensing submodule includes a high-definition infrared camera unit, a biometric identification unit, and a non-invasive bioelectric sensor unit. The high-definition infrared camera unit collects behavioral characteristic image data of the frog, and the non-invasive bioelectric sensor unit collects skin potential signal data of the frog. The external interference sensing submodule includes a sound vibration sensor unit, which collects characteristic vibration sound waves of airflow and air pressure preceding extreme weather and characteristic vibration sound waves of predator activity. The microecological sensing submodule includes a microbial sensor unit, which collects data on the community ratio of beneficial and harmful bacteria in the aquaculture pond water.

[0027] Specifically, the environmental parameter sensing submodule includes an infrared temperature sensor unit, a humidity sensor unit, a light sensor unit, and a water quality multi-parameter sensor unit. The infrared temperature sensor unit collects temperature data, the humidity sensor unit collects humidity data, the light sensor unit collects light intensity data, and the water quality multi-parameter sensor unit collects water quality data, including pH and dissolved oxygen in the aquaculture pond. This clarifies the hardware composition and specific data types of environmental parameter acquisition, achieving full-dimensional data coverage of temperature, humidity, light, and water quality (core physicochemical indicators), providing comprehensive and accurate basic data support for the AI ​​decision-making module to formulate environmental control instructions.

[0028] The frog state perception submodule includes a high-definition infrared camera unit, a biometric recognition unit, and a non-invasive bioelectric sensor unit. The high-definition infrared camera unit collects behavioral characteristic image data of the frog, while the non-invasive bioelectric sensor unit collects skin potential signal data of the frog. This non-invasive acquisition scheme defines the frog's state and clarifies the dual-dimensional data acquisition path of behavioral characteristics (images) and physiological characteristics (skin potential), providing core biological data for stress assessment and growth status quantification, and supporting precise regulation "centered on the frog".

[0029] The external interference sensing submodule includes an acoustic vibration sensor unit, which collects characteristic vibration sound waves of airflow and air pressure as precursors to extreme weather and characteristic vibration sound wave data of predator activity. This clarifies the hardware and target data for external interference collection, accurately captures two types of core risk signals (precursors to extreme weather and predator activity), and provides a triggering basis for the early response of the safety protection submodule (closing of windproof and heat-insulating sheds, ultrasonic avoidance), thereby achieving proactive risk prevention and control.

[0030] The microbial ecosystem sensing submodule includes a microbial sensor unit, which collects data on the community ratio of beneficial and harmful bacteria in the aquaculture pond water. This dedicated data collection hardware and core data for the microbial ecosystem directly obtains key indicators of microbial community balance, filling the gap in traditional water quality testing that "emphasizes physicochemical properties while neglecting ecology." It provides core input data for the microbial ecosystem balance sub-model to predict water quality deterioration risks and formulate probiotic addition strategies.

[0031] For example, the infrared temperature sensor unit has an accuracy of ±0.1℃. By limiting the accuracy to ±0.1℃, the problem of large temperature acquisition errors in existing technologies is solved. Forest frogs are sensitive to temperature at different growth stages (e.g., the suitable water temperature for tadpoles is 18-22℃). Accurate temperature data provides a reliable basis for the AI ​​decision-making module, avoiding inaccurate regulation due to temperature deviations (e.g., overheating causing stress in tadpoles), and ensuring that temperature regulation accuracy matches the growth needs of forest frogs. The humidity sensor unit has a sampling interval of no more than 5 seconds, overcoming the lag in humidity data in existing technologies through short-interval sampling. Humidity in the breeding environment easily fluctuates with the operation of the atomizer and changes in ventilation. Real-time sampling within 5 seconds can capture instantaneous humidity changes, ensuring that the AI ​​decision-making module generates adjustment instructions in a timely manner, avoiding sudden increases or decreases in humidity that could cause dehydration or mold growth in forest frogs, and adapting to the continuous high-humidity requirements of forest frogs. The multi-parameter water quality sensor unit simultaneously detects pH, dissolved oxygen, and ammonia nitrogen content to define the specific detection range of the multi-parameter water quality sensor, and has the characteristic of parallel acquisition of multiple indicators. pH affects the activity of metabolic enzymes in forest frogs, dissolved oxygen is related to respiratory safety, and excessive ammonia nitrogen (a product of excrement decomposition) can easily lead to poisoning. Simultaneous detection of these three factors can comprehensively reflect the water quality status, providing complete data support for the micro-ecological balance sub-module of the AI ​​decision-making module, avoiding the one-sidedness of detecting a single water quality indicator, and ensuring targeted water quality control.

[0032] In some preferred embodiments, the biometric identification unit operates on the image data acquired by the high-definition infrared camera unit: Target detection: The biometric recognition unit accurately locates individual forest frogs in an image using a biometric recognition algorithm. The formula is as follows: Confidence=P obj ×IOU; Where Confidence represents the target confidence level, used to determine the probability of whether it is a forest frog, and its value ranges from [0, 1]; P obj The probability of the target being predicted by the model is represented by IOU, which represents the intersection-union ratio between the predicted bounding box and the actual frog area. IOU is used to measure the localization accuracy and has a value range of [0, 1]. Application Notes: A confidence threshold of Threshold=0.7 (optimized based on real-world testing in aquaculture scenarios) is set, retaining only targets with a confidence ≥ 0.7 as valid frogs, filtering out low-confidence interference. The infrared camera function still works in low-light conditions; the algorithm enhances frog contour recognition through grayscale differences, ensuring P... obj Adaptability to nighttime images.

