Marine ranch biological behavior analysis method

Abstract

The invention discloses a marine ranch biological behavior analysis method, and belongs to the field of biological behavior analysis, and the method comprises the steps: collecting original image data through a high-resolution underwater camera device, extracting tiny motion features through a deep learning algorithm, achieving the fixed-point tracking of a key individual through the combination of acoustic positioning and visual recognition, and generating time sequence data; further analyzing association between behaviors and the environment through a rate quantification algorithm and an association model, decomposing a complex behavior sequence by adopting a transaction processing mechanism, extracting high-intensity activity features and evaluating energy consumption; and finally, in combination with the appearance change characteristic, mapping a behavior state, coding data and realizing seamless transmission and sharing through a standard communication protocol. According to the method, the accuracy of behavior recognition and the efficiency of data analysis are remarkably improved, and a scientific basis is provided for marine ecological research.

Description

Technical Field

[0001] This invention belongs to the field of biological behavior analysis, and in particular relates to a method for analyzing the biological behavior of marine ranches. Background Technology

[0002] Marine ranching, as an important direction for the development of modern fisheries, has irreplaceable value in ensuring the sustainable use of marine resources and ecological balance. In-depth research on the behavior of marine organisms can effectively improve the aquaculture efficiency and ecological management level of ranches. However, research and application in this field face many challenges, urgently requiring breakthroughs in existing technological limitations and the exploration of more precise and systematic analytical methods.

[0003] Currently, although some methods have been developed to monitor and analyze marine biological behavior, most remain at a superficial observation level, lacking in-depth analysis of behavioral details and comprehensive consideration of dynamic relationships. Especially in complex marine environments, the diversity and transient changes in biological behavior are often overlooked, making it difficult for analytical results to reflect the true ecological situation and providing precise guidance for ranch management. This limitation not only affects the assessment of biological health and environmental adaptability but also restricts the efficient allocation of ranch resources.

[0004] A deeper technical challenge lies in capturing and analyzing the subtle characteristics of marine organism behavior and dynamically linking them to environmental factors. Firstly, marine organism behavior often involves minute changes in movement, such as the subtle posture adjustments fish make while feeding or positional shifts during group interactions. Capturing these subtle characteristics requires extremely high technical precision, which current equipment and methods struggle to achieve. The inability to accurately identify these behavioral details further leads to a lack of understanding of the causal relationship between biological behavior and environmental conditions; for example, it's impossible to determine whether changes in water flow directly affect fish swimming patterns. This progressive problem, from subtle behavior to environmental correlation, has become a bottleneck for current technological breakthroughs.

[0005] Therefore, accurately capturing subtle characteristics of biological behavior in the complex and ever-changing marine environment and establishing a dynamic correlation mechanism between behavior and environmental factors has become a key issue in improving the management level of marine ranches. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method for analyzing the biological behavior of marine ranches, comprising:

[0007] Raw behavioral image data of marine organisms are collected, and deep learning algorithms are used to extract fine-grained features from the image sequences to obtain a quantitative description of the organisms' minute movements.

[0008] Based on the fine-grained features, acoustic positioning points are set in a specific preset area and combined with visual recognition to perform point tracking of biological individuals, record behavioral change patterns and generate time series data.

[0009] The behavior execution speed index is obtained from the time series data, and the numerical distribution of the execution speed index is calculated using a rate quantization algorithm to construct a correlation model between behavior rate and environmental conditions.

[0010] The complex behavioral sequences in the association model are decomposed into independent units using a transaction processing mechanism. Each unit contains pattern recognition and anomaly detection operations, resulting in a processed behavioral transaction dataset.

[0011] High-intensity activity features are extracted from the behavioral transaction dataset. The frequency and duration of the high-intensity activity features are analyzed using a high-energy consumption monitoring method. If the frequency exceeds a preset threshold, it is marked as a high-energy consumption state, and the energy consumption assessment result is obtained.

