Pseudosciaena crocea feeding intensity prediction method and system based on acoustic detection

Abstract

The invention discloses a large yellow croaker ingestion intensity prediction method and system based on acoustic detection, and is applied to the field of fishery breeding, and the method comprises the steps: constructing a large yellow croaker ingestion sound wave change curve based on a time sequence sound wave signal of a target breeding region, and carrying out the analysis, and obtaining a first feature separation curve of different ingestion behaviors; correcting and re-separating each first feature separation curve to obtain a second feature separation curve; performing ingestion feature extraction on the second feature separation curves, and calculating ingestion intensity indexes of the second feature separation curves according to ingestion features; performing statistical analysis on the ingestion intensity index to obtain an ingestion intensity average value and an ingestion intensity variable coefficient of the large yellow croaker; and according to the feeding intensity average value and the feeding intensity variation coefficient, generating a feeding intensity prediction report of the large yellow croaker. The large yellow croaker feeding intensity is quantified in real time through acoustic non-intrusive monitoring, the bait utilization rate is increased, manual disturbance is reduced, and precise feeding and growth synchronization is achieved.

Description

Technical Field

[0001] This invention relates to the field of aquaculture technology, and in particular to a method and system for predicting the feeding intensity of large yellow croaker based on acoustic detection. Background Technology

[0002] With the rapid development of aquaculture, the large yellow croaker, as an important economic fish, has seen its farming scale continuously expand. Accurately grasping the feeding status of large yellow croaker is crucial for improving farming efficiency and ensuring fish health.

[0003] Existing methods for detecting the feeding behavior of large yellow croaker mostly rely on manual observation or simple sensor data, which suffers from problems such as low efficiency, inaccurate data, and inability to monitor in real time. Manual observation is not only time-consuming and labor-intensive, but also makes it difficult to obtain continuous feeding data. At the same time, the acoustic signals in the aquaculture environment are complex and diverse, including not only the sound waves produced by the large yellow croaker feeding, but also noise generated by water flow, equipment operation, and other biological activities. This noise interference makes it difficult to accurately interpret the acoustic signals, resulting in inaccurate identification of feeding behavior and inaccurate quantification of feeding intensity. Consequently, methods for predicting feeding behavior based on acoustic signals are also unable to distinguish feeding behaviors.

[0004] Therefore, how to accurately identify and analyze the feeding behavior of large yellow croaker through acoustic detection in order to achieve accurate prediction of feeding intensity has become a technical problem that urgently needs to be solved by technical personnel in this field. Summary of the Invention

[0005] This invention provides a method and system for predicting the feeding intensity of large yellow croaker based on acoustic detection, in order to predict the feeding intensity of large yellow croaker.

[0006] To address the aforementioned technical problems, embodiments of the present invention provide a method for predicting the feeding intensity of large yellow croaker based on acoustic detection, comprising: The feeding sound wave variation curve of large yellow croaker was constructed based on the time-series acoustic wave signals of the target aquaculture area.

[0007] The peak sound pressure level and frequency distribution characteristics in the feeding sound wave variation curve of the large yellow croaker were analyzed to obtain the first feature separation curve for different feeding behaviors.

[0008] The consistency of sound pressure level and frequency range of each first feature separation curve is verified by the changes in sound pressure level and frequency distribution of adjacent first feature separation curves. Based on the verification results, each first feature separation curve is corrected and reseparated to obtain the second feature separation curve.

[0009] Feeding features are extracted from each of the corrected second feature separation curves, and the feeding intensity index of each second feature separation curve is calculated based on the feeding features.

[0010] Statistical analysis was performed on the feeding intensity index to obtain the average feeding intensity and the coefficient of variation of feeding intensity for large yellow croaker.

[0011] A feeding intensity prediction report for large yellow croaker is generated based on the average feeding intensity and the coefficient of variation of feeding intensity.

[0012] Furthermore, the construction of the feeding acoustic variation curve of large yellow croaker based on the time-series acoustic signal of the target aquaculture area includes: The acquired time-series acoustic signal is subjected to multi-scale wavelet transform to decompose the signal into first multi-scale characteristic signals of different frequency bands.

[0013] An adaptive filtering algorithm is applied to dynamically adjust the filtering parameters of the multi-scale feature signal to obtain a second multi-scale feature signal.

[0014] The second multi-scale feature signal was subjected to time-frequency analysis using Fourier transform and Hilbert transform to obtain an enhanced feature signal. Based on the enhanced feature signal, a feeding sound wave variation curve of the large yellow croaker was constructed.

[0015] Furthermore, the analysis of the sound pressure level peak value and frequency distribution characteristics in the feeding sound wave variation curve of the large yellow croaker yields the first feature separation curve for different feeding behaviors, including: Time-domain analysis was performed on the feeding sound wave variation curve of the large yellow croaker to obtain the peak sound pressure level and corresponding timestamp in the sound wave signal.

[0016] The frequency distribution of the acoustic signal in the feeding acoustic waveform of the large yellow croaker was calculated to obtain the frequency distribution characteristics.

[0017] Based on the results of time-domain analysis and the frequency distribution characteristics, the peak sound pressure level is matched with the corresponding frequency components to form sound pressure level-frequency feature points.

[0018] Based on the time series distribution of the sound pressure level-frequency feature points, different feeding behavior patterns are identified, and each feeding behavior pattern corresponds to a sequence of sound pressure level-frequency feature points, forming a first feature separation curve.

[0019] Further, the consistency of sound pressure level and frequency range of each of the first feature separation curves is verified by the changes in sound pressure level and frequency distribution of adjacent first feature separation curves. Based on the verification results, each of the first feature separation curves is corrected and re-separated to obtain the second feature separation curve, including: The sound pressure level variation of adjacent first feature separation curves is analyzed, and the sound pressure level difference between adjacent curves is calculated.

[0020] Frequency distribution variation analysis is performed on adjacent first feature separation curves to calculate the frequency range overlap between adjacent curves.

[0021] Based on the first difference between the sound pressure level difference and the preset sound pressure level threshold, and the second difference between the frequency range overlap and the preset frequency overlap threshold, the sound pressure level consistency and frequency range consistency of adjacent first feature separation curves are determined.

[0022] Based on the judgment results, each of the first feature separation curves is corrected, and the corrected first feature separation curves are re-identified and separated to obtain the second feature separation curve.

[0023] Furthermore, the step of correcting each of the first feature separation curves based on the judgment result includes: When the first difference and / or the second difference exceed the correction threshold, the sound pressure level of the first feature separation curve is dynamically adjusted point by point based on recursive dependence. The adjusted first feature separation curve is nonlinearly smoothed to eliminate abrupt changes introduced by the correction and to enhance the features.

