Bird nest reproduction state real-time monitoring system based on Internet of Things

Abstract

The invention relates to the field of the Internet of Things, and discloses a bird nest reproduction state real-time monitoring system based on the Internet of Things, and the system comprises a sensing layer which comprises a piezoelectric sensor group, a non-contact infrared temperature measurement array, an MEMS microphone array, and a gas sensor module; the edge calculation layer receives the multi-modal data of the sensing layer in real time, carries out localization processing, and extracts vibration energy, voiceprint features, temperature gradient and gas concentration change rate; a cloud analysis layer; and a feedback control layer. Multi-modal data such as egg turning times, nest temperature, acoustic signals and COconcentration are comprehensively collected through various sensors such as a piezoelectric sensor group and a non-contact infrared temperature measurement array of the sensing layer, the data are preprocessed in real time through algorithms such as wavelet transformation and Kalman filtering, key characteristics such as vibration energy and temperature gradient are extracted, and the detection accuracy is improved. The data volume per day is reduced from 8-12 GB to 200-500 MB, data comprehensiveness is guaranteed, processing efficiency is improved, and high-quality data support is provided for follow-up risk assessment.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) technology, specifically to a real-time monitoring system for the breeding status of bird nests based on IoT. Background Technology

[0002] Bird reproduction is a crucial stage in the bird's life cycle and is of great significance to the continuation of bird populations. Traditional methods of monitoring bird nesting and breeding status mostly rely on manual observation, which is not only time-consuming and labor-intensive, but also easily interferes with the birds' breeding activities, leading to inaccurate monitoring data and even affecting the birds' normal reproduction.

[0003] With the development of IoT technology, various sensors and monitoring systems are being used in wildlife monitoring. However, existing monitoring systems suffer from problems such as incomplete data collection, low monitoring accuracy, large data processing delays, and significant interference with birds. For example, some systems can only monitor single parameters such as nest temperature or vibration, failing to fully reflect the breeding status of nests. Some systems rely on cloud computing for data processing, resulting in high transmission delays and difficulty in responding to anomalies in real time. Furthermore, some systems have poorly deployed sensors that are easily detected by birds, interfering with their breeding behavior. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an Internet of Things-based real-time monitoring system for bird nest breeding status. This system can comprehensively collect multimodal data during the bird nest breeding process, enabling accurate and real-time monitoring of the bird nest breeding status. It can also perform risk assessment and feedback control based on the monitoring results, effectively improving the success rate of bird breeding.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a real-time monitoring system for bird nest breeding status based on the Internet of Things, comprising:

[0006] The sensing layer includes a piezoelectric sensor array, a non-contact infrared temperature measurement array, a MEMS microphone array, and a gas sensor module, which are deployed in different areas of the Bird's Nest's physical structure.

[0007] The edge computing layer receives multimodal data from the perception layer in real time and performs local processing to extract vibration energy, acoustic features, temperature gradient and gas concentration change rate, and optimizes communication efficiency through a three-level data diversion rule.

[0008] The cloud-based analysis layer includes a reproductive risk assessment module based on a hybrid model of XGBoost, LSTM, and Bayesian methods. It takes the preprocessed feature vector from the edge layer as input and outputs a quantitative value of the risk level and intervention suggestions.

[0009] The feedback control layer, which includes a nest temperature regulation module and a directional acoustic wave transmitter, performs closed-loop control according to cloud commands.

[0010] Preferably, the piezoelectric sensor group is embedded in the biomimetic tree branch bark layer or laid in the waterproof and breathable membrane interlayer at the bottom of the artificial nest box, with a sensitivity of 0.5mV / g, a response frequency band of 0.1-200Hz, and covers ≥80% of the nest material projection area in a ring or plane.

[0011] The infrared temperature measurement array is installed 30cm above the nest top, and the focus is calibrated by UWB positioning. It has a grid scanning mode of 9 points per hour and a continuous temperature measurement mode triggered by pressure.

