Method and system for predicting sow ovulation window

Abstract

This invention relates to the field of intelligent equipment technology for livestock reproduction, specifically to a method and system for predicting the ovulation window of sows. The method includes the following steps: acquiring ultrasound radio frequency signals from the sow's reproductive tract and extracting backscattering statistical feature data of the follicle region from the ultrasound radio frequency signals; acquiring electrical impedance spectrum signals from the sow's reproductive tract and fitting the electrical impedance spectrum signals to obtain Cole-Cole model parameter data characterizing the electrophysiological properties of the follicle wall tissue; inputting the backscattering statistical features and the Cole-Cole model parameters into a pre-trained time-series prediction model to output the ovulation probability of the sow within a preset future time period. This invention overcomes the limitations of single morphological indicators, combining ultrasound backscattering microscopic features with follicle electrophysiological parameters. Through a physically constrained model, the ovulation window prediction error is compressed to ±4 hours, achieving high accuracy in distinguishing between mature and abnormal follicles and reducing the probability of misjudgment using traditional methods.

Description

Technical Field

[0001] This invention relates to the field of intelligent equipment technology for livestock reproduction, specifically to a method and system for predicting the ovulation window of sows. Background Technology

[0002] my country's pig farming has entered a stage of large-scale, batch production. In the management of large-scale, batch-based sow production, accurately identifying the sow's ovulation window is crucial for timely insemination and improving conception rates. The optimal insemination window for sows is only 12-1-24 hours before ovulation; deviations in insemination timing can lead to decreased conception rates and fewer piglets. Current technology mainly relies on technicians using handheld veterinary ultrasound transabdominal or transrectal scanning, combined with manual interpretation of follicle diameter, and behavioral indicators such as standing reflexes for estrus detection. However, the aforementioned monitoring methods have several problems: Ultrasound measurements are single, static snapshots and cannot utilize continuous evolutionary information, leading to prediction errors for the ovulation window typically exceeding one or two days. Secondly, simple diameter measurement fails to reflect the maturity of the follicle's microstructure; follicles larger than 10mm in diameter may be cysts or arrested follicles, easily causing misjudgment. Furthermore, image interpretation is disconnected from physiological indicators such as hormones and behavior, making it impossible to rule out physiological misjudgments caused by false estrus or follicle arrest. Additionally, manual judgment relies on experience, has low standardization, and requires long training periods for personnel. Moreover, while laboratory hormone testing can control prediction errors within 6 hours, it is costly, time-consuming, complex, and cannot provide timely on-site monitoring.

[0003] In summary, existing technologies cannot balance accuracy, practicality, and low cost, making it difficult to meet the production needs of large-scale pig farms. Summary of the Invention

[0004] To address the aforementioned deficiencies and problems, this invention provides a method and system for predicting the ovulation window of sows. It characterizes follicle maturity from the dual perspectives of the electrical characteristics of the reproductive tract and the acoustic characteristics of follicles, and constructs a predictive model with physiological interpretability, thereby compressing the ovulation window prediction error to within ±4 hours, while meeting the needs of low-cost operation on pig farms.

[0005] The solution adopted by this invention to solve its technical problem is: a method for predicting the ovulation window of sows, comprising the following steps: S1. Acquire ultrasound radio frequency signals of the sow's reproductive tract and extract backscatter statistical feature data of the follicle region from the ultrasound radio frequency signals; S2. Collect electrical impedance spectrum signals of the sow's reproductive tract, fit the electrical impedance spectrum signals, and obtain Cole-Cole model parameter data characterizing the electrophysiological properties of the follicle wall tissue. S3. Input the backscattering statistical features and the Cole-Cole model parameters into the pre-trained time series prediction model, and output the ovulation probability of the sow in the future preset time period; The time-series prediction model is a physical information neural network, and the loss function of the physical information neural network includes physical constraint terms constructed based on electrophysiological laws.

[0006] Furthermore, in step S2, the step of acquiring the electrical impedance spectrum signal of the sow's reproductive tract is as follows: S21. Place the bipolar concentric ring electrode in the vaginal fornix or external os of the cervix of the sow, apply a sinusoidal sweep excitation signal with a frequency range of 100Hz to 1MHz, and collect the complex impedance data at the corresponding frequency. S22. The Cole-Cole model parameters are obtained by nonlinear least squares fitting. The Cole-Cole model parameters include at least the high-frequency limiting resistance RO, the low-frequency limiting resistance R∞, the relaxation time constant τ, and the dispersion coefficient α.