[0024] The enhanced first feature separation curve is re-identified and separated to obtain the second feature separation curve.

[0025] Further, the process of re-identifying and separating the enhanced first feature separation curve to obtain the second feature separation curve includes: Obtain the key feature points of the enhanced first feature separation curve.

[0026] Obtain feeding behavior data of large yellow croaker from a farmed fish database.

[0027] Using the key feature points as anchor points, cluster analysis is performed in conjunction with the feeding behavior feature data of the large yellow croaker. Based on the analysis results, the first feature separation curve is separated into multiple sub-segments, and each sub-segment corresponds to a potential feeding behavior event.

[0028] Calculate the feeding behavior characteristics of each sub-segment, classify each sub-segment according to the feeding behavior characteristics, and merge or separate each sub-segment according to the classification results to obtain the second feature separation curve.

[0029] Further, the step of extracting feeding features from each of the corrected second feature separation curves and calculating the feeding intensity index of each second feature separation curve based on the feeding features includes: Feature extraction is performed on each of the corrected second feature separation curves to obtain acoustic feature parameters characterizing the feeding behavior of large yellow croaker. The acoustic feature parameters include sound pressure level features, frequency features, and time features of the feeding behavior of large yellow croaker.

[0030] Based on the acoustic characteristic parameters, the feeding intensity index of each of the second characteristic separation curves is calculated.

[0031] Furthermore, the statistical analysis of the feeding intensity index to obtain the average feeding intensity and the coefficient of variation of feeding intensity for large yellow croaker includes: The feeding intensity indices are summarized, and the sum of the feeding intensity indices of all the second characteristic separation curves is calculated. Based on the sum and the number of the second characteristic separation curves, the average feeding intensity of the large yellow croaker is calculated.

[0032] Calculate the difference between the feeding intensity index of each of the second feature separation curves and the average feeding intensity, and square the difference to obtain the sum of squares of the differences. Based on the sum of squares of the differences and the number of the second feature separation curves, calculate the coefficient of variation of the feeding intensity of the large yellow croaker.

[0033] Further, generating a feeding intensity prediction report for large yellow croaker based on the average feeding intensity and the coefficient of variation of feeding intensity includes: The current feeding intensity level of the large yellow croaker is obtained by performing quantile normalization analysis on the average feeding intensity.

[0034] The coefficient of variation of the feeding intensity was analyzed using the entropy weight method to obtain the stability index of the feeding intensity of large yellow croaker.

[0035] Based on the changing trends of the average feeding intensity and the coefficient of variation of the feeding intensity, grey relational analysis is performed to calculate the fluctuation of the future feeding intensity of the large yellow croaker.

[0036] A feeding intensity prediction report for large yellow croaker is generated based on the feeding intensity level, the stability index, and the fluctuation.

[0037] This invention also provides a system for predicting the feeding intensity of large yellow croaker based on acoustic detection, comprising: The curve construction module is used to construct the feeding sound wave variation curve of large yellow croaker based on the time-series acoustic wave signal of the target aquaculture area.

[0038] The feature analysis module is used to analyze the peak sound pressure level and frequency distribution characteristics in the feeding sound wave change curve of the large yellow croaker to obtain the first feature separation curve for different feeding behaviors.

[0039] The curve separation module is used to verify the consistency of sound pressure level and frequency range of each first feature separation curve by the changes in sound pressure level and frequency distribution of adjacent first feature separation curves, and to correct and re-separate each first feature separation curve according to the verification results to obtain the second feature separation curve.

[0040] The index calculation module is used to extract feeding features from each of the corrected second feature separation curves and calculate the feeding intensity index of each of the second feature separation curves based on the feeding features.

[0041] The coefficient calculation module is used to perform statistical analysis on the feeding intensity index to obtain the average feeding intensity and the coefficient of variation of feeding intensity of large yellow croaker.

[0042] The intensity prediction module is used to generate a feeding intensity prediction report for large yellow croaker based on the average feeding intensity and the coefficient of variation of the feeding intensity.

[0043] Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following: By processing time-series acoustic signals using multi-scale wavelet transform and adaptive filtering algorithms, characteristic signals of large yellow croaker feeding behavior are effectively separated. Further integration with time-frequency analysis techniques enables precise extraction of sound pressure level and frequency distribution characteristics, thus accurately identifying different feeding behavior patterns. Dynamic correction and re-separation of feature curves ensure the accuracy and stability of feature extraction. Based on these accurate feature parameters, the calculated feeding intensity index more closely reflects actual feeding conditions. Furthermore, statistical analysis and trend prediction methods generate predictive reports containing average feeding intensity, coefficient of variation, and future fluctuations, providing a scientific basis for aquaculture management and enhancing the accuracy and practicality of feeding intensity prediction. This achieves real-time, precise monitoring of large yellow croaker feeding intensity, reducing the labor intensity and time cost of manual observation. Timely and accurate feeding intensity prediction allows aquaculture personnel to optimize feed delivery strategies, avoid feed waste, and reduce aquaculture costs. Simultaneously, precise feeding monitoring helps to promptly detect potential health problems, ensuring the healthy growth of large yellow croaker and thus improving the overall efficiency and economic benefits of aquaculture. Attached Figure Description

[0044] Figure 1 This is a flowchart illustrating the steps of a method for predicting the feeding intensity of large yellow croaker based on acoustic detection in one embodiment of the present invention. Figure 2 This is a comparison diagram of feeding patterns of the large yellow croaker feeding intensity prediction method based on acoustic detection in one embodiment of the present invention; Figure 3This is a structural block diagram of a large yellow croaker feeding intensity prediction system based on acoustic detection in one embodiment of the present invention. Detailed Implementation

[0045] The technical solutions of 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. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0046] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0047] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. The terms "vertical," "horizontal," "left," "right," "upper," "lower," and similar expressions used herein are for illustrative purposes only and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0048] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0049] One embodiment of the present invention provides a method for predicting the feeding intensity of large yellow croaker based on acoustic detection. For details, please refer to [link to documentation]. Figure 1 , Figure 1 The flowchart shown is a step-by-step flowchart of a method for predicting the feeding intensity of large yellow croaker based on acoustic detection in one embodiment of the present invention, including steps S11-S16: S11. Construct the feeding sound wave variation curve of large yellow croaker based on the time-series acoustic wave signal of the target aquaculture area.