[0012] The MEMS microphone array is suspended from the support branch outside the nest in a triangular layout, and uses beamforming to collect acoustic signals in the frequency band of 50-8000Hz, with a dynamic sound pressure level of 30-120dB.

[0013] The gas sensor module is nested at the rear of the ventilation grille of the artificial nest box, with the detection point 10-15cm horizontally away from the biological activity area.

[0014] Preferably, the data processing method of the edge computing layer includes:

[0015] Piezoelectric vibration signal processing: Short-time pulses in the 5-10Hz frequency band are located using wavelet transform; the determination formula is as follows:

[0016] , A value greater than 0.1 is recorded as a valid hatching action.

[0017] Acoustic signal processing: After eliminating environmental interference using an FIR bandpass filter, 13-dimensional MFCC coefficients are generated, and the anomaly threshold is calculated using the Euclidean distance formula.

[0018] in, For Euclidean distance, For the generated j-th dimension MFCC coefficients, For the j-th dimension MFCC coefficient of the preset template, when When this happens, it is judged as abnormal;

[0019] Temperature field processing: A continuous temperature field is constructed by bicubic spline interpolation of the temperature measurement data, and local hot spots are marked when the temperature difference is >1.5℃;

[0020] Gas analysis: Sensor drift is eliminated through Kalman filtering algorithm, and the rate of CO2 change is dynamically calculated.

[0021] Preferably, in the cloud analysis layer, the XGBoost classifier calculation process first sets the predicted values ​​of all samples as initial values; then, it successively constructs decision trees, with each tree trained based on the residual of the previous tree (the difference between the true label and the current predicted value); when constructing each tree, a greedy algorithm is used to select the optimal split point to minimize the objective function; after constructing t trees, the outputs of all trees are summed to obtain the final risk type prediction value. The predicted value is then compared with the true label.

[0022] Preferably, the LSTM uses a forget gate to determine which information to discard from the cell state, an input gate to determine which new information to store in the cell state, then updates the cell state, and finally uses an output gate to determine what value to output, with the output value based on the cell state.

[0023] Preferably, the forget gate formula is as follows:

[0024] The output of the forget gate at time t. It is the sigmoid activation function. Here is the weight matrix for the forget gate. The hidden state at time t-1 Let be the input feature vector at time t. For the bias term of the forget gate;

[0025] The input gate formula:

[0026] in, The input gate output is at time t. Here is the weight matrix of the input gate. For the bias term of the input gate, Let t represent the candidate cell state at time t. The hyperbolic tangent activation function is used. This is the weight matrix for the candidate cell states. This is a bias term for the candidate cell state.

[0027] Preferably, the cell state update formula is:

[0028] Let t represent the cell state at time t. The cell state at time t-1. This is element-wise multiplication.

[0029] Preferably, the feedback control layer includes a nest temperature regulation module and a directional acoustic wave transmitter. The nest temperature regulation module is composed of a carbon fiber heating film and a ceramic heat insulation layer. It adopts PWM duty cycle adjustment to achieve a temperature control accuracy of ±0.5℃. The directional acoustic wave transmitter responds to the infrared thermometry detection of a characteristic of 28-32℃ for 10 seconds and emits an 18kHz pulse wave to drive away predators.

[0030] Preferably, the MFCC coefficient generation method specifically includes: processing the bandpass filtered acoustic signal in 25ms frames, performing Fourier transform on each frame to obtain the power spectrum, extracting the logarithmic energy through a 40-channel Mel filter bank, and taking the first 13 coefficients as the voiceprint feature vector after performing discrete cosine transform.

[0031] Preferably, the three-level data diversion rules include: emergency data: acceleration > 0.15 m / s² or CO2 mutation > 300 ppm, transmission delay < 2 seconds; important data: temperature fluctuation > ±3℃ / hour or low-frequency vibration during incubation, transmission per hour after compression rate ≥ 80%; normal data: periodically stored for 24 hours and then batch synchronized.