[0007] Furthermore, the backscattering statistical characteristics include at least one of the backscattering coefficients: skewness, kurtosis, and energy entropy. The time-series prediction model includes two parallel feature encoding branches and a gated fusion unit. The first encoding branch is used to process the Cole-Cole model parameter data, the second encoding branch is used to process the backscattering statistical feature data, and the gated fusion unit is used to adaptively allocate the feature weights of the two branches according to the quality of the dual-mode signal.

[0008] Furthermore, the physical constraints include Cole-Cole equation residual constraints and temporal monotonicity constraints; The Cole-Cole equation residual constraint is used to limit the deviation between the theoretical impedance value corresponding to the Cole-Cole model parameters output by the model and the actual measured electrical impedance spectrum data. The temporal monotonicity constraint is used to penalize the decreasing trend of the high-frequency limiting resistance R0 over time during the 24-48 hours before ovulation.

[0009] Furthermore, the training steps of the time series prediction model include: Pre-training was conducted using vertical monitoring data from a group of no fewer than 1,000 sows; For individual sows, individual monitoring data from three or more consecutive estrous cycles are used to fine-tune the pre-trained model through transfer learning to adapt to individual physiological differences; After deployment, incremental learning is triggered to update the model parameters every time 50 sows' conception confirmation data are accumulated.

[0010] A sow ovulation window prediction system includes an ultrasound detection module, an electrical impedance measurement module, and an edge computing terminal; The ultrasonic detection module is used to emit ultrasonic waves and collect ultrasonic radio frequency signals from the sow's reproductive tract. The impedance measurement module includes a bipolar concentric ring electrode and an impedance measurement front end, used to collect the impedance spectrum signal of the sow's reproductive tract; The edge computing terminal is electrically connected to the ultrasonic detection module and the electrical impedance measurement module, respectively, and the edge computing terminal is configured to perform the method according to any one of claims 1-5.

[0011] Furthermore, the impedance measurement module is integrated inside the handle of the veterinary ultrasound probe, and an electromagnetic shielding layer is provided between the electrode surface of the bipolar concentric ring electrode and the transducer array of the ultrasound probe; the edge computing terminal is configured to execute a time-division multiplexing control strategy to control the impedance measurement module and the ultrasound detection module to work alternately in different time slots.

[0012] Furthermore, it also includes an interactive terminal, which is communicatively connected to the edge computing terminal. The interactive terminal is used to display ultrasound images, impedance spectrum curves, and ovulation probability curves, and to issue a counterweight reminder when the ovulation probability exceeds a preset threshold. The edge computing terminal is also used to detect the contact impedance between the bipolar concentric ring motor and the tissue in real time. When the contact impedance is greater than 10kΩ, it generates a probe adjustment and lifting signal and sends it to the interactive terminal.

[0013] The beneficial effects of this invention are as follows: This invention overcomes the limitations of single-morphological indicators, combining the microscopic features of ultrasound backscattering with follicular electrophysiological parameters. Through a physically constrained model, the prediction error of the ovulation window is compressed to ±4 hours, achieving high accuracy in distinguishing mature and abnormal follicles and reducing the probability of misjudgment compared to traditional methods. This invention employs a dual-modal gating fusion mechanism, which can adaptively allocate weights based on signal quality. It innovates a three-level training strategy of group pre-training, individual fine-tuning, and incremental learning, effectively solving the cold start problem for gilts. Prediction accuracy stabilizes and reaches the target after the third estrous cycle. This invention integrates the electrical impedance module with the ultrasound probe, using electromagnetic shielding and time-division multiplexing to resolve signal interference. The short single-detection time meets the production needs of large-scale pig farms. Attached Figure Description

[0014] Figure 1 A diagram showing the overall hardware architecture of a sow ovulation window prediction system; Figure 2 This is a schematic diagram of the internal structure of the integrated detection probe; Figure 3 The timing diagram for the time-division multiplexing control of the integrated detection probe; Figure 4The flowchart shows the extraction process for statistical features of backscattering. Figure 5 The overall architecture and data flow diagram of the physical information neural network; Figure 6 A time-series curve of ovulation probability in representative sows, and a comparison of changes in serum LH concentration and follicle diameter; Figure 7 The prediction error convergence curve for 20 replacement gilts during the cold start phase.