[0050] In deep-sea cages, the feeding rhythm of large yellow croaker directly determines whether feed is efficiently absorbed and whether fish in the same batch can grow synchronously. Once the rhythm is unbalanced, uneaten feed and stress will lead to water quality deterioration and individual differences. However, underwater light is weak and the window for artificial observation is short. Traditional methods of relying on the naked eye or weighing with nets are time-consuming and easily disturb the fish, making it difficult to obtain continuous and undisturbed feeding information. Therefore, this embodiment turns its attention to underwater sound waves: by deploying hydrophones in the target aquaculture area, the pulses and energy clusters generated by the fish feeding are transformed into continuous temporal sound wave signals. This allows for the construction of a "large yellow croaker feeding sound wave change curve" throughout the entire feeding process without disturbing the fish, providing a reliable basis for subsequent precise feeding and health management.

[0051] Before formally collecting acoustic signals, this embodiment first delineates the "target aquaculture area" based on the biological characteristics of large yellow croaker and aquaculture engineering experience. This area is located at a water depth of 4-8m in the center of the net cage, avoiding the main jet area of ​​the aerator and the reflection area of ​​the net wall, so as to reduce mechanical noise and boundary reflection interference. The size of the area should be 20m×20m to ensure that the feeding activities of the fish are concentrated and the sound wave propagation path can be approximated as a free field.

[0052] Specifically, in this embodiment, four wideband hydrophones (bandwidth 10Hz–60kHz, sensitivity -205dB re 1 V / μPa) are arranged in a diamond pattern in the target area with a spacing of 5m between the hydrophones. The sampling rate is uniformly 192kHz with 24-bit quantization to ensure that the 1–10kHz high-frequency short pulses generated by the large yellow croaker feeding can be completely captured.

[0053] Continuous monitoring for 24 hours, with fixed feeding three times a day (07:00, 12:00, 17:30), with the key period being 30 minutes before and 60 minutes after each feeding; this was carried out continuously for 7 days to obtain the raw dataset of "time-series acoustic signals". During the acquisition process, a GPS clock was used to synchronize the four hydrophones at the sub-millisecond level to ensure phase consistency during subsequent multi-channel fusion.

[0054] The feeding sound wave change curve is a continuous curve with time as the horizontal axis and acoustic characteristic values ​​representing feeding intensity as the vertical axis. In order to suppress background noise and highlight feeding events, it is necessary to complete the "denoising-decomposition-enhancement-reconstruction" process on the basis of the original sound wave signal. Finally, the discrete sound events are converted into a quantifiable and comparable curve shape for subsequent feeding behavior identification and intensity assessment.

[0055] The acquired raw acoustic signal was first subjected to multi-scale wavelet transform, decomposing it into 32 layers of wavelet coefficients at different frequencies, covering 10Hz–60kHz, to initially separate the feeding pulses from environmental noise, obtaining the "first multi-scale characteristic signal". Since the ambient noise varied drastically with feeding and external conditions, an adaptive filtering algorithm was then used to denoise the high-frequency feeding signal in real time, using low-frequency noise as a reference. The filter order was 128, and the step size was adaptively adjusted, outputting the "second multi-scale characteristic signal", which improved the signal-to-noise ratio by approximately 10 dB.

[0056] Based on this, a short-time Fourier transform was performed on the "second multi-scale feature signal" to lock the dominant frequency of 1–10 kHz. Then, a Hilbert transform was performed to extract the instantaneous envelope, and after smoothing for 11 ms, the "enhanced feature signal" was obtained. The enhanced feature signal was resampled on a second-by-second basis, the peak cumulative energy per second was calculated, and the spatial average was taken from the four hydrophones to finally generate a "large yellow croaker feeding sound wave change curve" with a time resolution of 1 second, which can be directly used for subsequent feeding behavior identification and intensity assessment.

[0057] When installing the hydrophone, a float-counterweight flexible suspension is used to ensure that the probe is always aligned with the water layer where the fish are active, while allowing it to sway slightly in the waves to avoid cable fatigue and breakage; the cable is covered with a wear-resistant flexible tube and fixed in sections to prevent additional noise from friction with the net cage.

[0058] Through the above process, the underwater sound field of the target aquaculture area is completely mapped into a continuous feeding sound wave variation curve. This curve not only records the entire process of the fish from "resting—finding food—concentrated feeding—gradually becoming satiated," but also eliminates false fluctuations caused by wind, waves, and mechanical noise. When the curve shows a gentle, slightly rising trend, it indicates that the fish's feeding rhythm is uniform, the feed utilization rate is the highest, and the amount of uneaten feed can be reduced to the lowest level. When the curve shows a sharp peak, it indicates that the peak feeding period has arrived, and farmers can appropriately increase feeding within a few minutes after the peak appears to avoid subsequent hunger. In addition, the temporal continuity of the curve makes cross-day and cross-batch comparisons possible. The differences in feeding in the same cage under different weather conditions or feed formulas are immediately apparent, providing an objective basis for long-term optimization of the feeding system and significantly reducing the frequency and labor intensity of manual inspections.

[0059] S12. Analyze the peak sound pressure level and frequency distribution characteristics in the feeding sound wave change curve of the large yellow croaker to obtain the first feature separation curve of different feeding behaviors.

[0060] like Figure 2As shown, in actual aquaculture scenarios, the feeding actions of large yellow croaker generate transient high-frequency pulses underwater. The amplitude, timing, and frequency components of these pulses together constitute the acoustic fingerprint that distinguishes three typical behaviors: "grabbing feeding," "steady feeding," and "swallowing." During grabbing feeding, the fish lunges rapidly, and the mouth flaps, generating sharp and short high-frequency spikes. During steady feeding, rhythmic pecking movements form pulse clusters with uniform energy and stable intervals in the mid-frequency range. During the swallowing stage, the fish opens its mouth wide and closes it for a long time, resulting in a continuous low-frequency broadband sound segment. Because the net cages simultaneously contain the noise of aerators, wave impacts, and boat echoes, these acoustic fingerprints are often submerged. Therefore, it is necessary to first extract quantifiable peak values ​​and spectral information from the entire feeding sound wave variation curve, and then, through peak-time alignment and frequency-energy correlation, merge the scattered pulses into corresponding behavioral segments to reconstruct the true feeding rhythm. Specifically, the feeding behavior of large yellow croaker schools is shown in Table 1. Table 1

[0061] Therefore, this embodiment performs time-domain analysis on the curve to capture each actual feeding action. Specifically, a sliding micro-observation window is set up on the time axis: the window width is slightly larger than the duration of a typical feeding pulse, and the step size is set to half the window size to ensure that any rapid consecutive attacks are not missed. Whenever the window slides to a new position, the algorithm first calculates the instantaneous sound pressure level within that window and compares it with the rolling average of the background noise in the same segment; if the sound pressure level exceeds the dynamic threshold and the rise slope meets the steep rise and fall pattern unique to "fish mouth-slapping," it is determined to be a valid peak. Subsequently, the algorithm not only records the peak amplitude but also locks its precise millisecond-level timestamp within the window and writes the "peak-amplitude-time" triplet into the sequence. To reduce interference from occasional mechanical noise, the algorithm further introduces short-time energy ratio verification: only when the energy proportion of the segment containing the peak is higher than a certain multiple of the adjacent silent segment is the peak finally confirmed. After such iterations, the entire curve is transformed into a continuous and clean "peak-time" sequence, which not only preserves the timing of the large yellow croaker's feeding actions, but also lays a reliable benchmark for subsequent alignment with spectral information.