[0033] Feature extraction: Activity level is calculated based on the changes in the center coordinates of the frog in different frames. The activity level is used to determine whether the frog is in a stressed or healthy state. The formula is as follows: ; Where Activity represents the average activity level of the frog group, with a value range of [0, 1], where 0 represents stillness and 1 represents vigorous movement; N represents the number of valid frogs identified in a single frame image; (x i，t y i，t (x) represents the center coordinates (in pixels) of the i-th frog in frame t; i，t-1 y i，t-1 ) represents the center coordinates of the i-th frog in frame t-1, and T represents the calculation time window (set to 5s, corresponding to a camera frame rate of 10fps, for a total of 50 frames); D scale This represents the coordinate scaling factor (converting pixel distance to actual distance, e.g., 1 pixel = 0.5 cm, D). scale =50 pixels / m); Application Notes: If Activity∈[0.05, 0.3], it is considered a normal state; Activity>0.6 indicates a stress state (such as violent swimming caused by sudden environmental changes), triggering the stress assessment submodule; the formula avoids misjudgment caused by abnormal jumping of a single frog (such as struggling due to illness) by using "group average", thus improving data reliability.

[0034] Calculating cluster density: Cluster density reflects the degree of adaptation of forest frogs to the environment (e.g., clustering for warmth in low temperatures, dispersing to escape when water quality is abnormal). The formula is: ; Where Density represents the density of frog colonies (unit: frogs / m², value range [0, 20], suitable for the carrying capacity of the breeding pond); N represents the effective number of frogs; R represents the radius of the monitoring area of ​​the breeding pond; M represents the number of pairwise frogs (M=N×(N-1) / 2); d i，j L represents the actual distance (m) between the i-th and j-th frogs. th represents the cluster distance threshold; I(·) represents the exponential function (takes 1 if the condition in parentheses is met, otherwise takes 0); Application Notes: If Density > 10 frogs / m², it is considered a high-density cluster (which may be due to a suitable environment or stress-induced aggregation). Further judgment is made by combining activity data (e.g., high density + low activity = normal heating, high density + high activity = abnormal stress). The formula takes into account both "number" and "density" to avoid misjudgment caused by calculating density based solely on the number (e.g., a small number of densely clustered forest frogs are also considered a high-density cluster).

[0035] Output frog status data: Combine activity level and cluster density to generate a frog status index, which is then used by the AI ​​decision-making module. The formula is as follows: State=α×Activity+β×(Density / D max ); Where State represents the overall state index of the forest frog (range [0, 1]); α and β represent weighting coefficients (α = 0.6, β = 0.4, determined through training with historical data, prioritizing the immediate state of activity response); D max This indicates the maximum carrying capacity of the aquaculture pond (animals / m²).

[0036] Application Description: The threshold is set as follows: State ∈ [0.1, 0.4] is the normal state, State > 0.7 is the stress state, and State < 0.1 is the low activity state (which may be due to illness or environmental discomfort). The output State is fused with the skin potential signal collected by the non-invasive bioelectric sensor to provide complete data support for the stress assessment submodule.

[0037] In some preferred embodiments, the AI ​​decision processing module includes an improved LSTM fusion prediction model submodule and a stress assessment submodule, and the execution module includes a stress relief submodule. The improved LSTM fusion prediction model submodule uses a three-level mathematical model of feature standardization, weighted fusion, and stress level quantization to unify and quantify heterogeneous data of skin potential signals in the behavioral dimension (State) and physiological dimension, and outputs a comprehensive emergency feature value that can be directly used by the stress assessment submodule. Feature standardization: Core features are extracted and standardized from the acquired skin potential signals to ensure consistency with the dimensional range [0, 1] of the State in the behavior dimension. The formula is as follows: ; Where V RMS The root mean square value of the skin potential signal (unit: mV) represents the average energy of the potential fluctuation; T represents the number of sampling points within the calculation window (sampling frequency 1kHz, window 5s, therefore T=5000); v k This represents the original value (mV) of the skin point at the kth sampling point. V RMS Mapping to the interval [0, 1] eliminates dimensional differences; the formula is: V norm =(V RMS -V min ) / (V max -V min ); Where V norm V represents the standardized skin potential characteristic value (range [0, 1]); min V represents the minimum RMS skin potential of a frog under normal conditions; max This represents the maximum RMS skin potential of a forest frog under severe stress. Application Notes: If V RMS <V minTake V norm =0; if V RMS >V max Take V norm =1 (to avoid interference from outliers).

[0038] Weighted fusion: A dynamic weighted summation model is used to fuse physiological features V. norm The weights of the behavioral characteristic State are dynamically adjusted according to the growth stage of the forest frog, and the formula is as follows: S fusion =ω1×V norm +ω2×State; ω1 + ω2 = 1; Where S fusion ω1 represents the comprehensive stress characteristic value (range [0, 1]), which is directly input into the stress assessment submodule; ω2 represents the weight coefficient of skin potential (physiological index weight, reflecting the immediacy of stress); ω3 represents the weight coefficient of behavioral index (behavioral index weight, reflecting the persistence of stress). Stress level quantification: The stress assessment submodule will quantify S fusion The mapping is divided into three stress levels, providing clear instructions for the precise execution layer. The formula is as follows: ; Where Stress_Level represents the stress level (0=normal, 1=mild, 2=moderate / severe); θ1 represents the mild stress threshold; θ2 represents the moderate / severe stress threshold.

[0039] The stress assessment submodule sends stress relief instructions to the stress relief submodule based on the stress level, and the stress relief submodule executes the stress relief instructions.

[0040] In some preferred embodiments, the stress relief submodule includes a low-frequency sound wave generator unit (20~30Hz) and a plant essential oil slow-release device unit. After receiving the relief instruction from the stress assessment submodule, the stress relief submodule reduces the stress response of the frog by releasing soothing sound waves and plant volatile essential oils that the frog prefers. The stress relief submodule outputs instructions according to the stress level: when Stress_Level=1, it outputs a "basic relief instruction", and the low-frequency sound wave generator unit starts releasing low-frequency sound waves (20~30Hz, 60~70dB); when Stress_Level=2, it outputs an "enhanced relief instruction", and simultaneously starts the low-frequency sound wave generator unit and the plant essential oil slow-release device unit to release low-frequency sound waves (20~30Hz, 60~70dB) and plant essential oils (release amount 1~3mL / h), and simultaneously triggers the temperature and humidity control unit to slow down the rate of change of environmental parameters.