[0032] This invention provides a real-time monitoring system for bird nest breeding status based on the Internet of Things (IoT). It has the following beneficial effects:

[0033] 1. This invention utilizes a system with multiple sensors, including a piezoelectric sensor array in the sensing layer and a non-contact infrared thermometer array, to comprehensively collect multimodal data such as the number of egg turnings, nest temperature, acoustic signals, and CO2 concentration. The system preprocesses the data in real time using algorithms such as wavelet transform and Kalman filtering to extract key features such as vibration energy and temperature gradients. This reduces the daily data volume from 8-12GB to 200-500MB, ensuring both data comprehensiveness and improved processing efficiency, providing high-quality data support for subsequent risk assessment.

[0034] 2. This invention combines three-level data diversion rules to ensure the timely transmission of emergency data and the efficient transmission of important data, avoiding the impact of cloud latency. The cloud analysis layer uses models such as XGBoost and LSTM to quickly classify risk types and predict evolution trends, while the Bayesian model quantifies risk levels and generates suggestions. The feedback control layer then precisely regulates nest temperature and triggers directional sound waves to drive away predators, forming a closed loop of "monitoring-analysis-response" to address issues such as temperature control failure and predator intrusion in real time, effectively reducing the risk of nest abandonment.

[0035] 3. This invention fully considers minimizing interference with bird breeding through sensor deployment. For example, the piezoelectric sensor is encapsulated with 3D-printed tree branch textures, the MEMS microphone is disguised as a plant seed, and non-contact infrared thermometry achieves non-invasive monitoring through a folding arm bracket. This ensures the effectiveness of data collection while reducing interference with the natural behaviors of parent birds such as nesting and incubation. It achieves accurate monitoring while maintaining the natural ecological environment of bird nests, conforming to the core principles of wildlife protection. Attached Figure Description

[0036] Figure 1 This is a system architecture diagram of the present invention. Detailed Implementation

[0037] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Example:

[0039] Please see the appendix Figure 1 This invention provides a real-time monitoring system for bird nest breeding status based on the Internet of Things, including a sensing layer. The sensing layer comprises a deployed piezoelectric sensor group, a non-contact infrared temperature measurement array, a MEMS microphone array, and a gas sensor group, specifically deployed in:

[0040] In one specific embodiment, the piezoelectric sensor group is embedded in the biomimetic tree branch bark layer in the tree nest, encapsulated with 3D printed tree branch texture, and arranged in three groups in a 120° ring at a distance of 2-3 cm from the bottom of the nest. The ground nest is laid in the waterproof and breathable membrane interlayer at the bottom of the artificial nest box, covering an area ≥ 80% of the projected area of ​​the nest material. Its parameters are: sensitivity 0.5mV / g, response frequency band 0.1-200Hz, used to detect the number of times the egg body turns over (accuracy ±1 time / hour) and the vibration intensity of the parent bird in the nest (resolution 0.01m / s²).

[0041] The non-contact infrared temperature measurement array has its infrared temperature sensor installed 30cm above the nest top on a deployable folding arm bracket with the lens facing the center of the nest. The monitoring focus is calibrated by UWB positioning for active temperature measurement: scanning the temperature of 9 grid points on the surface of the nest material every hour (accuracy ±0.3℃). When the piezoelectric sensor detects that the parent birds have left the nest, the continuous temperature measurement mode is activated.

[0042] It also includes a MEMS microphone array, whose microphones are disguised as wireless nodes of plant seeds, suspended on support branches 0.5-1m outside the nest in a triangular layout, beamforming to directionally collect acoustic signals inside the nest, with a frequency range of 50-8000Hz and a dynamic sound pressure level of 30-120dB.

[0043] The gas sensor module is nested behind the ventilation grille on the side wall of the artificial nest box. The detection point is 10-15cm horizontally away from the biological activity area inside the nest, and the detection index is CO2.