[0012] Based on the energy consumption assessment results and combined with changes in biological morphology, an association analysis model is used to map the relationship between morphological features and behavioral states, resulting in an enhanced and accurate dataset for behavior recognition.

[0013] The enhanced behavior recognition accurate dataset is encoded with behavior type, timestamp, and spatial coordinates. Data packets are formatted using communication protocol standards for seamless transmission and sharing between different devices, resulting in unified biological behavior analysis results.

[0014] Compared with the prior art, the present invention has the following advantages and technical effects:

[0015] This invention discloses a marine organism behavior analysis method based on deep learning and multimodal data fusion. Addressing the challenges of fine-grained feature extraction, complex modeling of behavior-environment correlations, and poor data sharing across devices in marine organism behavior monitoring, it proposes an integrated solution. By combining high-resolution underwater image acquisition with deep learning algorithms, it accurately extracts minute biological movement features and uses acoustic localization and visual recognition to achieve individual point tracking, constructing a correlation model between behavior and environmental conditions. Simultaneously, it employs a transaction processing mechanism to decompose complex behavior sequences, combining high-energy-consumption monitoring with correlation analysis to map the relationship between behavioral states and physical characteristics, improving recognition accuracy. Finally, it uses standard communication protocols to format data packets, ensuring seamless data transmission and sharing. This invention significantly improves the intelligence level of marine organism behavior monitoring through efficient multimodal data fusion and precise quantification of behavior analysis, providing reliable technical support for marine ecological research.

[0016] This invention deeply integrates multi-source data fusion with behavior analysis, which not only improves the accuracy of behavior recognition but also enables efficient data sharing across devices, providing scientific basis and intelligent support for marine ecological research. Attached Figure Description

[0017] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0018] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0019] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0021] like Figure 1 As shown, this embodiment provides a method for analyzing the behavior of marine ranch organisms, including:

[0022] We collect raw behavioral image data of marine organisms and use deep learning algorithms to extract fine-grained features from the image sequences to obtain a quantitative description of the organisms' minute movements.

[0023] Based on fine-grained characteristics, acoustic positioning points are set in a specific preset area and combined with visual recognition to track biological individuals at fixed points, record behavioral change patterns and generate time series data.

[0024] The behavior execution speed index is obtained from time series data, and the numerical distribution of the execution speed index is calculated using a rate quantization algorithm. A correlation model between behavior speed and environmental conditions is then constructed.

[0025] A transaction processing mechanism is used to decompose the complex behavioral sequences in the association model into independent units. Each unit contains pattern recognition and anomaly detection operations, resulting in a processed behavioral transaction dataset.

[0026] High-intensity activity features are extracted from behavioral transaction datasets. High-energy consumption monitoring methods are used to analyze the frequency and duration of high-intensity activity features. If the frequency exceeds a preset threshold, it is marked as a high-energy-consumption state, and energy consumption assessment results are obtained.

[0027] Based on the energy consumption assessment results and combined with changes in biological morphology, an association analysis model was used to map the relationship between morphological features and behavioral states, resulting in an enhanced and accurate dataset for behavior recognition.

[0028] The enhanced behavior recognition dataset is encoded with behavior type, timestamp, and spatial coordinates. Data packets are formatted using communication protocol standards for seamless transmission and sharing between different devices, resulting in unified biological behavior analysis results.

[0029] Furthermore, the process of obtaining a quantitative description of the minute movements of organisms includes:

[0030] Raw image data is collected using underwater camera equipment to obtain image sequences containing data on marine biological behavior, thus obtaining preliminary visual records;

[0031] A pre-trained deep learning model is used to process the image sequence, extracting fine-grained features from each frame to obtain the subtle details of the organism's movements.

[0032] Based on the minute movement details of organisms, temporal analysis is performed on consecutive frame images to obtain the change patterns of swimming posture and to determine the dynamic performance of posture features.