[0062] Next, to fully decode the "sound fingerprint" behind each peak, this embodiment, after peak locking, first extends a segment slightly longer than a complete feeding pulse forward and backward as the analysis window, centered on the peak. This covers both the rising and falling edges of the pulse while avoiding frequency leakage caused by an excessively long window. Subsequently, a Fast Fourier Transform is performed on this segment to obtain a continuous energy spectrum from low to high frequencies. Based on this, the highest energy dominant frequency position in the spectrum is first located—it often corresponds to the main resonant frequency of the fish's mouth flapping or gill cover vibration. Then, the percentage of energy within a certain bandwidth on both sides of the dominant frequency is calculated to eliminate interference from broadband environmental noise. To further confirm that the peak is indeed caused by feeding rather than mechanical noise, the algorithm also compares the dominant frequency with the center frequency offset of three typical modes in the database: "fighting - high-frequency narrow band," "stable feeding - mid-frequency concentration," and "swallowing - low-frequency broad peak." Only when the offset falls within the allowable tolerance and the energy concentration is higher than the threshold is the dominant frequency and its energy percentage officially recorded as the "frequency distribution characteristic" corresponding to the peak. Using this feature, it becomes possible to clearly distinguish, in complex contexts, the sharp high-frequency energy clusters during food grabbing, the uniform mid-frequency clusters during steady feeding, and the low-frequency broadband energy bands continuously released during swallowing.

[0063] Based on the results of time-domain analysis and frequency distribution characteristics, the peak sound pressure level is matched with the corresponding frequency components to form sound pressure level-frequency feature points. The system pairs each peak amplitude with its dominant frequency position to generate feature points carrying three-dimensional information of "time-amplitude-frequency". These feature points are arranged in chronological order to form a discrete but continuously evolving feature point stream, which preserves both the temporal sequence of feeding actions and the differences in acoustic properties.

[0064] Further examining the entire sound pressure level-frequency characteristic point flow along the time axis reveals a clear alternation of various aggregation patterns: when the fish spot bait, the characteristic points suddenly become dense, the dominant frequency shifts to the high-frequency range, but the amplitude remains low, and the intervals between adjacent points are extremely short, forming a typical "feeding pulse cluster"; subsequently, the fish enter a stable pecking phase, the rhythm of the characteristic points becomes more uniform, the dominant frequency stabilizes in the mid-frequency range, the amplitude is moderate, and the time intervals exhibit a regular beat, constituting a "stable feeding sequence"; when the bait is swallowed, the characteristic points suddenly shift to the low-frequency range, the amplitude increases significantly, and the duration lengthens, forming a "swallowing duration." To accurately segment these behaviors, this embodiment uses the density of points within the time window, the drift of the dominant frequency center, and the amplitude gradient as criteria, setting an adaptive threshold: when the density suddenly increases and the dominant frequency is higher than the set upper limit, it is marked as a feeding segment; when the density is stable and the dominant frequency falls in the mid-frequency range, it is marked as a stable feeding segment; when the density decreases, the dominant frequency is lower than the lower limit, and the amplitude increases, it is marked as a swallowing segment. Based on these criteria, the feature point sequence is automatically divided into several continuous segments, each of which contains only a single behavioral pattern of "sound pressure level-frequency feature point" sequence, and finally spliced ​​together into multiple clearly distinguishable "first feature separation curves".

[0065] The first feature separation curve breaks down the complex sound field into three clear segments: "grabbing—steady feeding—swallowing," eliminating the need for fish farmers to rely on experience to judge the fish's condition. The appearance of short, high-frequency clusters in the curve corresponds to a feeding frenzy at the moment of feeding, at which point the feeding frequency can be increased simultaneously. The mid-frequency uniform segment represents a stable pecking period, where regular feeding can be maintained. The appearance of a sustained low-frequency segment indicates the start of the swallowing stage, and feed should be gradually reduced to prevent overfeeding. This segmented presentation also immediately highlights abnormal behaviors (such as sudden silence or continuous high noise), facilitating early detection of fish stress or equipment malfunctions and reducing economic losses. Furthermore, manual verification of the curve and on-site video shows that the accuracy of identifying the three segments is within an acceptable range.

[0066] S13. Verify the consistency of sound pressure level and frequency range of each first feature separation curve by measuring the changes in sound pressure level and frequency distribution of adjacent first feature separation curves. Based on the verification results, correct and re-separate each first feature separation curve to obtain the second feature separation curve.

[0067] In actual cage environments, water flow disturbances, aerator start-up and shutdown, uneven feed distribution, and even minute sensor drift can all cause abrupt changes in acoustic conditions between adjacent time periods. Consequently, the same feeding event may exhibit fluctuating sound pressure levels, frequency drift, or insufficient energy band overlap on the two first characteristic separation curves. Directly feeding these seemingly independent segments into subsequent intensity calculations will misjudge environmental noise as differences in feeding intensity, leading to distorted feeding strategies. To avoid misjudging "different sounds from the same fish" or "same sounds from different fish," this embodiment must first "align" adjacent curves before proceeding further—by verifying the consistency of sound pressure levels and frequency ranges, identifying and correcting deviations caused by the environment rather than the behavior itself. Only through this self-checking and calibration can the acoustic segments truly generated by fish feeding be re-stitched together, forming a more coherent second characteristic separation curve.

[0068] First, the sound pressure level variation of adjacent first feature separation curves is analyzed, and the sound pressure level difference between adjacent curves is calculated. In actual operation, the system takes the same start and end time as the reference, takes a sliding window of the same length for the two adjacent curves, and calculates the average sound pressure level in each window. Then, the mean difference between the two curves is calculated, and a dynamic threshold (obtained from recent background noise statistics) is used as the judgment line. If the difference exceeds the line, it is marked as "sound pressure level abnormal segment", indicating that there may be sudden environmental noise or sensor drift.