[0041] Specifically, the stress assessment submodule quantifies the stress level of the forest frog by analyzing the correlation between skin potential signals and the rate of change of environmental parameters, and outputs a soothing instruction to the stress relief submodule of the execution module. The stress relief submodule includes a low-frequency sound wave generator unit (20~30Hz) and a plant essential oil slow-release device unit. After receiving the soothing instruction from the stress assessment submodule, the stress relief submodule controls the low-frequency sound wave generator unit to release soothing sound waves and controls the plant essential oil slow-release device unit to release plant volatile essential oils that the forest frog likes to eat, thereby reducing the stress response of the forest frog. The stress relief submodule outputs stress relief instructions based on the stress level: when Stress_Level=1, it outputs a "basic relief instruction", which is activated by the low-frequency sound wave generator unit to release low-frequency sound waves (20~30Hz, 60~70dB); when Stress_Level=2, it outputs an "enhanced relief instruction", which simultaneously activates the low-frequency sound wave generator unit and the plant essential oil slow-release device unit to release low-frequency sound waves (20~30Hz, 60~70dB) and plant essential oils (release rate 1~3mL / h), and simultaneously triggers the temperature and humidity control unit to slow down the rate of change of environmental parameters, thereby achieving the purpose of stress relief.

[0042] In some preferred embodiments, the improved LSTM fusion prediction model submodule uses historical data of aquaculture environment (over 3 years), frog growth cycle parameters, and real-time sensing data as the training set. It dynamically adjusts the weights according to the frog's regional species to generate standardized control command values ​​and destandardized actual control command values. The formula is as follows: Y=W o H att +b o ; ; Where Y represents the standardized control command value ([0, 1]); W o / b o These represent the weights / biases of the fully connected layer, respectively; H att Y represents the attention-weighted hidden layer output (integrating temporal and growth cycle features); reg This represents the actual control command value after destandardization (e.g., temperature 20℃, humidity 65%RH). This represents the inverse standard function (reduced to the actual physical quantity).

[0043] Specifically, the improved LSTM fusion prediction model submodule uses historical data from aquaculture environments (over 3 years), frog growth cycle parameters, and real-time sensing data as its training set. This sets define the scope and type of training data for the improved LSTM fusion prediction model submodule, addressing the prediction bias issues caused by insufficient training data and single-dimensionality in existing technologies. The historical data (over 3 years) covers environmental fluctuations across different seasons, climates, and aquaculture cycles; frog growth cycle parameters (differences in requirements between tadpoles / juveniles / adults) inject a frog-centric physiological logic into the model; and real-time sensing data ensures the timely response of the improved LSTM fusion prediction model submodule. The fusion of these three elements enables the improved LSTM fusion prediction model submodule to learn the complete correlation between "environment-frog state-growth," providing algorithmic support for accurate decision-making. Dynamically adjusting weights based on frog regionality and species refines the adaptive capability of the improved LSTM fusion prediction model submodule, overcoming the shortcomings of existing systems that apply a "one-size-fits-all" approach and cannot adapt to the characteristics of frogs in different regions. For example, by taking into account the growth differences between the Northeast Forest Frog and the Northwest Forest Frog (such as different cold resistance and humidity requirements), and by dynamically adjusting the data weights (such as increasing the weight of cold resistance-related parameters for the Northeast Forest Frog), the control instructions output by the model can be made more in line with the needs of specific species, thereby improving the versatility and control accuracy of the system in different breeding scenarios.

[0044] More specifically, the improved LSTM fusion prediction model submodule has a full-process operation function of "training set construction - model training - dynamic adjustment of regional species weights". It is implemented through four core formulas: data standardization, LSTM gating calculation, attention weight allocation, and iterative regional species weights. It takes into account the multi-dimensional data fusion characteristics and the adaptability of forest frog regional species. The specific formula system and application logic are as follows: Training set data standardization: Since the training set contains heterogeneous data such as "environmental parameters (℃ / %RH), growth cycle parameters (days / physiological threshold), and real-time sensing data (mV / animal / m²)," it needs to be standardized to the [0, 1] interval first. The formula is as follows: X norm，i =(X i -X min，i ) / (X max，i -X min，i ); Where X norm，i X represents the standardized value of the i-th feature (such as temperature, humidity, frog skin spots, etc.); i X represents the original value of the i-th feature; min，i / X max，i This represents the historical extreme value of the i-th type of feature (e.g., the temperature difference X in the Northeast Forest Frog breeding area). min =-10℃, X max =30℃; Northwest Forest Frog Humidity X min =30%RH, Xmax =85%RH); Application Notes: Standardize according to feature type (environment, growth cycle, perception) to avoid the large amplitude of single-dimensional data from obscuring key features (such as the core influence of the growth cycle parameters of the forest frog on regulation).

[0045] Improved LSTM Gating Computation: The improved LSTM integrates the temporal correlation of 3 years of historical data, growth cycle parameters, and real-time data through gating units and attention mechanisms. The basic LSTM gating formula is as follows: i t =σ(W xi X t +W hi h t-1 +b i ); f t =σ(W xf X t +W hf h t-1 +b f ); C t =tanh(W) xc X t +W hc h t-1 +b c ); C t =f t ⊙C t-1 +i t ⊙C t ′; O t =σ(W xo X t +W ho h t-1 +b o ); h t= o t ⊙tanh(C t ); Where i t / f t / o t These represent the input gate, forget gate, and output gate (controlling data input, forgetting, and output); C t ′ / C t Representing candidate cell state / final cell state respectively (feature association in the storage time dimension); h t / h t-1 W represents the current / previous hidden layer output (core feature carrier), respectively; xi / W hiEqual representation of the weight matrix (the core parameters for model training); b i =E represents the bias term; σ represents the Sigmoid activation function (output [0, 1], controlling the opening and closing of the gate), tanh represents the hyperbolic tangent function (output [-1, 1], normalized cell state); ⊙ represents the Hadamard product (element-wise multiplication); X t This represents the standardized training set data at time t (integrating environment, growth cycle, and real-time sensing features). Improved logic: Compared with traditional LSTM, "growth cycle feature gating" is added. The weight coefficient of growth cycle parameters is added to the input gate to prioritize the retention of key environmental features of tadpole stage / juvenile frog stage / adult frog stage (such as increasing the gating weight of dissolved oxygen parameter in tadpole stage).