[0033] If an abnormal frame is detected in the swimming posture change pattern, a second feature extraction is performed on the abnormal frame to obtain motion description information;

[0034] By analyzing the characteristic data related to mouth opening and closing, the frequency and amplitude of mouth movements are obtained, and quantitative indicators of mouth behavior are determined.

[0035] Based on quantitative indicators of swimming posture and mouth behavior, combined with time-series data, a comprehensive comparison is made to determine whether the organism's behavioral pattern conforms to the preset normal range.

[0036] If the behavior pattern exceeds the preset range, the relevant image sequences are marked to obtain detailed feature records of the abnormal behavior.

[0037] For example, in this embodiment, behavior recording is performed on a specific fish species, such as salmon. When acquiring high-resolution image data using underwater cameras, assuming the cameras are deployed in the salmon's migratory channel, 30 frames of high-definition images are captured per second, forming a continuous image sequence. These images have a resolution of 4K, ensuring that subtle details of the fish's movements, such as slight fin movements or minute tail twists, are captured. This high-frequency, high-definition acquisition method provides ample visual data support for subsequent analysis, significantly improving the accuracy of behavior recognition.

[0038] For example, when processing image sequences using pre-trained deep learning models, convolutional neural network-based models are used to identify feature points of various parts of a fish's body. For each frame, the model extracts fine-grained features such as fin angles and body curvature. For instance, if a salmon fin wagging angle of 15 degrees is detected in one frame, while it is 18 degrees in an adjacent frame, this minute change is recorded as the basis for subsequent temporal analysis. This fine-grained feature extraction helps capture instantaneous changes in biological behavior, laying the foundation for dynamic pattern analysis.

[0039] For example, when performing time-series analysis to obtain swimming posture change patterns, the system analyzes the salmon's body curvature and tail swaying frequency over a 5-second period using feature data from consecutive frames. Assuming a salmon normally sways its tail 3 times per second, a sudden drop in swaying frequency to 1 time per second, coupled with an abnormally increased body curvature, might indicate water resistance or a health problem. In this case, the system marks the relevant frames as anomalous and performs secondary feature extraction, refining the analysis of tail muscle contraction amplitude to obtain a more accurate description of the movement. This secondary extraction effectively improves the accuracy of identifying abnormal behavior.

[0040] For example, when analyzing mouth opening and closing characteristics, one can focus on the salmon's mouth movements, counting the number of times it opens and closes per minute and the amplitude of each opening and closing. Assuming that under normal circumstances, a salmon's mouth opens and closes 20 times per minute, with each opening and closing amplitude being approximately 2 centimeters, but in a certain sequence, the opening and closing frequency increases to 40 times per minute while the amplitude decreases to 1 centimeter, this may indicate that the salmon is in a state of foraging or respiratory distress. Recording quantitative indicators can provide data support for judging behavioral patterns and improve the scientific rigor of the analysis.

[0041] For example, when comprehensively comparing swimming postures and mouth behavior, the aforementioned quantitative indicators can be matched with preset normal ranges. Assuming the normal range is a tail wagging frequency of 2-4 times / second and a mouth opening and closing frequency of 15-25 times / minute, if a sequence is detected to exceed this range, the system will automatically label the image sequence and record detailed characteristics of the abnormal behavior, such as an abnormal duration of 10 seconds involving a tail wagging frequency reduced to 1 time / second. This comprehensive analysis and labeling mechanism helps to quickly locate problematic behaviors, providing a basis for subsequent research or conservation measures.

[0042] For example, after recording abnormal behavioral characteristics, this embodiment further analyzes their relationship with environmental factors, such as whether water temperature or water flow speed is a contributing factor. For instance, a sudden drop in water temperature from the normal 15 degrees Celsius to 10 degrees Celsius might lead to abnormal salmon behavior; this correlation analysis can provide more comprehensive data support for ecological conservation. Through the above multi-dimensional analysis, not only can abnormal behavior be accurately identified, but scientific evidence can also be provided for marine life research and conservation, significantly improving monitoring efficiency and effectiveness.

