Flexible monitoring sensor for actively sounding fish and signal analysis method

Abstract

The invention discloses a flexible monitoring sensor for actively sounding fish and a signal analysis method, and relates to the technical field of aquaculture, and the flexible monitoring sensor mainly comprises a flexible substrate, an interdigital electrode arranged on the substrate, and a piezoresistive sensitive layer covering the electrode. The sensitive layer is made of a PDMS and graphene composite material, and the surface of the sensitive layer is provided with a regularly arranged micro-pyramid structure array. The sensor is attached to the body surface of a fish sounding part, the micro-pyramid structure responds to body surface sound vibration in a preset frequency range to generate deformation, and the sound vibration is converted into a resistance change signal between the interdigital electrodes by utilizing the piezoresistive effect of the composite material. Through the design of the PDMS-graphene composite material and the surface micro-pyramid structure, high-fidelity and direct capture of body surface vibration signals generated by initiative sounding of individual fishes is realized, and the problems of group sound mixing and weak signal relevance in a traditional method are effectively avoided.

Description

Technical Field

[0001] This invention relates to the field of aquaculture technology, specifically to a flexible monitoring sensor and signal analysis method for actively vocalizing fish. Background Technology

[0002] As the global aquaculture industry develops towards large-scale and intensive operations, fish face increasingly complex stressors during farming and transportation, posing a severe challenge to their physiological health and product quality. Against this backdrop, developing real-time, accurate, and non-invasive physiological monitoring technologies has become a key requirement for ensuring the health of farmed animals and improving production efficiency. Fish species capable of active vocalization, such as the large yellow croaker, generate specific acoustic signals driven by the tympanic muscles on the swim bladder side, providing a unique biological basis for directly reflecting their physiological and behavioral states through acoustic means. Currently, research on fish physiological monitoring technologies mainly focuses on the following directions: First, wearable flexible sensing technologies for fish, such as strain or impedance sensors based on liquid metal composite materials, can be attached to the surface of fish to monitor changes in morphology and electrical properties caused by stress. However, these sensors capture indirect, derived physiological parameters with weak direct correlation to active behaviors such as feeding and courtship. Second, underwater acoustic detection technologies, such as hydrophones based on materials like graphene, are mainly used to collect broadband environmental underwater acoustic signals. However, in intensive aquaculture environments, it is difficult to separate and locate the vocalizations of specific individuals from the mixed sound field of the group, making individualized monitoring impossible. Third, optical monitoring methods, such as applying photoplethysmography to fish heart rate measurement, have performance highly dependent on the light transmittance of the water, resulting in a significant decrease in reliability in actual turbid and variable aquaculture environments. In addition, there are flexible devices that integrate multiple sensing functions. Their designs are generally more general and not specifically optimized for capturing and interpreting the specific biological signal of fish vocalization. In summary, existing technologies either fail to directly capture acoustic signals strongly correlated with fish's active behavior, are subject to interference from the aquaculture environment, or cannot distinguish individual sound sources. They also lack an integrated design from dedicated sensing to real-time analysis, making it difficult to meet the management needs for continuous, stable, and real-time intelligent judgment of the physiological state of individual fish in actual complex aquaculture scenarios. Summary of the Invention

[0003] To better capture the acoustic signals of fish's active behavior and to make continuous, stable, and real-time intelligent judgments on the physiological state of individual fish, this invention proposes a flexible monitoring sensor for actively vocalizing fish, comprising: Flexible substrate; Interdigitated electrodes disposed on the flexible substrate; and a piezoresistive sensitive layer covering the interdigitated electrodes; The piezoresistive sensitive layer is formed of a composite material of polydimethylsiloxane and graphene, and the surface of the piezoresistive sensitive layer has a regularly arranged array of micro pyramid structures. The flexible monitoring sensor is attached to the surface of the vocalization part of the active vocalizing fish. The micro pyramid structure array on the surface of the piezoresistive sensitive layer deforms in response to acoustic vibrations within a preset frequency range on the vocalization part surface. Through the piezoresistive effect of the piezoresistive sensitive layer, the acoustic vibrations are converted into resistance change signals between the interdigital electrodes.

[0004] This invention achieves high-fidelity, direct capture of surface vibration signals generated by the active vocalization of individual fish through the design of polydimethylsiloxane-graphene composite material and surface micropyramid structure, effectively avoiding the problems of mixed group sounds and weak signal correlation in traditional methods.