[0069] Next, frequency distribution variation analysis is performed on adjacent first feature separation curves to calculate the frequency range overlap between adjacent curves. Specifically, a short-time Fourier transform is performed on each curve to obtain the spectral envelope, and the dominant frequency interval and its energy proportion are extracted. Then, the frequency range overlap is obtained by calculating the ratio of the intersection length to the union length of the dominant frequency intervals of the two curves. If this ratio is lower than a preset lower limit, the two curves are considered to be significantly different in frequency structure and are marked as "frequency structure anomalous segment".

[0070] Subsequently, based on the first difference between the sound pressure level difference and the preset sound pressure level threshold, and the second difference between the frequency range overlap and the preset frequency overlap threshold, the system determines whether adjacent first feature separation curves are qualified in terms of sound pressure level consistency and frequency range consistency. If either the first difference or the second difference exceeds the set tolerance, the system determines that the adjacent curves are inconsistent in the corresponding dimension and writes the inconsistent segment and its corresponding curve number into the queue to be corrected.

[0071] Finally, based on the judgment results, each of the first feature separation curves is corrected, and the corrected first feature separation curves are re-identified and separated to obtain the second feature separation curves, specifically including: When the first difference and / or the second difference exceed the correction threshold, the sound pressure level of the first feature separation curve is dynamically adjusted point-by-point based on recursive dependency. First, adjacent curve segments judged as abnormal are read, and their peak sequences and corresponding background noise levels are extracted. Then, a recursive dependency model is constructed, using the peak amplitude of the previous moment as input to predict the theoretical amplitude at the current moment. If the measured value deviates from the predicted value and exceeds the threshold, the predicted trajectory is pulled back proportionally. This adjustment is only performed within a local time window to avoid over-correction that weakens the real feeding signal. At the same time, the model parameters are updated in real time as the window slides to adapt to the natural fluctuations in sound pressure level of the large yellow croaker at different feeding stages.

[0072] After recursive adjustment, the curve may exhibit minor steps or spikes; therefore, a nonlinear smoothing algorithm based on local weighted regression is introduced. The algorithm selects neighboring samples around each data point and suppresses the influence of outliers far from the center point through a weighting function, achieving a smooth transition while preserving the rising and falling edge characteristics of the feeding pulse. The smoothed curve not only eliminates the abrupt changes caused by the correction but also further highlights the energy concentration area directly related to feeding behavior.

[0073] The enhanced first feature separation curve is re-identified and separated to obtain a more robust second feature separation curve, specifically including: After smoothing the curve, a microwindow is slid along the time axis to capture the local maxima of the sound pressure level, the center of the dominant frequency, and the duration, which serve as key feature points characterizing a potential feeding action, and their precise positions and amplitudes within the entire curve are marked.

[0074] The database compiles standard feeding acoustic templates from multiple batches and different feeding scenarios, covering typical peak-frequency-duration combinations of three types of behaviors: snatching food, steady feeding, and swallowing. These are organized into feeding behavior characteristic data of large yellow croaker and used as a comparison benchmark.

[0075] This embodiment first maps the peak amplitude, dominant frequency center, and duration of key feature points into three-dimensional feature vectors. Then, it invokes density peak clustering, using three typical cluster centers from the database—"fighting for food—high frequency, low amplitude, and short bursts," "stable feeding—medium frequency, medium amplitude, and uniform," and "swallowing—low frequency, high amplitude, and sustained"—as initial centers to classify the feature vectors. After clustering, continuous feature points with the same label and intervals less than a set threshold are grouped into one interval along the time axis, while the remaining positions are naturally separated, thus dividing the entire curve into several continuous segments, each segment being considered a potential feeding behavior event.

[0076] For each segment, the average sound pressure level, dominant frequency energy ratio, and duration are re-extracted and compared with the database template for similarity. If the similarity is higher than the upper limit and adjacent segments are of the same category, they are merged to preserve the integrity of the behavior. If the similarity is lower than the lower limit or cross-class features appear within the segment, it is split again at the dominant frequency abrupt change. The continuous segments obtained after merging or separating are the second feature separation curves, and each curve corresponds to a single and pure feeding behavior event.

[0077] After consistency verification and correction, the second feature separation curve no longer exhibits irregular jumps in sound pressure level and frequency between adjacent time periods, resulting in a smooth transition across the entire curve. The correction process filters out "false peaks" caused by sudden acceleration of aerators, passing boats, or cage swaying, ensuring the continuity of the same feeding event in time and avoiding misjudgments such as "one feeding event being split into two segments" or "two swallowing events being merged." The smoothed curve not only improves the accuracy of subsequent intensity calculations but also significantly reduces the number of false alarms. Farmers no longer need to frequently adjust feeding amounts due to occasional noise, making the system more robust. Verification shows that the corrected curve exhibits higher consistency in cross-day comparisons, providing reliable raw data for establishing long-term feeding records.

[0078] S14. Extract feeding features from each of the corrected second feature separation curves, and calculate the feeding intensity index of each of the second feature separation curves based on the feeding features.

[0079] Specifically, a comprehensive feature extraction is first performed on each corrected second feature separation curve to obtain acoustic feature parameters that can directly characterize the feeding behavior of large yellow croaker. These parameters are uniformly defined as three dimensions—sound pressure level features, frequency features, and time features, which together depict the "fingerprint" of a complete feeding event in an underwater acoustic scene.

[0080] At the sound pressure level characteristic level, this embodiment calculates three indicators over the entire curve: peak sound pressure level, mean sound pressure level, and standard deviation of sound pressure level. Peak sound pressure level reflects the maximum energy release of a single feeding action, mean sound pressure level reflects the average energy level of the entire event, and standard deviation reveals the intensity of energy fluctuations. The combination of the three can avoid the random bias caused by using only the peak value.

[0081] At the frequency characteristic level, after performing a short-time Fourier transform on each curve, three indicators are extracted: dominant frequency position, dominant frequency energy ratio, and spectral centroid. The dominant frequency position indicates the concentrated frequency band of the feeding pulse in the frequency domain, the dominant frequency energy ratio measures the contribution of this frequency band to the total energy, and the spectral centroid further gives the overall frequency "center of gravity". The three indicators work together to distinguish the high-frequency spikes during food grabbing from the low-frequency broad peaks during swallowing.