[0046] To strengthen the weighting of "forest frog growth cycle parameters and regional species characteristics", an attention mechanism is introduced, with the following formula: ; e t =v T tanh(W h h t +W v V c +b e ); ; Where, α t e represents the attention weight (weight percentage of core features) at time t; t This represents the attention score (a measure of the importance of features at time t); v T / W h / W v Let b represent the attention layer weight matrix. e For bias terms; V c Represents the parameter vector of the frog's growth cycle (e.g., tadpole stage = [1, 0, 0], juvenile stage = [0, 1, 0], adult stage = [0, 0, 1]); H att This represents the hidden layer output after attention weighting (integrating temporal and growth cycle features). Application Notes: The higher the degree of matching between growth cycle parameters, the better the α... t The larger the attention weight (e.g., the attention weight for juvenile frog data is 1.2 times that for adult frog data), the more likely the model will learn the environmental regulation patterns of the corresponding stage.

[0047] By dynamically adjusting the weights of different varieties based on their geographical location, the system adapts to the characteristics of different varieties: Based on the attention fusion results, the weights are dynamically adjusted iteratively through variety feature vectors and gradient descent.

[0048] Initial weights are assigned to regional varieties, and feature weight vectors W for different regional varieties are preset.breed The formula is: W breed =[w T w H w L w W ]; W j =w j0 ×(1+δ j ); Among them, w T / w H / w L / w W These represent the variety weights for temperature, humidity, light intensity, and water quality parameters, respectively; w j0 Indicates the basic weight (e.g., the general frog w). T0 =0.25, w H0 =0.25); δ j This represents the variety correction factor (for the Northeast Forest Frog, which is cold-resistant, δ...). T =+0.15, δ H =-0.05; the Northwest Forest Frog is drought-resistant, therefore δ H =+0.1,δ T =-0.05).

[0049] By using error feedback from real-time sensing data, W is iteratively adjusted using the gradient descent method. breed The formula is: ; ; in / yk+1 / k represents the variety weight in the (k+1 / k)th iteration; η represents the learning rate (set to 0.01 to balance iteration speed and stability); Loss represents the mean squared error loss (the deviation between the model's predicted control effect and the actual effect, such as predicted survival rate minus actual survival rate); yn / This represents the actual value / predicted value of the nth sample (e.g., survival rate of forest frogs, stress level). This represents the partial derivative of the loss function with respect to the biased weights (indicating the direction of weight adjustment). Iterative logic: If, after adjusting the current weights, the survival rate of the Northeast Forest Frog breeding area remains below the threshold under low-temperature conditions, the model automatically increases the weight of the temperature parameter (w). T Increase by 0.02 per iteration); if the activity level of the Northwest Forest Frog meets the standard in a low humidity environment, the model reduces the weight of the humidity parameter (w). H Reduce by 0.01 per iteration).

[0050] The model inputs the weighted and fused feature Hatt into the fully connected layer to generate control command parameters (such as target values ​​for temperature and humidity), as shown in the formula: Y=W o H att +b o ; ; Where Y represents the standardized control command value ([0, 1]); W o / b o These represent the weights / biases of the fully connected layer, respectively; Y reg This represents the actual control command value after destandardization (e.g., temperature 20℃, humidity 65%RH). This represents the inverse standard function (reduced to the actual physical quantity).

[0051] The advantages of the improved LSTM mentioned above are: multi-dimensional fusion: through data standardization and attention mechanisms, it achieves deep fusion of 3 years of historical data, growth cycle parameters, and real-time sensing data, avoiding the limitations of a single data dimension; dynamic adaptation: regional species weights are iteratively optimized through gradient descent without human intervention, adapting to the growth characteristics of different species (such as the cold-resistant Northeast Forest Frog and the drought-resistant Northwest Forest Frog); engineering implementation: the formula is mainly based on matrix operations, which can run efficiently on cloud platforms / edge computing modules (single weight adjustment time ≤50ms), meeting the needs of real-time regulation; and controllable error: through iterative optimization of the loss function, it ensures that the deviation between the regulation instructions predicted by the model and the actual breeding needs is ≤±0.5℃ (temperature and humidity) and ≤5% (survival rate).

[0052] In some preferred embodiments, the execution module further includes a safety protection submodule, and the external interference sensing submodule is linked to the safety protection submodule; when the external interference sensing submodule detects a predator approaching, the safety protection submodule triggers the ultrasonic repelling device to release ultrasonic waves to repel the predator; when the external interference sensing submodule predicts extreme weather, the safety protection submodule controls the activation of the automatic closing device of the windproof and heat-insulating shed before the extreme weather, closing the windproof and heat-insulating shed. This measure can be considered a safety protection command, which belongs to one type of control command.

[0053] Specifically, the execution module also includes a security protection submodule and an external interference sensing submodule linked with the security protection submodule. By linking the two submodules, the defect of "independent operation of sensing and protection" in the existing technology is overcome, and the passive problem of "only being able to detect interference but not being able to respond automatically" in traditional aquaculture is solved. This establishes a hardware logic connection for the real-time response to "interference and protection", which is the basic architectural support for the system to realize "proactive risk prevention and control".