[0005] Furthermore, the flexible substrate is a polyethylene terephthalate film.

[0006] Furthermore, the interdigitated electrode is a chromium / gold electrode.

[0007] Furthermore, the preset frequency range is the vibration frequency band of the body surface when the vocalization part of the actively vocalizing fish emits sound.

[0008] Furthermore, it also includes a signal processing module, which is electrically connected to the interdigitated electrodes and is used to acquire the resistance change signal.

[0009] Furthermore, it also includes a waterproof encapsulation layer that covers the flexible substrate, interdigitated electrodes, piezoresistive sensing layer, and signal processing module.

[0010] This invention also includes a signal analysis method for a flexible monitoring sensor for actively vocalizing fish, comprising the following steps: S1: Acquire the resistance change signal after conversion by the flexible monitoring sensor; S2: Perform adaptive baseline estimation on the resistance change signal using a median filter, and subtract the baseline estimate from the resistance change signal; S3: Perform a short-time Fourier transform on the baseline-corrected signal to generate the corresponding time-spectrum graph; S4: Calculate the sound pressure level based on the amplitude value of the time-frequency spectrum and synthesize a multi-channel feature image containing time, frequency and sound pressure level information; S5: Input the multi-channel feature image into a pre-trained lightweight convolutional neural network model to obtain the classification results of the physiological state corresponding to the fish's active vocalization behavior.

[0011] Furthermore, in step S2, the window size of the median filter is determined based on the data length of the resistance change signal, and the window size is set to an odd number not greater than 201.

[0012] Furthermore, in step S4, the sound pressure level is calculated using the following formula: In the formula, Sound pressure level, The amplitude value is the short-time Fourier transform value. To prevent small constants from taking the logarithm of zero, This is the offset coefficient.

[0013] Furthermore, in step S5, the classification result corresponds to the stress state or active behavior state of the actively vocalizing fish; the stress state includes at least one of capture, cold, heat, and hypoxia; the active behavior state includes at least one of feeding and courtship.

[0014] Compared with the prior art, the present invention has at least the following beneficial effects: (1) The present invention proposes a flexible monitoring sensor and signal analysis method for actively vocalizing fish. By using a flexible piezoresistive sensor with a micro pyramid array structure and fixing it close to the vocalizing part of the fish, it achieves high-fidelity and source-based capture of the active acoustic vibration signal of the target individual. This not only overcomes the defect of traditional hydrophones that cannot distinguish between group and individual sound sources, but also ensures that weak bioacoustic signals directly related to behaviors such as feeding, courtship, and stress can be clearly acquired. (2) The flexible attachment measurement method and waterproof encapsulation design get rid of the dependence on the light transmittance of the water body, and can achieve long-term, stable and continuous signal acquisition under the actual aquaculture conditions of turbidity and variable light. (3) By generating a time-frequency-sound pressure level multi-channel feature image from the original resistance signal through adaptive baseline calibration and short-time Fourier transform, and inputting it into a lightweight convolutional neural network with significantly reduced parameters for real-time analysis, the state recognition capability is brought forward to the edge equipment of the breeding site, realizing the rapid identification and early warning of various stress states such as capture, cold, heat, and hypoxia, providing timely and reliable technical support for real-time management decisions such as precise feeding and stress intervention. Attached Figure Description

[0015] Figure 1 This is a schematic diagram showing the structure of a flexible monitoring sensor for actively vocalizing fish. Figure 2 This is a schematic diagram of the structure of a flexible monitoring sensor for actively vocalizing fish. Figure 3This is a step diagram of a signal analysis method for a flexible monitoring sensor used in actively vocalizing fish. Figure 4 This is a schematic diagram of the local deformation of the piezoresistive sensitive layer under pressure. Figure reference numerals: 1-flexible substrate, 2-interdigitated electrode, 3-piezoresistive sensitive layer, 31-polydimethylsiloxane, 32-graphene, 33-micropyramid structure. Detailed Implementation

[0016] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings to further illustrate the technical solutions of the present invention. However, the present invention is not limited to these embodiments.