[0082] At the temporal characteristics level, the event duration, mean pulse interval, and coefficient of variation of pulse interval are recorded: the duration describes the physical length of a feeding behavior from start to finish, the mean pulse interval reflects the speed of the behavior rhythm, and the coefficient of variation reflects the rhythm stability; through these three indicators, continuous rapid pecking and a single long swallow can be clearly distinguished.

[0083] Finally, the nine acoustic characteristic parameters in three categories—sound pressure level, frequency, and time—are input into a normalized weighted model. Based on the weight coefficients of each parameter in the database, a linear combination is performed to output a dimensionless feeding intensity index. The larger the value of this index, the more intense the feeding behavior corresponding to that curve segment, and vice versa. This provides a quantitative basis for direct comparison in subsequent statistical analysis.

[0084] Finally, after normalizing the above nine acoustic characteristic parameters, they are assigned different weights according to their contribution to feeding intensity, and then weighted and combined into a dimensionless feeding intensity index; the higher the index, the more active the feeding, and the lower the index, the more gradual the feeding.

[0085] The dimensionless feeding intensity index condenses three characteristics—sound pressure level, frequency, and time—into a single value, allowing on-site personnel to quickly determine feeding status without needing to understand complex acoustic parameters. A consistently high index with minimal fluctuations indicates the fish are in a vigorous feeding period, and feeding can continue as planned. A high index with significant fluctuations suggests concentrated but short-lived feeding behavior, requiring a shift to a strategy of small, frequent feedings to reduce food loss. A low and stable index indicates the fish are nearing satiation, and timely reduction in feed intake can prevent uneaten food from accumulating and worsening water quality. Furthermore, this index shows an inverse correlation with manually observed uneaten food levels, validating the effectiveness of acoustic quantification and providing a decision threshold for unattended automated feeding.

[0086] S15. Perform statistical analysis on the feeding intensity index to obtain the average feeding intensity and the coefficient of variation of feeding intensity for large yellow croaker.

[0087] First, the feeding intensity indices of all the second-characteristic separation curves are summarized. In this embodiment, after the end of the fixed daily feeding cycle, all the curves obtained by correction and reseparation within that cycle are numbered one by one, and the feeding intensity index corresponding to each curve is written into the same summary sequence; this sequence completely covers the entire process from the start of feeding to the end of feeding, which avoids omissions and ensures continuity in time.

[0088] The sum of feeding intensity indices from all second-characteristic separation curves is calculated, and the average feeding intensity of large yellow croaker is calculated based on the sum and the number of curves. After summing, all indices within the sequence are added together and then divided by the total number of curves; the result is the average feeding intensity. This average value directly reflects the overall feeding activity level of the current cultured batch within a specific time period and can be compared horizontally between different dates or different cages to assess feed attractiveness, environmental comfort, or population health.

[0089] After obtaining the average value, the difference between the index and the average value is taken for each curve. Then, the difference is squared to amplify abnormally high or low fluctuations. All squared terms are summed to form the sum of squared differences to quantify the contribution of each curve to the overall fluctuation and provide the necessary intermediate quantity for subsequent dispersion calculation.

[0090] The coefficient of variation of feeding intensity of large yellow croaker was calculated based on the number of the sum of squared differences and the second characteristic separation curve.

[0091] The dimensionless coefficient of variation of feeding intensity is obtained by dividing the sum of squared differences by the total number of curves and then taking the square root. The smaller the coefficient, the more consistent the feeding rhythm of the fish group and the more stable the food utilization; the larger the coefficient, the more significant the fluctuations in the feeding rhythm, which may be caused by uneven food distribution, sudden environmental changes, or increased individual competition. This provides a quantitative basis for the next step of adjusting the feeding amount or environmental intervention.

[0092] The combined evaluation of average feeding intensity and coefficient of variation quantifies both "group feeding level" and "individual behavior consistency." A high average and low coefficient of variation indicate that the fish are feeding actively and in unison, with optimal feed conversion ratio. A high average and high coefficient of variation suggest that while the fish are generally active, internal competition is intense, requiring an examination of food distribution or the addition of feeding points. A low average and high coefficient of variation reflect insufficient feeding in some individuals or environmental stress, necessitating immediate investigation of water quality, disease, or food palatability. By combining these two indicators, on-site personnel can draw conclusions before daily feed collection, eliminating the need to wait until the next day to weigh the fish, significantly shortening the feedback cycle and reducing the risk of feed waste and water quality deterioration.

[0093] S16. Generate a feeding intensity prediction report for large yellow croaker based on the average feeding intensity and the coefficient of variation of feeding intensity.

[0094] First, quantile normalization analysis is performed on the average feeding intensity to obtain the current feeding intensity level of the large yellow croaker. Specifically, in this embodiment, the average value sequence obtained over several consecutive days is divided into percentiles, and the current value is mapped to a dimensionless interval of 0–100; the mapping result is the "feeding intensity level", which intuitively shows whether the fish population is currently at a low, middle, or high level in the historical range, facilitating both longitudinal comparison and horizontal benchmarking of different cages or batches.

[0095] Subsequently, the coefficient of variation of feeding intensity was analyzed using the entropy weight method to obtain a stability index of the feeding intensity of large yellow croaker. The coefficient of variation sequence was regarded as an information source, and the entropy weight method was used to measure the amount of information carried by its fluctuations: the more disordered the fluctuations, the higher the entropy value and the greater the weight. The final output stability index is expressed as a percentage, with a high value indicating a stable feeding rhythm and a low value indicating potential disturbances.

[0096] Next, based on the changing trends of the average feeding intensity and the coefficient of variation, grey relational analysis is performed to calculate the fluctuation of the future feeding intensity of large yellow croaker. By constructing a grey relational model, the average value and coefficient of variation of the past few days are used as a reference sequence, and the candidate values ​​of the prediction date are used as a comparison sequence; by calculating and sorting the correlation coefficients, the possible upward, stable, or downward range of feeding intensity in the short term can be given, forming an interval estimate of the "fluctuation".

[0097] Finally, by comprehensively considering the feeding intensity level, stability indicators, and fluctuations, a feeding intensity prediction report for large yellow croaker is generated. The report is presented in a combination of charts and text: the charts visually display the current level's position within the historical range, the stability level, and the future fluctuation range; the text provides graded recommendations such as "maintain the current feeding amount," "appropriately reduce feed," or "increase feeding in advance and monitor water quality," enabling farmers to quickly make precise feeding and environmental control decisions based on the quantitative results.