[0054] When the external interference sensing submodule detects an approaching predator, the safety protection submodule triggers an ultrasonic repelling device to release ultrasonic waves to repel the predator. This is used to clarify the specific linkage logic and execution actions under the "predator interference" scenario. Combined with the definition in the solution that "the external interference sensing submodule includes a sound and vibration sensor unit," the sound and vibration sensor unit can accurately identify intrusions (such as the vibration of snakes crawling or the sound of rodents moving) through the vibration and sound characteristics generated by predator activity. This non-harmful control method against predators of forest frogs avoids the lag and labor costs of traditional manual hunting and killing of predators, and prevents the pollution of forest frogs and the breeding environment by chemical repellents. It achieves a closed-loop control of "predator intrusion - automatic identification - immediate repellency," directly improving the safety of forest frog breeding. When the external interference sensing submodule predicts extreme weather, the safety protection submodule controls the automatic closing device of the windproof and heat-insulating shed to be activated before the extreme weather. Closing the windproof and heat-insulating shed is used to define the predictive protection logic under the "extreme weather interference" scenario, highlighting the system's "foresight" advantage. By combining the "monitoring of precursors to extreme weather" function of the external interference sensing submodule in the solution (such as air pressure changes before heavy rain and airflow vibrations before strong winds), "extreme weather prediction" is achieved through the linkage of sensor data and AI decision processing module; "activating windproof and heat-insulating sheds before extreme weather" solves the problem of existing technologies "passively responding only after extreme weather arrives". The windproof and heat-insulating sheds that are closed in advance can maintain stable temperature and humidity in the breeding area, prevent extreme weather such as heavy rain and strong winds from directly impacting the frog habitat, prevent frogs from having stress reactions or dying due to sudden environmental changes, and ensure the continuity and stability of the breeding environment.

[0055] In some preferred embodiments, the AI ​​decision processing module further includes a microecological balance submodule, and the execution module further includes a water quality control submodule; the microecological sensing submodule detects the community ratio of beneficial and harmful bacteria in the water body through a microbial sensor unit; the microecological balance submodule predicts the risk of water quality deterioration based on the community ratio data, and the water quality control submodule adds corresponding probiotics according to the prediction results of the microecological balance submodule. This measure can be identified as a water quality control instruction, which is one type of control instruction.

[0056] For example, the data processing of the micro-ecological balance submodule forms a closed loop around "micro-ecological data quantification → risk index calculation → precise dosage derivation → feedback optimization," which is achieved through a data quantification process of "perception-decision-execution." By calculating the proportion of beneficial bacteria / harmful bacteria in the microecological community, the microecological data is quantified. The formula is as follows: B=B count / T count ×100%, H=H count / T count ×100%; T count =B count +Hcount ; Where B represents the percentage of beneficial bacteria (%, such as photosynthetic bacteria and nitrifying bacteria); H represents the percentage of harmful bacteria (%, such as Vibrio and Escherichia coli); B count Indicates the number of beneficial bacteria detected by the microbial sensor (unit: CFU / mL); H count This indicates the number of harmful bacteria detected by the microbial sensor (unit: CFU / mL); T count This indicates the total number of bacteria in the water (unit: CFU / mL).

[0057] Application Notes: The sensor collects data every 15 minutes and directly outputs B. count and H count The formula is used to convert the percentage index to eliminate the impact of water volume differences on the absolute number; the baseline thresholds for different growth stages of forest frogs (such as B≥60% and H≤10% during the tadpole stage) are preset to provide a reference for subsequent risk assessment.

[0058] Based on the above formula, the community proportion is standardized, and the formula is: B norm =(B-B min ) / (B max -B min ), H norm =(H-H min ) / (H max -H min ); Among them B norm H norm These represent the standardized percentage of beneficial bacteria / harmful bacteria (value range [0, 1]); B = 30%, B max =80% indicates a reasonable range for the proportion of beneficial bacteria, as determined through aquaculture experiments; H min =1%, H max =20% indicates the risk range of harmful bacteria.

[0059] Application Description: The percentage data is mapped to the [0, 1] interval, which is consistent with the dimensions of environmental parameters and frog status data, providing standardized input for the micro-ecological balance submodule.

[0060] The formula for predicting water quality deterioration risk, i.e., calculating the microecological risk index, is as follows: Risk = ω1 × (1 - B) norm ) + ω2×H norm +ω3×(ΔH / Δt); Where Risk represents the water quality deterioration risk index (value range [0, 1], the larger the value, the higher the risk); ω1=0.4, ω2=0.5, ω3=0.1 represent the weighting coefficients (the proportion of harmful bacteria has the highest weight, followed by the proportion of beneficial bacteria, and finally the growth rate of harmful bacteria); (ΔH / Δt) represents the rate of change of the proportion of harmful bacteria (% / h, for example, if H increases from 8% to 10% in two consecutive tests, Δt=0.25h, then ΔH / Δt=8% / h).

[0061] Application notes: Risk grading thresholds: Risk<0.3 (low risk, no need to add probiotics), 0.3≤Risk<0.6 (medium risk, regular addition), Risk≥0.6 (high risk, intensive addition); Introduce ΔH / Δt to avoid misjudgment by "only looking at the instantaneous percentage" (e.g., H=10% but the growth rate reaches 10% / h, early intervention is required).

[0062] The formula for deriving and calculating the amount of probiotics to be added is as follows: V=(H target -H current )×V water ×106 / (C pro ×η); Where V represents the amount of probiotics added (unit: mL); H target Indicates the percentage of target harmful bacteria (%), with H representing medium risk. target =5), high risk, take H target =3%); H current Indicates the current percentage of harmful bacteria (%); V water This indicates the volume of water in the aquaculture pond (unit: m³); 10⁶ represents the unit conversion factor (1 m³ = 10⁶ mL); C pro Indicates the effective concentration of probiotics (unit: CFU / mL, such as the effective concentration of commercial photosynthetic bacteria CFU / mL). pro =108 CFU / mL); η represents the degradation efficiency of probiotics against target harmful bacteria (value 0.6-0.8, determined experimentally, such as eta=0.75 for Bdellovibrio against Vibrio).

[0063] Application Notes: Example: V water =100m³, H current =12%, H target =5%, C pro =108 CFU / mL; η=0.7, then V=(5-12)×100×106 / (108×0.7)=100mL; for high risk, multiply by the enhancement coefficient k=1.5 (i.e., V 强化 =V×1.5), and at the same time, the aeration device is linked to increase dissolved oxygen and enhance the activity of probiotics.