[0043] Furthermore, the process of recording behavioral patterns and generating time series data includes:

[0044] Within a specific preset area, an area layout method is used to set acoustic positioning points and combine them with visual recognition technology to initially locate biological individuals, obtain spatial location information, and determine the initial tracking range.

[0045] Based on the initial tracking range, a fixed-point tracking method is used to continuously monitor key organisms. Real-time location signals are obtained from acoustic positioning points, and fine-grained features are extracted by combining visual recognition techniques to obtain a preliminary record of behavioral change patterns.

[0046] Based on the initial record of behavioral change patterns, the movement paths of key organisms are continuously collected using the tracking and recording method to obtain path change data and determine whether the activity patterns conform to the preset patterns.

[0047] If the path change data deviates from the preset pattern, high-frequency signals are collected from the abnormal location through acoustic positioning points, and images of the abnormal behavior are captured by visual recognition technology to obtain behavior change information.

[0048] Based on behavioral change information, time series data generation methods are used to correlate behavioral change patterns with the time dimension to obtain continuous time series data and determine long-term behavioral trends.

[0049] The time series data is segmented and processed using a data generation method. The behavioral change patterns of each time period are stored independently to obtain structured behavioral record data.

[0050] If an abnormal trend is detected in the structured behavioral record data, a pre-established convolutional neural network model is used to perform in-depth analysis of the data during the abnormal time period to obtain potential behavioral change patterns and determine whether further attention is needed.

[0051] For example, in this embodiment, when studying the behavioral characteristics of marine organisms, multiple acoustic positioning points are set up in a specific sea area using a regional deployment method, forming a monitoring grid covering an area of ​​500 square meters to initially locate the spatial position of the target organism. For key organisms such as a certain cetacean, the initial tracking range may be set within a radius of 100 meters, and its approximate location is estimated by using the time difference of acoustic signal reflection. The core of this method lies in using the speed of sound wave propagation and the time difference of reception to construct three-dimensional spatial coordinates, thereby laying the foundation for subsequent precise tracking.

[0052] For example, in the fixed-point tracking method, underwater cameras and sonar equipment are deployed at key points within an initial range to continuously monitor the movement trajectory of the target organism. Assuming acoustic signals are collected every 5 seconds, this is combined with visual recognition technology to capture changes in the organism's posture, such as the frequency of tail fin wagging or changes in body tilt angle. This dual-method approach can acquire behavioral data from both acoustic and visual dimensions, forming a preliminary record of behavioral change patterns.

[0053] For example, continuous acquisition of migration paths involves tracking and recording the target organism's activity trajectory over 24 hours. If its path deviates from a pre-defined straight-line migration pattern and instead exhibits frequent circular movements, the signal acquisition frequency can be increased to once per second using acoustic localization points. Simultaneously, cameras can capture images of abnormal behavior, such as whether it is being disturbed by external forces or exhibiting foraging behavior. This combination of high-frequency acquisition and image recording facilitates a more comprehensive analysis of the causes of abnormal behavior.

[0054] For example, in the generation of time series data, behavioral patterns can be correlated with the time dimension. Assuming an hourly unit, the activity intensity changes of a target organism at different time periods can be recorded. By segmenting the data, a day can be divided into 24 time periods, and the behavioral data for each period can be stored separately, forming a structured record. This segmented storage method facilitates subsequent targeted analysis of behavioral trends within a specific time period.

[0055] For example, for in-depth analysis of abnormal trends, this embodiment utilizes a pre-built neural network model to process data from abnormal time periods. Suppose that within a certain time period, the activity frequency of a target organism suddenly decreases by 50%. The model can compare this with historical data to determine whether it may be related to environmental changes or health problems. This in-depth analysis can provide important references for subsequent research and promptly identify potential problems.