[0017] As the global aquaculture industry moves towards intensification and intelligentization, real-time, accurate, and non-invasive monitoring of the physiological state of farmed organisms has become crucial for ensuring fish health and improving farming efficiency and product quality. Against this backdrop, biosignal-based behavioral and health monitoring technologies have attracted significant attention. This is particularly true for fish like the large yellow croaker, which possess the ability to actively vocalize. Their vocalization behavior is directly related to various physiological states such as feeding, courtship, and stress response, providing a unique biosignal source for health monitoring through acoustic means. However, existing monitoring technologies have significant limitations when addressing this specific need. For example, while wearable flexible sensors based on impedance or strain can achieve continuous monitoring, the signals they collect are mostly indirect physiological parameter changes caused by stress, which have a weak correlation with active vocalization behavior and are difficult to reflect the instantaneous behavioral intentions of individuals. Specialized underwater acoustic sensors (such as hydrophones) can collect acoustic signals, but their design is geared towards environmental or group sound field monitoring. Under intensive farming conditions, they cannot effectively separate and locate the vocalizations of specific individuals acoustically, lacking individual discrimination capabilities. Furthermore, optical monitoring technologies, such as photoplethysmography (PPG), are highly dependent on the light transmittance of the water, resulting in insufficient reliability in actual turbid and variable aquaculture environments. Meanwhile, some multifunctional integrated sensing systems are often complex in structure, computationally burdensome, and rely on cloud processing, making it difficult to achieve low-latency real-time analysis and decision-making at the aquaculture site. To address these issues, this invention proposes a flexible monitoring sensor for actively vocalizing fish. The core of this sensor lies in a specially designed flexible piezoresistive sensing structure that can closely adhere to the surface of the fish's vocalization area (such as near the gills of the large yellow croaker), directly capturing the extremely weak surface vibrations caused by the fish's active vocalization and converting them into electrical signals.

[0018] The flexible sensor proposed in this invention, such as Figure 1 and Figure 2As shown, it is essentially a multilayer flexible piezoresistive device. Its bottom layer is a flexible substrate, which needs to possess good flexibility, biocompatibility, and chemical stability to adapt to the underwater environment and the curvature changes of the fish's surface. In a preferred embodiment, a polyethylene terephthalate (PET) film with a thickness of approximately 0.05 mm is used as the flexible substrate 1. The PET film not only has moderate flexibility, allowing it to adhere well to the fish's surface, but also possesses excellent insulation and water resistance, providing reliable mechanical support and electrical isolation for the sensor.

[0019] On the upper surface of the flexible substrate, a pair of interdigitated electrodes 2 were fabricated using thin-film deposition processes such as thermal evaporation and magnetron sputtering, combined with photolithography or masking techniques. The design of these interdigitated electrodes aims to increase the contact area between the electrodes and the upper sensitive material, thereby more effectively collecting electrical signals. The electrode material is preferably a chromium / gold (Cr / Au) bilayer structure, where the chromium layer acts as an adhesion layer to enhance the adhesion between the gold electrode and the PET substrate, while the gold layer serves as the main conductive layer due to its excellent conductivity and corrosion resistance. The electrode width was optimized, set to 50 micrometers in one specific embodiment. This is because an excessively wide electrode would inhibit the deformation of the upper sensitive material due to its rigidity, leading to uneven stress distribution and thus degrading the piezoresistive performance of the sensor; while an excessively narrow electrode might increase fabrication difficulty and resistance. Therefore, the 50-micrometer-wide interdigitated electrodes achieve a good balance between strain transfer and electrical performance while ensuring manufacturability.

[0020] The key functional layer of the sensor is the piezoresistive sensing layer 3, which covers the interdigitated electrodes. This piezoresistive sensing layer 3 is made of a composite material of polydimethylsiloxane 31 (PDMS) and graphene 32 (Gr). PDMS is an elastomer material widely used in flexible electronics, possessing excellent flexibility and stretchability. Graphene is known for its excellent conductivity and mechanical strength. The combination of the two aims to give the piezoresistive sensing layer 3 both the flexibility of PDMS and the piezoresistive properties of graphene. During preparation, a certain proportion of graphene nanosheets are uniformly dispersed in uncured PDMS prepolymer, and the mixture is ensured to be uniform by stirring and ultrasonication, and then cast onto a substrate with a specific pattern template. In a preferred embodiment of the invention, the template surface has a regularly arranged array of micro-pyramid structure pits. Therefore, after the PDMS-Gr composite material cures and is demolded, its surface forms an array of micro-pyramid structures corresponding to the template. The size of each micro-pyramid structure 33 is on the order of micrometers, and its tip has an extremely small radius of curvature. Figure 4As shown, when external sound pressure or vibration is applied to the sensor surface, stress is highly concentrated at the tip of the micro-pyramid structure 33, causing significant local deformation in this region. This deformation drastically alters the contact state, spacing, and tunneling effect between the graphene sheets, resulting in a significant change in the overall resistance of the composite material, from the resistance R under relatively low pressure conditions. L Resistance R under relatively high pressure H Compared to flat surfaces, the micro-pyramid structure 33 can amplify weak mechanical stimuli into detectable changes in electrical signals, which is crucial for capturing low-amplitude signals such as fish vocalizations.