[0044] The edge computing layer receives data collected by the perception layer in real time and performs real-time preprocessing on the raw multimodal data collected by the perception layer, including vibration, acoustics, and temperature, extracting key features such as vibration energy, acoustic signature features, and temperature gradients. This reduces the daily data volume from 8-12GB in traditional solutions to 200-500MB. Through local algorithms such as wavelet transform and MFCC recognition, the average processing latency is reduced to <2 seconds, avoiding the delays caused by round-trip transmission to the cloud. Furthermore, data grading and communication optimization rules are established, including a three-level data distribution rule.

[0045] Urgent data: Abnormal egg displacement, such as acceleration threshold > 0.15 m / s² or sudden change in CO2 concentration > 300 ppm, must be transmitted to the cloud immediately, with the highest priority.

[0046] Important data: Temperature fluctuations > ±3℃ / hour or low-frequency vibrations, such as the incubation rhythm of parent birds, are transmitted in batches per hour after compression.

[0047] Routine data: Baseline temperature and humidity, average noise, and other routine parameters are stored locally and synchronized as needed, with a 24-hour cycle.

[0048] The methods for processing raw multimodal data acquired by the perception layer include:

[0049] Piezoelectric vibration signal processing, based on time-frequency joint analysis using wavelet transform algorithm, locates short-time pulse signals in the 5-10Hz frequency band, lasting 0.5-2 seconds. The specific formula is as follows:

[0050] in:

[0051] ;

[0052] ;

[0053] ;

[0054] ;when A value greater than 0.1 is considered a valid hatching behavior.

[0055] Acoustic signal processing, using chick foraging sounds for identification, involves: identifying chick foraging sounds, using an FIR bandpass filter (200-8000Hz) to eliminate environmental wind noise interference, generating 13-dimensional MFCC coefficients in 25ms frames, and calculating the Euclidean distance to a preset template. A threshold >1.2 indicates an anomaly. The calculation process is as follows: First, the acoustic signal processed by the FIR bandpass filter is framed, with each frame lasting 25ms. Then, a Fourier transform is performed on each frame to obtain the power spectrum. Next, the power spectrum is passed through a Mel filter bank to obtain the Mel spectrum. Finally, the logarithm of the Mel spectrum is taken, and a discrete cosine transform is performed to obtain the 13-dimensional MFCC coefficients. The Euclidean distance calculation formula is as follows:

[0056] in, For Euclidean distance, For the generated j-th dimension MFCC coefficients, Let be the j-th dimension MFCC coefficient of the preset template. When... When this occurs, it is considered abnormal.

[0057] Temperature field data processing involves spatial interpolation of the measured temperature data using a bicubic spline algorithm to identify local hotspots where the temperature difference between adjacent points is greater than 1.5℃. The bicubic spline interpolation calculation process is as follows: Given the temperature values ​​and coordinates (x_k, y_k, T_k) of nine points (k=1,2,…,9), a bicubic spline function T(x,y) is constructed such that T(x_k,y_k)=T_k at each known point, while ensuring the continuity of the function and its first and second partial derivatives. This function can be used to calculate the temperature value at any point within the Bird's Nest area, thus determining whether the temperature difference between adjacent points is greater than 1.5℃.

[0058] Gas concentration analysis: Kalman filtering is used to remove sensor drift error, and the CO2 change rate is calculated. Threshold: ΔCO2 / 450ppm / Δt. Kalman filtering calculation process: Prediction stage: Update phase:

[0059] in, Let A be the predicted value at time k, and let A be the state transition matrix. Let B be the optimal estimate at time k-1, and let B be the control input matrix. This is the control input at time k-1. To predict the error covariance, Let Q be the estimation error covariance at time k-1, and let Q be the process noise covariance. Here, H is the Kalman gain, H is the observation matrix, and R is the observation noise covariance. Let k be the observation value at time k. Let be the optimal estimate at time k. Let be the covariance of the estimation error at time k, and I be the identity matrix. After Kalman filtering, the filtered CO2 concentration value is obtained. Then, the change per unit time, ΔCO2 / 450ppm / Δt, is calculated. When ΔCO2 / 450ppm / Δt is reached, it is judged as an anomaly. The edge computing layer can process the data in real time, with an average processing latency of <2 seconds, avoiding the delay of round-trip transmission to the cloud and ensuring the real-time monitoring of bird nest breeding status and risk response.