[0015] In the diagram: 1. Integrated detection probe; 11. Bipolar concentric ring electrode; 12. Electromagnetic shielding layer; 13. Ultrasonic transducer array; 14. Shielded cable; 2. Edge computing terminal; 3. Interactive terminal. Detailed Implementation

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0017] Please see Figure 1-7 This invention provides a technical solution for a method and system for predicting the ovulation window of sows: Example

[0018] like Figure 1 As shown, the sow ovulation window prediction system mainly comprises four functional entities: an ultrasound detection module, an electrical impedance measurement module, an edge computing terminal, and an interactive terminal. The ultrasound detection module and the electrical impedance measurement module are integrated into the same handheld housing via an integrated detection probe 1, and both are connected to the edge computing terminal via shielded cables 14. The edge computing terminal performs data aggregation and model inference via a local bus. The interactive terminal communicates with the edge computing terminal via wired or local wireless means, and is used to present ovulation probability curves to technicians and issue mating reminders.

[0019] like Figure 2 As shown, the handle of the ultrasound probe integrates a bipolar concentric ring electrode 11 with an impedance measurement module. The electrode is made of medical-grade 316L stainless steel with a gold-plated surface, and has an outer diameter of 8mm, an inner ring diameter of 3mm, and a ring spacing of 1mm. An electromagnetic shielding layer 12 made of copper foil or conductive polymer is placed between the electrode surface and the ultrasound transducer array 13 of the ultrasound probe to suppress electromagnetic interference from high-voltage pulses during ultrasound emission on the impedance measurement signal. During operation, the technician inserts the probe through the vagina, positioning the bipolar concentric ring electrode at the vaginal fornix or the external os of the cervix. These two locations are chosen because: the vaginal fornix is ​​close to the pelvic inlet, near the ovary and fimbriae of the fallopian tubes, allowing the impedance signal from the reproductive tract wall tissue to couple with changes in electrical properties caused by blood perfusion and extracellular matrix changes in the ovarian region; the mucosa at the external os of the cervix is ​​dense, providing good contact stability and making it suitable as a reference point for impedance baseline measurement.

[0020] To resolve the signal conflict between ultrasound and impedance measurement, the edge computing terminal incorporates a time-division multiplexing control strategy, sending interlocking trigger signals to both the ultrasound front-end and the impedance front-end, enabling them to work alternately in different time slots. For example, a 200ms cycle is divided into two parts: the first 150ms is for ultrasound transmission and echo reception, and the last 50ms is for impedance excitation and sampling, thus avoiding mutual interference of electroacoustic signals in the physical channel.

[0021] like Figure 3 As shown, the impedance measurement module is centered around a front-end impedance measurement chip. The edge computing terminal controls this chip to output a constant-voltage sinusoidal sweep frequency excitation signal, covering a frequency range of 100Hz to 1MHz. After passing through a series resistor and coupling capacitor, the signal is injected into the tissue through the outer loop of the electrode, while the inner loop collects the response current. The front-end chip internally performs current-to-voltage conversion and synchronous demodulation, outputting the real and imaginary parts of the impedance. A bidirectional clamping protection circuit is installed at the electrode interface to absorb spike interference that may be coupled from electrostatic discharge and ultrasonic high-voltage pulses. After the signal is filtered by a fourth-order Butterworth low-pass filter to remove high-frequency noise, it is sent to the front-end chip for analog-to-digital conversion.

[0022] During each detection, the system performs measurements at three frequency points: 1 kHz, 10 kHz, and 100 kHz, with a dwell time of no more than 2 seconds at each frequency point. After obtaining the complex impedance spectroscopy data, a nonlinear least-squares fitting is performed using the Cole-Cole equivalent circuit model to extract the following four characteristic parameters: high-frequency limiting resistance R0, reflecting the extracellular fluid ion concentration and channel density; low-frequency limiting resistance R... ∞ The impedance characteristic parameters reflect cell membrane integrity and lipid bilayer impedance; the relaxation time constant τ reflects the average cell diameter and membrane capacitance effect; and the diffusion coefficient α reflects the heterogeneity of cell size distribution. These four parameters together constitute the time-series data of impedance characteristic parameters, which serve as inputs for subsequent prediction models.