[0064] Adjustments and optimizations are made based on feedback on the application results, specifically by calculating the application effect correction coefficient, using the following formula: K corr =(H target -H actual ) / (H target -H current ); Where K corr This represents the correction factor (range [0.8, 1.2], deviation from 1 will adjust subsequent dosage); H actual This indicates the percentage of harmful bacteria detected 1 hour after addition (%).

[0065] Application Notes: If K corr <0.9 (actual effect not met), the next dosage should be V. next =V×1.1; if K corr >1.1 (too strong), the next dosage should be V. next =V×0.9, to avoid excessive probiotics leading to microecological imbalance.

[0066] In some preferred embodiments, the execution module further includes a temperature and humidity control submodule, a shading and supplemental lighting submodule, and a feeding linkage submodule. The temperature and humidity control submodule executes the temperature and humidity control commands issued by the AI ​​decision processing module, the shading and supplemental lighting submodule executes the light intensity control commands issued by the AI ​​decision processing module, and the feeding linkage submodule executes the feeding commands issued by the AI ​​decision processing module. The remote interaction module automatically generates growth reports and equipment maintenance reminders through the cloud platform and pushes them to the user terminal via the APP.

[0067] Specifically, the execution module also includes a temperature and humidity control submodule, a shading and supplemental lighting submodule, and a feeding linkage submodule. The temperature and humidity control submodule executes the temperature and humidity control commands issued by the AI ​​decision processing module, the shading and supplemental lighting submodule executes the light intensity control commands issued by the AI ​​decision processing module, and the feeding linkage submodule executes the feeding commands issued by the AI ​​decision processing module. This improves the core functional coverage of the execution module, ensures basic environmental adaptation through the temperature and humidity control submodule, accurately matches the light requirements of the frogs at different growth stages (such as weak light during the tadpole stage and light avoidance during the adult frog stage), and enables the coordination of nutrient supply and growth cycle through the feeding linkage submodule. Together, these three modules provide functional support for the integrated and precise control of "environment-nutrition".

[0068] The remote interaction module automatically generates growth reports and equipment maintenance reminders through a cloud platform and pushes them to user terminals via an APP. The core functions and implementation path of the remote interaction module are clearly defined. The cloud platform automatically integrates data to generate growth reports (quantifying the growth status of forest frogs) and equipment maintenance reminders (predicting fault risks). The APP push breaks spatial limitations, enabling users to keep track of breeding dynamics and system status in real time, improving management convenience and emergency response efficiency, and supporting unattended breeding.

[0069] For example, the temperature and humidity control submodule includes a variable frequency heating element unit and a misting humidifier unit; the shading and supplemental lighting submodule includes a sunshade unit and an LED supplemental lighting unit. The temperature and humidity control submodule is linked together by the variable frequency heating element unit, the misting humidifier unit, the sunshade unit, and the LED supplemental lighting unit to define the specific hardware composition and collaborative working mode of the temperature and humidity control submodule, solving the problem of independent operation of temperature, humidity, and light control equipment in the prior art, which easily leads to conflicts in environmental parameters. The variable frequency heating element unit precisely controls the temperature, the misting humidifier unit provides stable humidification, and the sunshade unit and the LED supplemental lighting unit indirectly assist in balancing temperature and humidity by adjusting the light (such as shading to cool down in strong light). The linkage of multiple device units forms a collaborative control mechanism of "temperature-humidity-light", providing a stable and suitable microenvironment for the forest frog; the implementation of temperature and humidity control accuracy of ±0.5℃ and ±3%RH is used to clarify the quantitative indicators of temperature and humidity control, responding to the sensitive needs of forest frogs at different growth stages in the technical solution (such as water temperature fluctuations during the tadpole stage needing to be strictly controlled within ±1℃). This precision index breaks through the limitations of existing technologies, which generally control within ±1℃ and ±5%RH. Through hardware linkage and precise implementation of AI decision-making commands, it ensures that temperature and humidity fluctuations are within the tolerance threshold of the forest frog, directly improving the survival rate and growth stability of the breeding. This is the core quantitative manifestation of the system's "precise control" characteristics.

[0070] The remote interaction module features a cloud platform that automatically generates growth reports and equipment maintenance reminders. The "automatic growth report generation" represents a valuable transformation of data across the entire system chain. The cloud platform integrates frog status data (activity level, growth cycle, etc.) collected by multi-dimensional sensing modules with environmental control data to create a visualized report, solving the problems of fragmented manual records and difficulty in quantifying growth in traditional aquaculture. The "automatic equipment maintenance reminder generation" uses equipment operating parameters (such as sensor sampling frequency and the number of times execution units start and stop) to predict faults, avoiding the interruption of control caused by "behind-the-scenes detection of equipment failure" in existing technologies, thus ensuring stable system operation. Furthermore, the reminders are pushed to user terminals via an app, clarifying the interaction path between the cloud platform and the user, breaking through the behavioral limitations of traditional aquaculture that require on-site data viewing. The "APP push" system synchronizes growth reports (such as statistics on frog molting cycles and growth rate comparisons) and maintenance reminders (such as sensor calibration prompts and warnings about the battery level of avoidance modules) to mobile phones and other terminals in real time, allowing users to keep track of the breeding dynamics without supervision. At the same time, the "push" mode is more timely than "user-initiated queries," ensuring that users can quickly respond to emergency reminders (such as equipment malfunctions), reducing breeding risks, improving the system's ease of operation and management efficiency, and is a key link in realizing "unattended operation and remote control."