[0060] The cloud-based analysis layer includes a breeding risk assessment model. Edge preprocessing is used as input, and the feature vector after edge preprocessing contains rich parameters related to bird nest breeding, specifically temperature gradients, obtained through temperature field data processing, reflecting temperature differences in different locations within the nest; egg-turning frequency, obtained through piezoelectric vibration signal processing, reflecting the pattern of how many times the parent birds turn the eggs, the number of egg-turning times, the intensity of parent birds' nest vibration, and acoustic signal characteristics within the nest, extracted through acoustic signal processing; CO2 concentration and rate of change, etc., are used with an XGBoost classifier, achieving an accuracy of 92.7% and an F1 value of 0.89. The nest abandonment risk level includes 1-5 levels and suggested intervention measures. The core design logic is as follows: the first layer uses XGBoost to quickly classify risk types, such as temperature control failure, predator intrusion, and parent bird abandonment; the second layer uses LSTM to predict the risk evolution trend in the next 3-72 hours; and the third layer uses Bayesian methods to comprehensively calculate the risk level (quantified value R∈[0,1]) and generate treatment suggestions.

[0061] The training process specifically includes collecting a large amount of bird nest monitoring data under different bird species, different breeding stages including incubation and brooding periods, and different environmental conditions. This includes data on normal breeding status and various abnormal statuses, such as temperature control failure, predator intrusion, and parent birds abandoning the nest. This data covers the raw data collected by the perception layer and the feature vectors processed by the edge computing layer, and is also labeled with the corresponding risk type, risk level, and other information.

[0062] First, the training data is preprocessed, including data cleaning (removing noise and outliers) and data standardization (converting feature vectors of different magnitudes to the same numerical range). Then, the preprocessed data is input into the models, training XGBoost, LSTM, and Bayesian models respectively. For the XGBoost model, parameters such as the number of trees, depth, and learning rate are adjusted to improve the accuracy of risk type classification. For the LSTM model, parameters such as the number of network layers, hidden units, and iterations are optimized to improve the accuracy of risk evolution trend prediction. For the Bayesian model, the prior probability distribution is appropriately set to make risk level calculation more reasonable. During training, cross-validation is used to evaluate model performance, and parameters are continuously adjusted until the model reaches the preset performance indicators.

[0063] The calculation process specifically includes:

[0064] The first-layer XGBoost classifier's core idea is to construct multiple weak classifiers (decision trees) and combine them into a strong classifier. In this system, XGBoost is used to quickly classify risk types. It takes the feature vector output from the edge computing layer as input, and through analysis and learning of these features, it determines the type of risk that the nest may face during breeding, such as temperature control failure (when temperature fluctuations exceed the normal range and persist for a certain period of time), predator intrusion (detection of specific acoustic signals or abnormal vibrations), parent birds abandoning the nest (a significant reduction or disappearance of parent birds' nesting time), and zero egg-turning frequency. The objective function formula for XGBoost is:

[0065]

[0066] in, Let be the objective function. For the sample size, Let be the loss function for the i-th sample. Let be the true label (risk type) of the i-th sample. Let be the predicted risk type value of the i-th sample after prediction by t trees. For the number of trees, Let be the regularization term for the k-th tree. The formula is:

[0067]

[0068] in, and For regularization parameters, The number of leaf nodes in the tree. Let be the weight of the j-th leaf node. The calculation process is as follows: First, initialize the model by setting the predicted values ​​of all samples as initial values; then, construct decision trees sequentially, with each tree trained based on the residuals of the previous tree, using the difference between the true label and the current predicted value; during the construction of each tree, a greedy algorithm is used to select the optimal split point to minimize the objective function; after constructing t trees, the outputs of all trees are summed to obtain the final risk type prediction. By comparing the predicted values ​​with the true labels, the structure and parameters of the trees are continuously adjusted to improve classification accuracy.