[0021] To ensure the sensor can operate stably underwater for extended periods, it requires waterproof encapsulation. In a preferred embodiment, a medical-grade polyurethane (PU) film is used as the waterproof encapsulation layer. During encapsulation, the assembled sensor core (including a flexible substrate, interdigitated electrodes, and a piezoresistive sensitive layer) and the micro-signal processing module electrically connected to the electrodes (which may include preamplifier, filter, and other circuitry) are encapsulated together with the PU film, leaving only the necessary electrical interfaces or wireless antenna area exposed. The encapsulated sensor is then fixed to a specific location on the target fish's body surface using biocompatible adhesive or medical dressing, such as near the lateral line region behind the gill covers of a large yellow croaker, where acoustic vibrations are most readily transmitted to the body surface. It is important to ensure that the sensor is attached in close contact with the fish's skin to maximize the coupling of vibration signals.

[0022] After introducing the structural components of the flexible monitoring sensor, its working principle will be explained. When fish with the flexible monitoring sensor attached are stimulated by external factors (such as human capture, sudden changes in water temperature, or a decrease in dissolved oxygen) or engage in proactive behaviors (such as feeding or courtship), their sound-producing organs (such as the spiral tympanic muscle of the large yellow croaker) contract and vibrate, driving the swim bladder to produce sound. The generated sound waves propagate through the fish's body tissue and the surrounding water. The vibration component directly transmitted to the fish's skin acts on the attached flexible monitoring sensor. The micro-pyramid array on the surface of the flexible monitoring sensor undergoes periodic deformation under the influence of sound pressure, causing a synchronous change in the resistance of the PDMS-Gr composite material. This resistance change is detected in real time by the interdigitated electrodes and transmitted to the signal processing module via leads. The signal processing module is typically based on a microcontroller (such as the STM32 series) and integrates an analog-to-digital converter (ADC), a storage unit, and a wireless communication module (such as Wi-Fi or Bluetooth). The ADC converts the analog resistance change signal into a digital signal and stores it, while the wireless module is used to send the data to a host computer or edge computing device. By monitoring the relative rate of change of resistance (ΔR / R0, where ΔR is the change in resistance and R0 is the static resistance), the acoustic vibration waveform of the fish's surface can be derived by analyzing the resistance change signal, thereby achieving direct capture of its acoustic signal.

[0023] In order to extract the active / passive physiological state information of fish from the resistance change signal, such as Figure 3 As shown, this invention proposes a complete signal analysis method, which mainly includes the following steps: S1: Acquire the resistance change signal after conversion by the flexible monitoring sensor; S2: Perform adaptive baseline estimation on the resistance change signal using a median filter, and subtract the baseline estimate from the resistance change signal; S3: Perform a short-time Fourier transform on the baseline-corrected signal to generate the corresponding time-spectrum graph; S4: Calculate the sound pressure level based on the amplitude value of the time-frequency spectrum and synthesize a multi-channel feature image containing time, frequency and sound pressure level information; S5: Input the multi-channel feature image into a pre-trained lightweight convolutional neural network model to obtain the classification results of the physiological state corresponding to the fish's active vocalization behavior.

[0024] Specifically, the first step is signal preprocessing, primarily aimed at eliminating baseline drift. Because the sensor possesses static resistance, which may slowly change due to factors such as temperature and stress relaxation during long-term underwater monitoring, the signal baseline may shift. Therefore, this invention employs an adaptive baseline estimation method based on median filtering. Let the acquired one-dimensional resistance time-domain signal sequence be... First, based on the total length of the data Dynamically determine the window size of the median filter ,generally Set to an odd number no greater than 201, for example, it can be taken as... Then, regarding Window length is Median filtering yields the baseline estimated sequence. Finally, the baseline estimate is subtracted from the original signal to obtain the baseline-corrected signal. This method can remove low-frequency drift while preserving the high-frequency acoustic characteristics of the signal.