[0056] For example, the combined application of acoustic positioning and visual recognition technologies throughout the process not only improves the accuracy of data acquisition but also characterizes biological behavior from multiple dimensions. This multi-pronged approach helps build a more comprehensive behavioral database, providing a reliable basis for the protection and research of marine life, and also providing technical support for early warning of abnormal behavior.

[0057] Furthermore, the process of constructing a correlation model between behavioral rate and environmental conditions includes:

[0058] Behavioral indicators are obtained from time series data, and data processing tools are used to clean and format the raw data to obtain a preliminary behavioral dataset.

[0059] Based on the preliminary organized behavioral dataset, the rate quantization algorithm was applied to calculate the numerical distribution of swimming speed and feeding frequency in the execution speed index, and the range of change of each index in different time periods was determined.

[0060] If the range of change exceeds the preset threshold, the numerical distribution is segmented to obtain the time points of abnormal fluctuations and the corresponding behavioral indicator values ​​to determine whether there is an abnormal rate.

[0061] By comparing the time points and behavioral indicator values ​​of abnormal fluctuations, data records related to environmental conditions are obtained, and the specific manifestations of environmental conditions at the time points of abnormal fluctuations are determined.

[0062] Based on the specific manifestations of environmental conditions, a regression analysis algorithm is used to construct a correlation model between behavioral rate and environmental conditions.

[0063] For example, in this embodiment, when studying behavioral changes in key organisms within a specific region, time-series data is used to analyze their behavioral indicators. First, the raw data is cleaned and formatted. Data processing tools are used to denoise the collected behavioral data, such as removing invalid records caused by device vibration, to ensure data accuracy. The cleaned data is then organized into a standard format according to time sequence for easy subsequent analysis.

[0064] For example, in calculating the swimming speed and feeding frequency, this embodiment uses a rate quantification algorithm to divide the swimming speed records for different time periods throughout the day into multiple intervals. Assuming an organism's average swimming speed is 0.5 m / s in the morning and decreases to 0.2 m / s in the evening, this numerical distribution provides a clear visual indication of the changes in its activity intensity. Regarding feeding frequency, assuming 10 feeding behaviors are recorded in the morning and only 2 at night, this distribution difference can help determine its lifestyle habits.

[0065] For example, if the variation in swimming speed exceeds a preset threshold—for instance, if the normal range is 0.3 to 0.6 meters per second, but reaches 0.8 meters per second at a certain time—this abnormal fluctuation needs to be segmented to identify the specific time point, such as 10:00 AM, and the corresponding behavioral indicator value recorded to determine whether the abnormal speed is caused by external interference. This method helps to detect potential problems in a timely manner.

[0066] For example, when comparing abnormal fluctuation times with environmental conditions, this embodiment obtains data records such as water temperature and flow rate at that time. Suppose that the water temperature suddenly rises to 25 degrees Celsius at 10:00 AM, while the normal temperature is 20 degrees Celsius. This environmental change may be associated with abnormal swimming speed. Through this comparison, the influence of the environment on behavior can be preliminarily inferred.

[0067] For example, in constructing a correlation model between behavior rate and environmental conditions, this embodiment uses a regression analysis algorithm to analyze the relationship between swimming speed and water temperature, deriving a correlation coefficient. If the coefficient shows a high correlation, such as above 0.8, it indicates that water temperature may be an important factor affecting behavior rate. After optimizing and adjusting the model, this relationship can be reflected more accurately.

[0068] For example, when making predictions based on new time-series data, optimized models can be used to infer future trends in behavioral rates. Assuming a predicted water temperature of 24 degrees Celsius tomorrow, the model might show swimming speeds increasing to 0.7 meters per second. The degree to which these predictions match actual environmental conditions can provide a reference for subsequent research.