[0069] The second-layer LSTM model: This model takes the risk type obtained from XGBoost classification and the related feature vector time series as input to predict the risk evolution trend over the next 3-72 hours. For example, for the risk of temperature control failure, LSTM can predict whether the temperature will deviate further from the normal range in the future, as well as the degree and speed of deviation, based on historical temperature change data and current temperature gradients; for the risk of predator intrusion, it can predict whether predators will continue to exist or reappear.

[0070] The core of LSTM lies in cell state and three gating mechanisms, and its key formulas are as follows:

[0071] Forget Gate Formula:

[0072] in, The output of the forget gate at time t. It is the sigmoid activation function. Here is the weight matrix for the forget gate. The hidden state at time t-1 The input feature vector at time t (containing risk type and related features) is given. This is the bias term for the forget gate.

[0073] Input gate formula:

[0074] in, The input gate output is at time t. Here is the weight matrix of the input gate. For the bias term of the input gate, Let t represent the candidate cell state at time t. The hyperbolic tangent activation function is used. This is the weight matrix for the candidate cell states. This is a bias term for the candidate cell state.

[0075] Cell state update formula:

[0076] in, Let t represent the cell state at time t. The cell state at time t-1. This is element-wise multiplication.

[0077] Output gate formula:

[0078] in, The output of the gate at time t is the output of the gate. Here is the weight matrix of the output gate. This is the bias term for the output gate. The hidden state at time t is used to predict future risk evolution trends.

[0079] The computation process: LSTM uses a forget gate to determine which information to discard from the cell state, an input gate to determine which new information to store in the cell state, then updates the cell state, and finally an output gate to determine the output value, which is based on the cell state. Through this gating mechanism, LSTM can effectively capture long-term dependencies in time series data and predict risk evolution trends over the next 3-72 hours.

[0080] The third layer of the Bayesian model: Based on Bayes' theorem, the Bayesian model calculates the posterior probability through prior probability and likelihood function, enabling inference of uncertain events. In this system, the Bayesian model integrates the risk type results of XGBoost classification and the risk evolution trend predicted by LSTM, combined with relevant prior knowledge, to calculate the posterior probability distribution of the risk level, thereby obtaining a comprehensive risk level quantification value. It also generates disposal suggestions.

[0081] Bayes' theorem formula is:

[0082]

[0083] in, In terms of risk type and the evolution trend of risks Risk level under the conditions is The posterior probability, In order to be at the risk level Risk type occurs under the following conditions and the evolution trend of risks likelihood function The risk level is The prior probability, Risk type and the evolution trend of risks The joint probability. Calculation process: First, determine the risk level based on historical data and expert knowledge. Prior probability Then, the risk type is obtained based on the XGBoost classification. and the risk evolution trend predicted by LSTM Calculate the likelihood function That is, the probability of the occurrence of this risk type and risk evolution trend under different risk levels is then calculated using the law of total probability. , Finally, the posterior probability is calculated using Bayes' theorem. The quantified value corresponding to the risk level with the highest posterior probability is taken as the comprehensive risk level quantified value.