[0071] Through the above technical solutions, the intelligent control system for frog farming environment provided in this application has the following advantages: ① Comprehensive sensing coverage, solidifying the data foundation for intelligent control: Through four sensing sub-modules—environment, frog status, external interference, and micro-ecology—combined with dedicated hardware such as infrared thermometry, bioelectric sensors, and microbial sensors, it achieves comprehensive data collection of temperature and humidity, water quality physicochemical indicators, frog behavior / physiological characteristics, predator / extreme weather signals, and the proportion of microbial communities in the water body. This breaks through the limitations of traditional systems' "single and partial data collection," providing comprehensive and accurate data source support for AI decision-making; ② Precise quantification of frog behavior and... Intelligent stress relief enhances adaptability to aquaculture: Biometric recognition algorithms quantify frog activity, cluster density, and overall condition index, combined with a two-dimensional fusion model of skin potential signals to accurately determine stress levels. This is complemented by a non-invasive soothing solution using low-frequency sound waves and plant essential oils, forming a closed loop of "frog state perception - stress assessment - precise relief," addressing the problems of traditional regulation neglecting frog physiological states and indiscriminate stress intervention, ensuring stable frog growth. ③ AI-driven decision-making ensures precise adaptation to diverse aquaculture needs: An improved LSTM fusion prediction model submodule incorporates over 3 years of historical data, growth cycle parameters, and real-time data. The data serves as a training set, allowing for dynamic weight adjustments based on the geographical distribution and species of forest frogs to generate targeted regulatory instructions, breaking through the traditional "one-size-fits-all" regulatory model. Simultaneously, through algorithms such as feature standardization and weighted fusion, heterogeneous data is uniformly quantified, improving the scientific rigor and adaptability of decision-making to meet the needs of forest frog farming in different regions and at different growth stages. ④ Multi-dimensional proactive risk control reduces farming losses: Through the linkage of external interference perception and safety protection sub-modules, it proactively responds to predator invasions and extreme weather; through microecological perception and precise probiotic addition, it predicts and controls the risk of water quality deterioration; combined with real-time intervention in stress states, it constructs... The comprehensive risk control system covering "natural enemies, extreme weather, water quality, and frog behavior" addresses the pain points of traditional aquaculture, such as "passive risk response and high loss rates." ⑤ Precise zoning and convenient management enhance the practicality of aquaculture: The execution module encompasses multiple functions including temperature and humidity control, shading and supplemental lighting, feeding, stress relief, and water quality regulation, enabling personalized zoning control of the aquaculture area to meet the differentiated needs of frogs at different growth stages. The remote interaction module generates growth reports and equipment maintenance reminders via a cloud platform and pushes them to the app, breaking spatial limitations, supporting unattended operation and remote manual intervention, reducing aquaculture management costs, and improving operational convenience.

[0072] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. An intelligent environmental control system for forest frog farming, characterized in that, It includes a multi-dimensional perception module, an AI decision processing module, an execution module, and a remote interaction module. The multi-dimensional perception module is communicatively connected to the AI ​​decision processing module, and the AI ​​decision processing module is connected to both the execution module and the remote interaction module. The multi-dimensional sensing module collects environmental parameter data, frog status data, external disturbance data, and micro-ecological data. The multi-dimensional sensing module includes an environmental parameter sensing submodule, a forest frog state sensing submodule, an external disturbance sensing submodule, and a micro-ecological sensing submodule. The environmental parameter sensing submodule collects environmental parameter data; The frog status perception submodule collects frog status data; The external interference sensing submodule collects data on extreme weather and interference from natural enemies. The microecological sensing submodule collects bacterial colony data in the aquaculture pond water. The environmental parameter sensing submodule includes an infrared temperature sensor unit, a humidity sensor unit, a light sensor unit, and a water quality multi-parameter sensor unit. The infrared temperature sensor unit collects temperature data, the humidity sensor unit collects humidity data, the light sensor unit collects light intensity data, and the water quality multi-parameter sensor unit collects water quality data including pH and dissolved oxygen in the aquaculture pond. The frog state perception submodule includes a high-definition infrared camera unit, a biometric recognition unit, and a non-invasive bioelectric sensor unit. The high-definition infrared camera unit collects behavioral characteristic image data of the frog, and the non-invasive bioelectric sensor unit collects skin potential signal data of the frog. The external interference sensing submodule includes an acoustic vibration sensor unit, which collects characteristic vibration sound waves of airflow and air pressure as precursors to extreme weather and characteristic vibration sound wave data of predator activities. The microecological sensing submodule includes a microbial sensor unit, which collects data on the community ratio of beneficial and harmful bacteria in the aquaculture pond water. The AI ​​decision processing module generates specific control instructions based on historical and real-time data. The control instructions include at least one of the following: emergency relief instructions, temperature and humidity control instructions, light intensity control instructions, and feeding instructions. The execution module executes specific control commands; The remote interaction module provides data visualization and manual intervention control command functions based on the established cloud platform; The biometric identification unit operates on the image data acquired by the high-definition infrared camera unit to generate a frog state index, including: Target detection: The biometric recognition unit accurately locates individual forest frogs from the image using a biometric recognition algorithm, the formula of which is: Confidence=P obj ×IOU; Where Confidence represents the target confidence level, used to determine the probability of whether it is a forest frog, and its value ranges from [0, 1]; P obj The probability of the target being predicted by the model is represented by IOU, which represents the intersection-union ratio between the predicted bounding box and the actual frog area. IOU is used to measure the localization accuracy and has a value range of [0, 1]. Feature extraction: Activity level is calculated based on the changes in the center coordinates of the frog in different frames. The activity level is used to determine whether the frog is in a stressed or healthy state. The formula is as follows: ； Where Activity represents the average activity level of the frog group, with a value range of [0, 1], where 0 represents stillness and 1 represents vigorous movement; N represents the number of valid frogs identified in a single frame image; (x i，t y i，t (x) represents the center coordinates (in pixels) of the i-th frog in frame t; i，t-1 y i，t-1 ) represents the center coordinates of the i-th frog in frame t-1, and T represents the calculation time window (set to 5s, corresponding to a camera frame rate of 10fps, for a total of 50 frames); D scale This represents the coordinate scaling factor (converting pixel distance to actual distance, e.g., 1 pixel = 0.5 cm, D). scale =50 pixels / m); Calculating cluster density: Cluster density reflects the degree of adaptation of forest frogs to the environment (e.g., clustering for warmth in low temperatures, dispersing to escape when water quality is abnormal). The formula is: ； Where Density represents the density of frog colonies (unit: frogs / m², value range [0, 20], suitable for the carrying capacity of the breeding pond); N represents the effective number of frogs; R represents the radius of the monitoring area of ​​the breeding pond; M represents the number of pairwise frogs (M=N×(N-1) / 2); d i，j L represents the actual distance (m) between the i-th and j-th frogs. th represents the cluster distance threshold; I(·) represents the exponential function (takes 1 if the condition in parentheses is met, otherwise takes 0); Output frog status data: Combine activity level and cluster density to generate a frog status index, which is then used by the AI ​​decision-making module. The formula is as follows: State=α×Activity+β×(Density / D max ); Where State represents the overall state index of the forest frog (range [0, 1]); α and β represent weighting coefficients; D max This indicates the maximum carrying density of the aquaculture pond (animals / m²). The AI ​​decision processing module includes an improved LSTM fusion prediction model submodule and a stress assessment submodule, and the execution module includes a stress relief submodule. The improved LSTM fusion prediction model submodule uses a three-level mathematical model of feature standardization, weighted fusion, and stress level quantization to unify and quantify heterogeneous data of skin potential signals in the behavioral dimension (State) and physiological dimension, and outputs a comprehensive emergency feature value that can be directly used by the stress assessment submodule. Feature standardization: Core features are extracted and standardized from the acquired skin potential signals to ensure consistency with the dimensional range [0, 1] of the State in the behavior dimension. The formula is as follows: ； Where V RMS The root mean square value of the skin potential signal (unit: mV) represents the average energy of the potential fluctuation; T represents the number of sampling points within the calculation window (sampling frequency 1kHz, window 5s, therefore T=5000); v k This represents the original value (mV) of the skin point at the kth sampling point. V RMS Mapping to the interval [0, 1] eliminates dimensional differences; the formula is: V norm =(V RMS -V min ) / (V max -V min ); Where V norm V represents the standardized skin potential characteristic value (range [0, 1]); min V represents the minimum RMS skin potential of a frog under normal conditions; max This represents the maximum RMS skin potential of a forest frog under severe stress. Weighted fusion: A dynamic weighted summation model is used to fuse physiological features V. norm The weights of the behavioral characteristic State are dynamically adjusted according to the growth stage of the forest frog, and the formula is as follows: S fusion =ω1×V norm +ω2×State; ω1 + ω2 = 1; Where S fusion ω1 represents the comprehensive stress characteristic value (range [0, 1]); ω2 represents the weighting coefficient of skin potential (physiological index weight, reflecting the immediacy of stress); ω3 represents the weighting coefficient of behavioral index (behavioral index weight, reflecting the persistence of stress). Stress level quantification: The stress assessment submodule will quantify S fusion The mapping is based on a three-level stress response, providing clear instructions for the execution module. The formula is as follows: ； Where Stress_Level represents the stress level (0=normal, 1=mild, 2=moderate / severe); θ1 represents the mild stress threshold; θ2 represents the moderate / severe stress threshold; The stress assessment submodule sends stress relief instructions to the stress relief submodule based on the stress level, and the stress relief submodule executes the stress relief instructions.