[0069] By employing the methods described above, we can gain a more comprehensive understanding of the behavioral patterns of key organisms and their interactions with environmental factors. This analysis not only helps to reveal the driving factors behind behavior but also provides data support for conservation and management, bringing long-term practical value.

[0070] Furthermore, the process of obtaining the processed behavioral transaction dataset includes:

[0071] The complex behavior is broken down into multiple sub-sequence segments using a segmentation technique, resulting in a preliminary set of decomposed sequences;

[0072] Based on the initially decomposed sequence set, a pre-established pattern recognition rule base is applied to each sub-sequence segment to determine whether there is a pattern that conforms to the rules, and to determine the pattern classification result of each segment.

[0073] If the pattern classification result shows that a certain subsequence segment does not match any rules, it is marked as a potential anomalous segment, and the marked anomalous candidate set is obtained;

[0074] The support vector machine algorithm is used to extract features and classify each potential anomaly fragment in the marked anomaly candidate set to determine whether it is a real anomaly and obtain the anomaly confirmation set.

[0075] Based on the abnormal confirmation set and combined with the transaction processing mechanism, abnormal fragments and normal fragments are encapsulated into independent units to generate a structured set of behavioral transaction units.

[0076] By integrating the behavioral transaction unit sets and adopting a unified formatted storage method, a complete behavioral transaction dataset is constructed.

[0077] Each cell in the behavioral transaction dataset is validated. If the validation finds missing data or inconsistent format, an automatic completion and correction process is triggered to obtain a complete and consistent behavioral transaction dataset.

[0078] For example, when studying the behavioral patterns of aquatic organisms, this embodiment breaks down complex swimming and feeding behaviors into multiple sub-sequence segments through preliminary analysis of the behavioral sequences. Assuming a fish's activities throughout the day are recorded as a long sequence, preliminary analysis can divide it into sub-sequence segments for three time periods: morning, noon, and evening. Each segment contains specific action data. This segmentation technique helps to break down complex overall behavior into smaller, more easily analyzable units, providing a foundation for subsequent pattern recognition.

[0079] For example, for the initially decomposed sequence set, this embodiment applies a pre-established pattern recognition rule base to determine whether each sub-sequence segment conforms to a known behavioral pattern. Assuming the rule base defines the frequency range of normal swimming behavior as 10 to 20 times per minute, if a morning segment has a frequency of 15 times, it is classified as a normal pattern; if an evening segment has a frequency of only 5 times, it does not match the rule and is marked as a potentially abnormal segment. This method can quickly filter out behavioral segments that may have problems.

[0080] For example, when processing potentially anomalous segments, this embodiment employs a support vector machine (SVM) algorithm for feature extraction and classification, which can more accurately determine whether a segment is a genuine anomaly. Suppose a segment exhibits a swimming frequency as low as 5 times per minute, accompanied by excessively long periods of inactivity. By extracting both frequency and inactivity time as features, the algorithm might classify it as a genuine anomaly, possibly due to reduced activity caused by a sudden drop in water temperature. This method effectively distinguishes between genuine anomalies and data noise.

[0081] For example, for anomaly confirmation sets, this embodiment combines transaction processing mechanisms to encapsulate abnormal segments and normal segments into independent units. Assuming the abnormal segment is labeled "low temperature impact" and the normal segment is labeled "routine activity," they are stored as independent behavioral transaction units. This structured processing facilitates subsequent tracing and analysis of the causes of anomalies.

[0082] For example, when integrating behavioral transaction unit sets, this embodiment adopts a unified formatted storage method to ensure data consistency. It assumes that all units are stored using timestamps and behavior types as primary keys to construct a complete behavioral transaction dataset. This approach facilitates data retrieval and comparative analysis across time periods.

[0083] For example, during the final verification of the behavioral transaction dataset, if a cell is found to be missing timestamp data, an automatic completion process is triggered, inferring the missing value and correcting the format based on the preceding and following time points. This verification and correction mechanism ensures the integrity of the dataset, providing reliable data support for subsequent research. Through these multi-stage refined processing steps, the accuracy and practicality of behavioral data analysis can be significantly improved, laying a solid foundation for exploring the relationship between behavior and environmental factors.