[0084] The risk levels of nest abandonment and corresponding intervention measures are specifically classified as follows: Level 1 The risk is extremely low, and the nesting and breeding are proceeding normally. Recommendation: Maintain the usual monitoring frequency; no special intervention is required; Level 2. The risk is low; there may be slight abnormal signs, but the impact on breeding is minimal. Recommendation: Increase monitoring frequency appropriately, closely monitor changes in nest condition, and refrain from proactive intervention for now; Level 3 The risk is moderate, with obvious anomalies present. Failure to address this promptly could escalate the risk. Recommendation: Strengthen monitoring and arrange regular on-site inspections to assess the need for intervention, such as checking the nest temperature regulation module for proper functioning; Level 4. The risk is high; nesting and breeding are under significant threat, and nest abandonment is highly likely. Recommendations: Take immediate targeted intervention measures, such as adjusting the nest temperature to a suitable range using a nest temperature regulation module; if signs of predators are detected, activate a directional acoustic emitter to drive them away; Level 5 The risk is extremely high, with a very high risk of nest abandonment, and the bird's breeding process may be interrupted at any time. Recommendation: In addition to implementing Level 4 risk intervention measures, arrange for professional personnel to handle the situation on-site urgently, and if necessary, take measures such as artificial breeding assistance to try to save the bird's breeding process. Through multiple sensors in the sensing layer, multimodal data such as vibration, temperature, acoustics, and gas within the bird's nest can be collected simultaneously, comprehensively reflecting the breeding status of the nest and overcoming the shortcomings of traditional monitoring systems with their single data collection method. The breeding risk assessment model in the cloud analysis layer can accurately assess and predict the breeding risk of the bird's nest, and through the feedback control layer, it can achieve intelligent regulation of the nest environment and effective repulsion of predators, improving the success rate of bird breeding and providing strong technical support for wildlife conservation.

[0085] The feedback control layer includes a nest temperature regulation module, which is installed in the carbon fiber heating layer of the artificial nest box. It is separated from the nest material by a 2mm ceramic insulation layer and has a control accuracy of ±0.5℃, which is adjusted by PWM duty cycle. It also includes a directional sound wave transmitter: it generates an 18kHz pulse wave with a range of 3m, which does not affect the hearing of birds. The trigger condition is that infrared thermometry detects the body temperature characteristics of snakes (28-32℃ for more than 10 seconds).

[0086] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A real-time monitoring system for bird nest breeding status based on the Internet of Things, characterized in that: include: The sensing layer includes a piezoelectric sensor array, a non-contact infrared temperature measurement array, a MEMS microphone array, and a gas sensor module, which are deployed in different areas of the Bird's Nest's physical structure. The edge computing layer receives multimodal data from the perception layer in real time and performs local processing to extract vibration energy, acoustic features, temperature gradient and gas concentration change rate, and optimizes communication efficiency through a three-level data diversion rule. The cloud-based analysis layer includes a reproductive risk assessment module based on a hybrid model of XGBoost, LSTM, and Bayesian methods. It takes the preprocessed feature vector from the edge layer as input and outputs a quantitative value of the risk level and intervention suggestions. The feedback control layer, which includes a nest temperature regulation module and a directional acoustic wave transmitter, performs closed-loop control according to cloud commands.

2. The real-time monitoring system for bird nest breeding status based on the Internet of Things according to claim 1, characterized in that: The piezoelectric sensor array is embedded in the biomimetic tree bark layer or laid in the waterproof and breathable membrane interlayer at the bottom of the artificial nest box, with a sensitivity of 0.5mV / g and a response frequency band of 0.1-200Hz, covering ≥80% of the nest material projection area in a ring or plane. The infrared temperature measurement array is installed 30cm above the nest top, and the focus is calibrated by UWB positioning. It has a grid scanning mode of 9 points per hour and a continuous temperature measurement mode triggered by pressure. The MEMS microphone array is suspended from the support branch outside the nest in a triangular layout, and uses beamforming to collect acoustic signals in the frequency band of 50-8000Hz, with a dynamic sound pressure level of 30-120dB. The gas sensor module is nested at the rear of the ventilation grille of the artificial nest box, with the detection point 10-15cm horizontally away from the biological activity area.