2. The intelligent control system for the forest frog breeding environment according to claim 1, characterized in that, The stress relief submodule includes a low-frequency sound wave generator unit and a plant essential oil slow-release device unit. The stress relief submodule outputs stress relief commands based on the stress level: when Stress_Level=1, it outputs a "basic relief command", which is activated by the low-frequency sound wave generator unit to release low-frequency sound waves; when Stress_Level=2, it outputs an "enhanced relief command", which simultaneously activates the low-frequency sound wave generator unit and the plant essential oil slow-release device unit to release low-frequency sound waves and plant essential oils, and simultaneously triggers the temperature and humidity control unit to slow down the rate of change of environmental parameters.

3. The intelligent control system for the breeding environment of forest frogs according to claim 2, characterized in that, The improved LSTM fusion prediction model submodule uses historical data of aquaculture environment (over 3 years), frog growth cycle parameters, and real-time sensing data as the training set. It dynamically adjusts the weights according to the frog's regional species to generate standardized control command values ​​and destandardized actual control command values. The formula is as follows: Y=W o H att +b o ; ； Where Y represents the standardized control command value ([0, 1]); W o / b o These represent the weights / biases of the fully connected layer, respectively; H att Y represents the attention-weighted hidden layer output (integrating temporal and growth cycle features); reg This represents the actual control command value after destandardization; This represents the inverse standard function (reduced to the actual physical quantity).

4. The intelligent control system for the breeding environment of forest frogs according to claim 3, characterized in that, The execution module also includes a security protection submodule, and the external interference sensing submodule is linked to the security protection submodule. When the external interference sensing submodule detects an approaching predator, the safety protection submodule triggers the ultrasonic repelling device to release ultrasonic waves to repel the predator; when the external interference sensing submodule predicts extreme weather, the safety protection submodule controls the automatic closing device of the windproof and heat-insulating shed to close the windproof and heat-insulating shed before the extreme weather.

5. The intelligent control system for the breeding environment of forest frogs according to claim 4, characterized in that, The AI ​​decision processing module also includes a micro-ecological balance sub-module, and the execution module also includes a water quality control sub-module; The microecological sensing submodule detects the community ratio of beneficial and harmful bacteria in the water body through a microbial sensor unit; the microecological balance submodule predicts the risk of water quality deterioration based on the community ratio data; and the water quality regulation submodule adds corresponding probiotics based on the prediction results of the microecological balance submodule.

6. The intelligent control system for the breeding environment of forest frogs according to claim 5, characterized in that, The execution module also includes a temperature and humidity control submodule, a shading and supplemental lighting submodule, and a feeding linkage submodule. The temperature and humidity control submodule executes the temperature and humidity control commands issued by the AI ​​decision processing module, the shading and supplemental lighting submodule executes the light intensity control commands issued by the AI ​​decision processing module, and the feeding linkage submodule executes the feeding commands issued by the AI ​​decision processing module. The remote interaction module automatically generates growth reports and equipment maintenance reminders through the cloud platform and pushes them to the user's terminal via the APP.

Images