[0023] like Figure 4 As shown, the raw radio frequency output signal of the ultrasound probe is not processed by envelope detection and is directly amplified to a suitable level by a low-noise amplifier. After being filtered out by a passive bandpass filter to remove out-of-band noise, it is digitized by an analog-to-digital converter. The digitized signal is then sent to a programmable logic device, where quadrature demodulation and downsampling are performed internally to generate baseband data or a downsampled radio frequency envelope sequence. Finally, it is cached in the memory of the edge computing terminal through a high-speed data interface.

[0024] See Figure 5The extraction process of backscatter statistical features is as follows: In the ultrasound image, the region of interest (ROI) is manually or semi-automatically defined, centered on the target follicle region; simultaneously, a reference background region is delineated within the surrounding tissue at the same depth. Hilbert transforms are performed on the radiofrequency signals of both regions to extract the signal envelopes, and then the skewness, kurtosis, and energy entropy of the envelope probability density function are calculated. Skewness reflects the asymmetry of cell debris distribution in the follicular fluid, kurtosis reflects the degree of protein concentration aggregation, and energy entropy reflects the randomness of the scattered signal. Finally, the skewness difference and kurtosis difference between the two regions are calculated to eliminate the influence of differences in tissue acoustic baselines between different individuals, forming time-series data of backscatter statistical features.

[0025] This embodiment uses a physical information neural network as the time series prediction model. This network consists of five layers: an input layer, an encoding layer, a physical constraint layer, a fusion layer, and a prediction layer, as follows: Figure 6 As shown. The input layer receives two sets of parallel timing data: one set is the timing data of impedance characteristic parameters, including the high-frequency limiting resistance R. O Low-frequency limiting resistance R ∞ One set consists of the relaxation time constant τ and the diffusion coefficient α; the other set consists of ultrasonic characteristic time series data, including backscattering coefficient, skewness, kurtosis and energy entropy.

[0026] The encoding layers perform deep encoding on two sets of temporal features. The impedance feature encoding branch employs a bidirectional long short-term memory network superimposed with a multi-head self-attention mechanism to capture the long-term dependencies of slowly evolving electrical features during follicle development; the ultrasound feature encoding branch employs a temporal convolutional network superimposed with a temporal attention mechanism to address local rapid fluctuations in ultrasound scattering signals caused by sow respiration or intestinal peristalsis. Both encoders output deep temporal feature vectors of the same dimension.

[0027] The physical constraint layer embeds two types of physical constraints based on electrophysiological knowledge. The first type is the Cole-Cole equation residual constraint: the impedance parameters output by the encoding layer are substituted back into the Cole-Cole theoretical equation to calculate the theoretical complex impedance value. The deviation between this theoretical and measured complex impedance values ​​is used as a penalty term in the loss function, forcing the parameter combination output by the network to conform to the physical laws of dielectric relaxation and avoiding physically unreasonable prediction results. The second type is the temporal monotonicity constraint: according to the electrophysiological laws of the terminal stage of follicular development, in the 24 to 48 hours before ovulation, the granulosa cell layer of the follicular wall thickens, intercellular junctions increase, and the extracellular fluid resistance R0 should show a monotonically decreasing trend. When the R0 value predicted by the model shows an abnormal increase that violates this law, a penalty term is introduced into the loss function, thereby improving the stability of the prediction of the critical ovulation window period.

[0028] The fusion layer employs gated fusion units, which adaptively allocate weights based on the quality of the dual-modal features at the current moment. When the signal quality of a certain mode deteriorates due to poor electrode contact or ultrasonic artifacts, the gating mechanism automatically reduces the contribution of that mode and increases the weight of the other mode, ensuring the overall robustness of the system.

[0029] The prediction layer consists of a fully connected layer and an activation function, and ultimately outputs the ovulation probability value within a preset time period in the future.

[0030] The model training employs a strategy of group pre-training followed by individual fine-tuning. First, pre-training is performed using historical data from the group (complete estrous cycle records of no fewer than 1000 sows) to learn common patterns at the group level. When deployed to specific sows, the pre-trained model is fine-tuned using monitoring data from three or more consecutive cycles for that individual, adapting to differences in individual anatomical structure and physiological characteristics through transfer learning.