3. The real-time monitoring system for bird nest breeding status based on the Internet of Things according to claim 1, characterized in that: The data processing method of the edge computing layer includes: Piezoelectric vibration signal processing: Short-time pulses in the 5-10Hz frequency band are located using wavelet transform; the determination formula is as follows: , A value greater than 0.1 is recorded as a valid hatching action. Acoustic signal processing: After eliminating environmental interference using an FIR bandpass filter, 13-dimensional MFCC coefficients are generated, and the anomaly threshold is calculated using the Euclidean distance formula. in, For Euclidean distance, For the generated j-th dimension MFCC coefficients, For the j-th dimension MFCC coefficient of the preset template, when When this happens, it is judged as abnormal; Temperature field processing: A continuous temperature field is constructed by bicubic spline interpolation of the temperature measurement data, and local hot spots are marked when the temperature difference is >1.5℃; Gas analysis: Sensor drift is eliminated through Kalman filtering algorithm, and the CO2 change rate is dynamically calculated.

4. The real-time monitoring system for bird nest breeding status based on the Internet of Things according to claim 1, characterized in that: In the cloud-based analysis layer, the XGBoost classifier first sets the predicted values ​​of all samples as initial values; then it successively builds decision trees, with each tree trained based on the residual of the previous tree, using the difference between the true label and the current predicted value; when building each tree, a greedy algorithm is used to select the optimal split point to minimize the objective function; after building t trees, the outputs of all trees are summed to obtain the final risk type prediction value.

5. The real-time monitoring system for bird nest breeding status based on the Internet of Things according to claim 1, characterized in that: The LSTM uses a forget gate to determine which information to discard from the cell state, an input gate to determine which new information to store in the cell state, then updates the cell state, and finally uses an output gate to determine what value to output, with the output value based on the cell state.

6. The real-time monitoring system for bird nest breeding status based on the Internet of Things according to claim 5, characterized in that: The forget gate formula is as follows: The output of the forget gate at time t. It is the sigmoid activation function. Here is the weight matrix for the forget gate. The hidden state at time t-1 Let be the input feature vector at time t. For the bias term of the forget gate; The input gate formula: in, The input gate output is at time t. Here is the weight matrix of the input gate. For the bias term of the input gate, Let t represent the candidate cell state at time t. The hyperbolic tangent activation function is used. This is the weight matrix for the candidate cell states. This is a bias term for the candidate cell state.

7. The real-time monitoring system for bird nest breeding status based on the Internet of Things according to claim 5, characterized in that: The cell state update formula is as follows: Let t represent the cell state at time t. The cell state at time t-1. This is element-wise multiplication.

8. The real-time monitoring system for bird nest breeding status based on the Internet of Things according to claim 1, characterized in that: The feedback control layer includes a nest temperature regulation module and a directional acoustic wave transmitter. The nest temperature regulation module is composed of a carbon fiber heating film and a ceramic heat insulation layer. It uses PWM duty cycle adjustment to achieve a temperature control accuracy of ±0.5℃. The directional acoustic wave transmitter responds to the infrared thermometry detection of a 28-32℃ temperature for 10 seconds and emits an 18kHz pulse wave to drive away predators.

9. The real-time monitoring system for bird nest breeding status based on the Internet of Things according to claim 3, characterized in that: The MFCC coefficient generation method specifically includes: processing the bandpass filtered acoustic signal in 25ms frames, performing Fourier transform on each frame to obtain the power spectrum, extracting the logarithmic energy through a 40-channel Mel filter bank, and taking the first 13 coefficients as the voiceprint feature vector after performing discrete cosine transform.

10. The real-time monitoring system for bird nest breeding status based on the Internet of Things according to claim 1, characterized in that: The three-level data diversion rules include: Emergency data: Acceleration > 0.15 m / s² or CO2 mutation > 300 ppm, transmission delay < 2 seconds; Important data: Temperature fluctuation > ±3℃ / hour or low-frequency vibration during incubation, transmission per hour after compression rate ≥ 80%; Normal data: Stored periodically for 24 hours and then synchronized in batches.

Images