[0025] Preprocessed signal This is a one-dimensional time-domain sequence. To reveal the frequency characteristics that change over time, time-frequency analysis is required. This invention uses Short-Time Fourier Transform (STFT) to convert the one-dimensional signal into a two-dimensional time-frequency spectrum. STFT is achieved by performing Fourier transforms on windowed signal segments. Specific parameter settings need to balance time resolution and frequency resolution. In a preferred embodiment, a Hamming window is used, and the length of the Fast Fourier Transform (FFT) is adaptively selected between 64 and 512 points based on the signal characteristics, with the window overlap rate set to 75%. After performing STFT, a complex matrix is ​​obtained. ,in Represents a time frame. Representing frequency points. To more intuitively reflect the energy intensity of sound, this invention uses amplitude values. Convert to sound pressure level (SPL, unit: decibels dB). The conversion formula is: , In the formula, It is an extremely small constant (e.g., 1×10⁻⁶). -10 ), used to prevent taking the logarithm of zero; This is an offset coefficient used to adjust the calculated relative sound pressure level to a suitable numerical range for easier subsequent processing and visualization. Ultimately, a time-frequency graph containing three dimensions—time, frequency, and sound pressure level—is obtained. To facilitate processing by convolutional neural networks, this information from these three dimensions is further synthesized into a multi-channel feature image. For example, a three-channel "image" can be generated, where the first channel represents time (or a normalized frame number), the second channel represents frequency, and the third channel represents the corresponding sound pressure level value. This representation method transforms a one-dimensional time-series signal into a two-dimensional image with spatial structure, effectively preserving the spectral pattern differences of vocal signals under different physiological states.

[0026] After the feature images are generated, they are fed into a lightweight convolutional neural network (CNN) model for physiological state classification. This CNN model, named the FishSound classifier, is specifically designed for edge deployment. Its design goal is to greatly compress the number of model parameters and computational cost while ensuring high classification accuracy, enabling it to run on embedded devices or edge computing boxes in the farm and achieve real-time inference. The total number of parameters in the model has been optimized to only about one-eighth of that of traditional large CNNs such as VGG. This is mainly achieved through the following strategies: using a shallower network depth to avoid excessively deep layers; using a large number of small-sized convolutional kernels (such as 3×3); introducing pooling layers after appropriate layers to reduce the dimension of feature maps; and significantly simplifying fully connected layers. The input to the model is the aforementioned three-channel time-frequency-SPL feature image. The input image first passes through several convolutional and pooling layers for feature extraction. These layers can automatically learn abstract representations of spectral features under different stress states, such as energy patches in specific frequency bands and textures of time-frequency patterns. The extracted high-level features are then flattened and fed into a fully connected layer, and finally output by a Softmax classifier the probability distribution of each physiological state category.

[0027] In the specific application of this invention, the physiological states of fish are defined into seven categories, including four typical stress states: capture stress, cold stress, heat stress, and hypoxia, as well as three different recovery stages after being subjected to a knocking stimulus: 1 minute, 5 minutes, and 10 minutes after the knocking. Of course, depending on actual monitoring needs, the classification categories can be extended to include active behavioral states such as feeding and courtship. To train the FishSound CNN model, this invention collected a large amount of vocalization data of large yellow croaker under the above different states and generated corresponding feature images to form a training dataset. The dataset was divided into a training set, a validation set, and a test set. The model uses the cross-entropy loss function and is trained using the Adam optimizer. Because the model is very lightweight, its single inference time is extremely short, achieving millisecond-level response times on common edge computing devices (such as Raspberry Pi, Jetson Nano, etc.), thus fully meeting the needs of real-time early warning in aquaculture sites.

[0028] In summary, the flexible monitoring sensor and signal analysis method proposed in this invention for actively vocalizing fish achieves high-fidelity, source-based capture of the active acoustic vibration signals of the target individual by employing a flexible piezoresistive sensor with a micro-pyramid array structure and fixing it close to the vocalization part of the fish. This not only overcomes the shortcomings of traditional hydrophones in distinguishing between group and individual sound sources, but also ensures that weak bioacoustic signals directly related to behaviors such as feeding, courtship, and stress are clearly acquired.