[0031] Example 2: Combined transabdominal and transvaginal detection mode for multiparous sows This embodiment is applicable to multiparous sows weighing over 180kg, with thicker abdominal fat, allowing for clear localization of the ovarian region via transabdominal ultrasound. The operating steps are as follows: The sow is lightly restrained in a gestation crate. A technician holds an integrated detection probe and first scans the ovarian region via the abdomen, identifying the largest follicle in the ultrasound image and marking its spatial location. Then, the probe is slowly pushed into the vagina, positioning the bipolar concentric ring electrode at the vaginal fornix, approximately aligned with the target follicle. Time-division multiplexing control is activated to simultaneously acquire impedance data and ultrasound radiofrequency signals, with a single complete detection time controlled within 30 seconds. The edge computing terminal performs feature extraction, temporal alignment, and model inference in real time, outputting the ovulation probability time-series curve for the next 4 hours.

[0032] To verify the effectiveness of the technology, 30 multiparous Landrace sows were selected as the experimental group in a large-scale pig farm, and 30 sows were selected as the control group (using traditional B-ultrasound diameter measurement combined with behavioral assessment). The experimental group underwent the combined testing of this embodiment every 4 hours during estrus, while the control group had the same technician determine the insemination time according to standard procedures. Using simultaneous blood sampling to detect the LH hormone peak and the actual ovulation time confirmed by continuous B-ultrasound monitoring within 24 hours as the gold standard, statistical results showed that the ovulation window prediction error in the experimental group was ±4 hours, and the accuracy rate in distinguishing mature follicles from cysts reached 92%; the prediction error in the control group was ±18 hours, and the accuracy rate in distinguishing between mature follicles and cysts was approximately 65%. After batch insemination, the conception rate in the experimental group reached 92%, while it was 82% in the control group.

[0033] like Figure 7As shown, the ovulation probability time-series curve of a representative sow in the experimental group, compared with the changes in serum LH concentration and follicle diameter, shows that the ovulation probability curve begins to rise sharply about 2 to 4 hours before the LH peak, providing a clear mating decision window before the follicle diameter reaches its peak; while the traditional diameter measurement method cannot effectively distinguish whether the follicle subsequently ovulates normally or becomes cystic.

[0034] Example 3: Cold Start and Individual Adaptation Mode of Replacement Gilts This embodiment is applicable to gilts without historical monitoring data (monitoring at puberty), solving the cold start problem of the model caused by the lack of training samples for newly added individuals. The operation process is divided into two stages: Phase 1 (1st to 2nd estrous cycle): Due to the deep location of the ovaries and thin abdominal fat in gilts, the quality of transabdominal ultrasound images is unstable. In this phase, only electrical impedance data is collected; ultrasound guidance is not mandatory, simplifying the operation. At this time, the edge computing terminal uses the population baseline model for prediction. This model, pre-trained from population data of over 1000 gilts, can capture breed-level common patterns. The prediction error during the cold start phase is approximately ±8 hours. Phase 2 (from the 3rd cycle onwards): As historical data for the gilt accumulates, the system gradually introduces ultrasound backscatter statistical moment data to activate the individual adaptive model. Based on the temporal impedance data from the first two cycles, ultrasound feature data is added from the third cycle onwards. The population baseline model is fine-tuned through transfer learning, and the prediction error converges to ±4 hours after the third cycle.

[0035] like Figure 7 As shown, the prediction error convergence of 20 replacement gilts during the cold start phase shows that when only electrical impedance data was used in the first and second cycles, the error was distributed in the range of 6 to 10 hours. After introducing bimodal data and activating the individual adaptive model in the third cycle, the error decreased significantly and stabilized within ±4 hours in the fourth and fifth cycles, proving that the group-individual transfer learning strategy can effectively solve the cold start problem.

[0036] It should be noted that the above embodiments and accompanying drawings are merely illustrative examples of the core principles and key structures of the present invention, "A Method and System for Predicting Ovulation Window in Sows." The accompanying drawings are simplified schematic diagrams, intended to clearly illustrate the structural, process, or data flow relationships related to the innovative points of the technical solution, and are not intended to limit the complete form of the actual product. This specification focuses on the innovative technical means necessary to achieve the invention's objectives and solve the technical problems; while auxiliary or common-sense details that can be implemented by those skilled in the art without creative effort are not elaborated upon, they should be understood as naturally included in the specific implementation of this invention and fall within the scope of protection and implementation of this technical solution.