[0098] The forecast report integrates historical percentile levels, stability grades, and future fluctuation ranges into a graphical format, allowing fish farmers to grasp the entire feeding landscape from past to present to future on a single screen. When the report shows that the current level is high and the future range is narrowing, the current feeding rhythm can be maintained with peace of mind, saving manpower; when the level drops and the range widens, it indicates that feeding may decline in the future, and reducing the amount of feed in advance can avoid uneaten food; if the stability index drops sharply, it indicates that there are potential risks to the environment or the health of the fish population, and water quality testing and disease screening can be initiated simultaneously to achieve "prevention before disease."

[0099] This invention presents a method for predicting the feeding intensity of large yellow croaker based on acoustic detection. It processes time-series acoustic signals using multi-scale wavelet transform and adaptive filtering algorithms to effectively separate characteristic signals of the large yellow croaker's feeding behavior. Further integration with time-frequency analysis technology enables precise extraction of sound pressure level and frequency distribution characteristics, thereby accurately identifying different feeding behavior patterns. Dynamic correction and re-separation of feature curves ensure the accuracy and stability of feature extraction. Based on these accurate feature parameters, the calculated feeding intensity index more closely reflects actual feeding conditions. Furthermore, through statistical analysis and trend prediction methods, a prediction report containing the average feeding intensity, coefficient of variation, and future fluctuations can be generated, providing a scientific basis for aquaculture management and enhancing the accuracy and practicality of feeding intensity prediction. This method achieves real-time and accurate monitoring of the large yellow croaker's feeding intensity, reducing the labor intensity and time cost of manual observation. Timely and accurate feeding intensity prediction allows aquaculture personnel to optimize feed delivery strategies, avoid feed waste, and reduce aquaculture costs. Simultaneously, accurate feeding monitoring helps to promptly detect potential health problems, ensuring the healthy growth of large yellow croaker and thus improving the overall efficiency and economic benefits of aquaculture.

[0100] This invention also provides a system for predicting the feeding intensity of large yellow croaker based on acoustic detection, used to execute the method for predicting the feeding intensity of large yellow croaker based on acoustic detection as described above. Figure 3 This is a block diagram of a large yellow croaker feeding intensity prediction system based on acoustic detection, according to an embodiment of the present invention. The system includes: Curve construction module 21 is used to construct a feeding sound wave variation curve of large yellow croaker based on the time-series acoustic wave signal of the target aquaculture area; Feature analysis module 22 is used to analyze the peak sound pressure level and frequency distribution characteristics in the feeding sound wave change curve of the large yellow croaker to obtain the first feature separation curve for different feeding behaviors; The curve separation module 23 is used to verify the consistency of sound pressure level and frequency range of each first feature separation curve by the sound pressure level change and frequency distribution change of adjacent first feature separation curves, and to correct and re-separate each first feature separation curve according to the verification results to obtain the second feature separation curve. The index calculation module 24 is used to extract feeding features from each of the corrected second feature separation curves and calculate the feeding intensity index of each of the second feature separation curves based on the feeding features. The coefficient calculation module 25 is used to perform statistical analysis on the feeding intensity index to obtain the average feeding intensity and the coefficient of variation of feeding intensity of large yellow croaker; The intensity prediction module 26 is used to generate a feeding intensity prediction report for large yellow croaker based on the average feeding intensity and the coefficient of variation of feeding intensity.

[0101] The technical features and effects of the device proposed in the embodiments of the present invention are the same as those of the method proposed in the embodiments of the present invention, and will not be repeated here. Each module in the above-described device can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0102] In summary, compared with the prior art, the beneficial effects of the method and system for predicting the feeding intensity of large yellow croaker based on acoustic detection provided by the embodiments of the present invention are at least one of the following: By processing time-series acoustic signals using multi-scale wavelet transform and adaptive filtering algorithms, characteristic signals of large yellow croaker feeding behavior are effectively separated. Further integration with time-frequency analysis techniques enables precise extraction of sound pressure level and frequency distribution characteristics, thus accurately identifying different feeding behavior patterns. Dynamic correction and re-separation of feature curves ensure the accuracy and stability of feature extraction. Based on these accurate feature parameters, the calculated feeding intensity index more closely reflects actual feeding conditions. Furthermore, statistical analysis and trend prediction methods generate predictive reports containing average feeding intensity, coefficient of variation, and future fluctuations, providing a scientific basis for aquaculture management and enhancing the accuracy and practicality of feeding intensity prediction. This achieves real-time, precise monitoring of large yellow croaker feeding intensity, reducing the labor intensity and time cost of manual observation. Timely and accurate feeding intensity prediction allows aquaculture personnel to optimize feed delivery strategies, avoid feed waste, and reduce aquaculture costs. Simultaneously, precise feeding monitoring helps to promptly detect potential health problems, ensuring the healthy growth of large yellow croaker and thus improving the overall efficiency and economic benefits of aquaculture.

[0103] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for predicting the feeding intensity of large yellow croaker based on acoustic detection, characterized in that, include: Construct a feeding acoustic waveform variation curve for large yellow croaker based on time-series acoustic signals from the target aquaculture area; The peak sound pressure level and frequency distribution characteristics in the feeding sound wave variation curve of the large yellow croaker were analyzed to obtain the first feature separation curve for different feeding behaviors; The consistency of sound pressure level and frequency range of each first feature separation curve is verified by the changes in sound pressure level and frequency distribution of adjacent first feature separation curves. Based on the verification results, each first feature separation curve is corrected and reseparated to obtain the second feature separation curve. Feeding features are extracted from each of the corrected second feature separation curves, and feeding intensity index is calculated for each of the second feature separation curves based on the feeding features; Statistical analysis was performed on the feeding intensity index to obtain the average feeding intensity and the coefficient of variation of feeding intensity for large yellow croaker; A feeding intensity prediction report for large yellow croaker is generated based on the average feeding intensity and the coefficient of variation of feeding intensity.

2. The method for predicting the feeding intensity of large yellow croaker based on acoustic detection as described in claim 1, characterized in that, The construction of the feeding acoustic variation curve of large yellow croaker based on the time-series acoustic signals of the target aquaculture area includes: The acquired time-series acoustic signal is subjected to multi-scale wavelet transform to decompose the signal into first multi-scale characteristic signals of different frequency bands; An adaptive filtering algorithm is applied to dynamically adjust the filtering parameters of the multi-scale feature signal to obtain a second multi-scale feature signal; The second multi-scale feature signal was subjected to time-frequency analysis using Fourier transform and Hilbert transform to obtain an enhanced feature signal. Based on the enhanced feature signal, a feeding sound wave variation curve of the large yellow croaker was constructed.