[0029] The flexible attachment measurement method and waterproof encapsulation design eliminate the dependence on water transparency, enabling long-term, stable, and continuous signal acquisition under actual aquaculture conditions of turbidity and variable lighting. By generating a time-frequency-sound pressure level multi-channel feature image from the raw resistance signal through adaptive baseline calibration and short-time Fourier transform, and inputting it into a lightweight convolutional neural network with significantly reduced parameters for real-time analysis, the system brings state recognition capabilities to the edge devices in the aquaculture field. This enables rapid identification and early warning of various stress states such as capture, cold, heat, and hypoxia, providing immediate and reliable technical support for real-time management decisions such as precise feeding and stress intervention.

[0030] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0031] Furthermore, in this invention, descriptions involving terms such as "first," "second," and "a" are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0032] In this invention, unless otherwise explicitly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0033] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

Claims

1. A flexible monitoring sensor for actively acoustic fish, characterized in that, include: Flexible substrate (1); Interdigitated electrodes (2) disposed on the flexible substrate; And a piezoresistive sensitive layer (3) covering the interdigitated electrodes; The piezoresistive sensitive layer (3) is formed by a composite material of polydimethylsiloxane (31) and graphene (32), and the surface of the piezoresistive sensitive layer (3) has a regularly arranged array of micro pyramid structures (33). The flexible monitoring sensor is attached to the surface of the vocal part of the active vocal fish. The micro pyramid structure (33) array on the surface of the piezoresistive sensitive layer (3) deforms in response to the acoustic vibration within a preset frequency range on the surface of the vocal part. Through the piezoresistive effect of the piezoresistive sensitive layer (3), the acoustic vibration is converted into a resistance change signal between the interdigital electrodes (2).

2. The flexible monitoring sensor for actively acoustic fish as described in claim 1, characterized in that, The flexible substrate (1) is a polyethylene terephthalate film.

3. A flexible monitoring sensor for actively acoustic fish as described in claim 1, characterized in that, The interdigitated electrode (2) is a chromium / gold electrode.

4. A flexible monitoring sensor for actively acoustic fish as described in claim 1, characterized in that, The preset frequency range is the vibration frequency band of the body surface when the vocalization part of an actively vocalizing fish emits sound.

5. A flexible monitoring sensor for actively acoustic fish as described in claim 1, characterized in that, It also includes a signal processing module, which is electrically connected to the interdigitated electrode (2) and is used to collect the resistance change signal.

6. A flexible monitoring sensor for actively acoustic fish as described in claim 5, characterized in that, It also includes a waterproof encapsulation layer, which covers the flexible substrate (1), interdigitated electrodes (2), piezoresistive sensitive layer (3), and signal processing module.

7. A signal analysis method applied to a flexible monitoring sensor for actively vocalizing fish as described in any one of claims 1 to 6, characterized in that, Including the following steps: S1: Acquire the resistance change signal after conversion by the flexible monitoring sensor; S2: Perform adaptive baseline estimation on the resistance change signal using a median filter, and subtract the baseline estimate from the resistance change signal; S3: Perform a short-time Fourier transform on the baseline-corrected signal to generate the corresponding time-spectrum graph; S4: Calculate the sound pressure level based on the amplitude value of the time-frequency spectrum and synthesize a multi-channel feature image containing time, frequency and sound pressure level information; S5: Input the multi-channel feature image into a pre-trained lightweight convolutional neural network model to obtain the classification results of the physiological state corresponding to the fish's active vocalization behavior.

8. The signal analysis method for a flexible monitoring sensor for actively vocalizing fish as described in claim 7, characterized in that, In step S2, the window size of the median filter is determined based on the data length of the resistance change signal, and the window size is set to an odd number not greater than 201.

9. The signal analysis method for a flexible monitoring sensor for actively vocalizing fish as described in claim 7, characterized in that, In step S4, the sound pressure level is calculated using the following formula: In the formula, Sound pressure level, This represents the amplitude value of the short-time Fourier transform. To prevent small constants from taking the logarithm of zero, This is the offset coefficient.

10. The signal analysis method for a flexible monitoring sensor for actively vocalizing fish as described in claim 7, characterized in that, In step S5, the classification result corresponds to the stress state or active behavior state of the actively vocalizing fish; the stress state includes at least one of capture, cold, heat, and hypoxia; the active behavior state includes at least one of feeding and courtship.

Images