Claims

1. A method for predicting the ovulation window of sows, characterized in that, Includes the following steps: S1. Acquire ultrasound radio frequency signals of the sow's reproductive tract and extract backscatter statistical feature data of the follicle region from the ultrasound radio frequency signals; S2. Collect electrical impedance spectrum signals of the sow's reproductive tract, fit the electrical impedance spectrum signals, and obtain Cole-Cole model parameter data characterizing the electrophysiological properties of the follicle wall tissue. S3. Input the backscattering statistical features and the Cole-Cole model parameters into the pre-trained time series prediction model, and output the ovulation probability of the sow in the future preset time period; The time-series prediction model is a physical information neural network, and the loss function of the physical information neural network includes physical constraint terms constructed based on electrophysiological laws.

2. The method for predicting the ovulation window of sows according to claim 1, characterized in that, In step S2, the step of obtaining the electrical impedance spectrum signal of the sow's reproductive tract is as follows: S21. Place the bipolar concentric ring electrode in the vaginal fornix or external os of the cervix of the sow, apply a sinusoidal sweep excitation signal with a frequency range of 100Hz to 1MHz, and collect the complex impedance data at the corresponding frequency. S22. Obtain the Cole-Cole model parameters through nonlinear least squares fitting. The Cole-Cole model parameters include at least the high-frequency limiting resistance R. O Low-frequency limiting resistance R ∞ The relaxation time constant τ and the dispersion coefficient α.

3. The method for predicting the ovulation window of sows according to claim 1, characterized in that, The backscattering statistical characteristics include at least one of the backscattering coefficients: skewness, kurtosis, and energy entropy. The time-series prediction model includes two parallel feature encoding branches and a gated fusion unit. The first encoding branch is used to process the Cole-Cole model parameter data, the second encoding branch is used to process the backscattering statistical feature data, and the gated fusion unit is used to adaptively allocate the feature weights of the two branches according to the quality of the dual-mode signal.

4. The method for predicting the ovulation window of sows according to claim 3, characterized in that, The physical constraints include Cole-Cole equation residual constraints and temporal monotonicity constraints; The Cole-Cole equation residual constraint is used to limit the deviation between the theoretical impedance value corresponding to the Cole-Cole model parameters output by the model and the actual measured electrical impedance spectrum data. The temporal monotonicity constraint is used to penalize the decreasing trend of the high-frequency limiting resistance R0 over time during the 24-48 hours before ovulation.

5. The method for predicting the ovulation window of sows according to claim 1, characterized in that, The training steps of the time series prediction model include: Pre-training was conducted using vertical monitoring data from a group of no fewer than 1,000 sows; For individual sows, individual monitoring data from three or more consecutive estrous cycles are used to fine-tune the pre-trained model through transfer learning to adapt to individual physiological differences; After deployment, incremental learning is triggered to update the model parameters every time 50 sows' conception confirmation data are accumulated.

6. A sow ovulation window prediction system, characterized in that, It includes an ultrasonic detection module, an electrical impedance measurement module, and an edge computing terminal; The ultrasonic detection module is used to emit ultrasonic waves and collect ultrasonic radio frequency signals from the sow's reproductive tract. The impedance measurement module includes a bipolar concentric ring electrode and an impedance measurement front end, used to collect the impedance spectrum signal of the sow's reproductive tract; The edge computing terminal is electrically connected to the ultrasonic detection module and the electrical impedance measurement module, respectively, and the edge computing terminal is configured to perform the method according to any one of claims 1-5.

7. The sow ovulation window prediction system according to claim 6, characterized in that, The electrical impedance measurement module is integrated inside the handle of the veterinary ultrasound probe, and an electromagnetic shielding layer is provided between the electrode surface of the bipolar concentric ring electrode and the transducer array of the ultrasound probe; the edge computing terminal is configured to execute a time-division multiplexing control strategy to control the electrical impedance measurement module and the ultrasound detection module to work alternately in different time slots.

8. The sow ovulation window prediction system according to claim 6, characterized in that, It also includes an interactive terminal, which is communicatively connected to the edge computing terminal. The interactive terminal is used to display ultrasound images, impedance spectrum curves, and ovulation probability curves, and to issue a counterweight reminder when the ovulation probability exceeds a preset threshold. The edge computing terminal is also used to detect the contact impedance between the bipolar concentric ring motor and the tissue in real time. When the contact impedance is greater than 10kΩ, it generates a probe adjustment and lifting signal and sends it to the interactive terminal.

Images