3. The method for predicting the feeding intensity of large yellow croaker based on acoustic detection as described in claim 1, characterized in that, The analysis of the sound pressure level peak value and frequency distribution characteristics in the feeding sound wave variation curve of the large yellow croaker yields the first feature separation curves for different feeding behaviors, including: Time-domain analysis was performed on the feeding sound wave variation curve of the large yellow croaker to obtain the peak sound pressure level and corresponding timestamp in the sound wave signal; The frequency distribution of the acoustic signal in the feeding acoustic waveform change curve of the large yellow croaker was calculated to obtain the frequency distribution characteristics. Based on the results of time-domain analysis and the frequency distribution characteristics, the peak sound pressure level is matched with the corresponding frequency components to form sound pressure level-frequency feature points; Based on the time series distribution of the sound pressure level-frequency feature points, different feeding behavior patterns are identified, and each feeding behavior pattern corresponds to a sequence of sound pressure level-frequency feature points, forming a first feature separation curve.

4. The method for predicting the feeding intensity of large yellow croaker based on acoustic detection as described in claim 1, characterized in that, The consistency of sound pressure level and frequency range of each of the first feature separation curves is verified by the changes in sound pressure level and frequency distribution of adjacent first feature separation curves. Based on the verification results, each of the first feature separation curves is corrected and reseparated to obtain the second feature separation curve, including: The sound pressure level variation of adjacent first feature separation curves is analyzed, and the sound pressure level difference between adjacent curves is calculated. Perform frequency distribution variation analysis on adjacent first feature separation curves and calculate the frequency range overlap between adjacent curves; Based on the first difference between the sound pressure level difference and the preset sound pressure level threshold, and the second difference between the frequency range overlap and the preset frequency overlap threshold, the sound pressure level consistency and frequency range consistency of adjacent first feature separation curves are determined. Based on the judgment results, each of the first feature separation curves is corrected, and the corrected first feature separation curves are re-identified and separated to obtain the second feature separation curve.

5. The method for predicting the feeding intensity of large yellow croaker based on acoustic detection as described in claim 4, characterized in that, The step of correcting each of the first feature separation curves based on the judgment result includes: When the first difference and / or the second difference exceed the correction threshold, the sound pressure level of the first feature separation curve is dynamically adjusted point by point based on recursive dependence. The adjusted first feature separation curve is nonlinearly smoothed to eliminate abrupt changes introduced by the correction and to enhance the features. The enhanced first feature separation curve is re-identified and separated to obtain the second feature separation curve.

6. The method for predicting the feeding intensity of large yellow croaker based on acoustic detection as described in claim 5, characterized in that, The process of re-identifying and separating the enhanced first feature separation curve to obtain the second feature separation curve includes: Obtain the key feature points of the enhanced first feature separation curve; Obtain feeding behavior data of large yellow croaker from a farmed fish database; Using the key feature points as anchor points, cluster analysis is performed in conjunction with the feeding behavior feature data of the large yellow croaker. Based on the analysis results, the first feature separation curve is separated into multiple sub-segments, and each sub-segment corresponds to a potential feeding behavior event. Calculate the feeding behavior characteristics of each sub-segment, classify each sub-segment according to the feeding behavior characteristics, and merge or separate each sub-segment according to the classification results to obtain the second feature separation curve.

7. The method for predicting the feeding intensity of large yellow croaker based on acoustic detection as described in claim 1, characterized in that, The step of extracting feeding features from each of the corrected second feature separation curves and calculating the feeding intensity index of each second feature separation curve based on the feeding features includes: Feature extraction is performed on each of the corrected second feature separation curves to obtain acoustic feature parameters characterizing the feeding behavior of large yellow croaker. The acoustic feature parameters include sound pressure level features, frequency features, and time features of the feeding behavior of large yellow croaker. Based on the acoustic characteristic parameters, the feeding intensity index of each of the second characteristic separation curves is calculated.

8. The method for predicting the feeding intensity of large yellow croaker based on acoustic detection as described in claim 1, characterized in that, The statistical analysis of the feeding intensity index yields the average feeding intensity and the coefficient of variation of feeding intensity for large yellow croaker, including: The feeding intensity indices are summarized, and the sum of the feeding intensity indices of all the second characteristic separation curves is calculated. Based on the sum and the number of the second characteristic separation curves, the average feeding intensity of the large yellow croaker is calculated. Calculate the difference between the feeding intensity index of each of the second feature separation curves and the average feeding intensity, and square the difference to obtain the sum of squares of the differences. Based on the sum of squares of the differences and the number of the second feature separation curves, calculate the coefficient of variation of the feeding intensity of the large yellow croaker.

9. The method for predicting the feeding intensity of large yellow croaker based on acoustic detection as described in claim 1, characterized in that, The step of generating a feeding intensity prediction report for large yellow croaker based on the average feeding intensity and the coefficient of variation of feeding intensity includes: The current feeding intensity level of the large yellow croaker is obtained by performing quantile normalization analysis on the average feeding intensity. The coefficient of variation of the feeding intensity was analyzed using the entropy weight method to obtain a stability index of the feeding intensity of large yellow croaker; Based on the changing trends of the average feeding intensity and the coefficient of variation of the feeding intensity, grey relational analysis is performed to calculate the fluctuation of the future feeding intensity of the large yellow croaker; A feeding intensity prediction report for large yellow croaker is generated based on the feeding intensity level, the stability index, and the fluctuation.

10. A system for predicting the feeding intensity of large yellow croaker based on acoustic detection, characterized in that it is used for, include: The curve construction module is used to construct a feeding acoustic wave variation curve of large yellow croaker based on the time-series acoustic wave signal of the target aquaculture area. The feature analysis module is used to analyze the peak sound pressure level and frequency distribution characteristics in the feeding sound wave change curve of the large yellow croaker to obtain the first feature separation curve for different feeding behaviors; The curve separation module is used to verify the consistency of sound pressure level and frequency range of each first feature separation curve by the changes in sound pressure level and frequency distribution of adjacent first feature separation curves, and to correct and re-separate each first feature separation curve according to the verification results to obtain the second feature separation curve. The index calculation module is used to extract feeding features from each of the corrected second feature separation curves and calculate the feeding intensity index of each of the second feature separation curves based on the feeding features. The coefficient calculation module is used to perform statistical analysis on the feeding intensity index to obtain the average feeding intensity and the coefficient of variation of feeding intensity of large yellow croaker; The intensity prediction module is used to generate a feeding intensity prediction report for large yellow croaker based on the average feeding intensity and the coefficient of variation of the feeding intensity.

Images