[0084] Furthermore, the process of obtaining energy consumption assessment results includes:

[0085] Raw records are obtained from behavioral transaction data. Information related to high-intensity activities is filtered to separate activity segments that involve frequent swimming and rapid chasing, resulting in a preliminary classification of the data set.

[0086] For the initially classified dataset, a time window segmentation method is used to calculate the frequency and duration of frequent swimming and fast chasing, and to determine the quantitative indicators for each activity segment;

[0087] Based on quantitative indicators and compared with preset thresholds, if the frequency or duration exceeds the preset threshold, the corresponding activity segment is marked as a high-energy-consuming state, and a set of marked activity states is obtained.

[0088] By using the marked set of activity states and employing high-energy-consumption monitoring methods, we can analyze the distribution of high-energy-consumption states in the overall data and obtain the proportion and temporal distribution characteristics of high-energy-consumption segments.

[0089] Based on the proportion and temporal distribution characteristics of high-energy-consuming segments, the energy consumption is predicted using a support vector machine model to determine the energy consumption level for each time period.

[0090] By combining the predicted energy consumption levels with the duration and frequency of activity segments, energy consumption assessment results for each activity segment are generated, and the final energy consumption distribution is determined.

[0091] Based on the final energy consumption distribution, correlation analysis is performed on the activity segments in the high-energy-consuming state to obtain the behavioral patterns of frequent swimming and rapid chasing in the high-energy-consuming state, and to obtain the mapping relationship between behavior and energy consumption.

[0092] For example, this embodiment focuses on the screening and analysis of high-intensity activities by extracting data segments related to frequent swimming and rapid chasing from raw records. Assuming an aquatic organism monitoring scenario, the raw data records the activity trajectory of a certain fish over 24 hours. By filtering segments with speeds exceeding 2 meters per second and segments showing continuous swimming for more than 5 minutes, a preliminary set of high-intensity activities is identified. This process focuses on feature extraction from behavioral data, laying the foundation for subsequent analysis.

[0093] For example, this embodiment uses a time window segmentation method to divide 24 hours of data into 30-minute time windows, and counts the number of frequent swimming events and the duration of rapid chasing events within each window. Suppose that within a certain window, frequent swimming occurs 10 times, each lasting an average of 2 minutes, while rapid chasing occurs 3 times, each lasting 1 minute. By comparing this with preset thresholds such as 8 swimming events and 1.5 minutes of duration, the activity segment within that window is determined to be a high-energy-consuming state. This quantification method helps to accurately locate the time distribution of high-intensity activities.

[0094] For example, after marking high-energy-consuming states, high-energy-consuming monitoring methods are used to analyze their distribution. Assuming that high-energy-consuming segments account for 20% of the daily data, and are mainly concentrated in the time periods of 6:00-8:00 AM and 5:00-7:00 PM, this indicates that these periods may be peak times for fish to forage or escape predators. Extracting this distribution characteristic provides an important basis for subsequent energy consumption prediction.

[0095] For example, in this embodiment, when using a support vector machine model to predict energy consumption, the model is trained based on historical data. The frequency and duration of high-energy-consuming segments are input, and the energy consumption level for each time period is output. For instance, the segment from 6 AM to 8 AM is predicted as high-level consumption, while other time periods are predicted as medium-to-low levels. This helps identify the energy demand characteristics of key time periods.

[0096] For example, when generating energy consumption assessment results, combining duration and frequency, assuming a high-energy-consuming segment lasts for 10 minutes and occurs 5 times per hour, the assessment results may show that the energy consumption of this segment accounts for 15% of the total energy consumption of the day, thus intuitively reflecting the impact of different activity segments on the overall energy distribution.

[0097] For example, this embodiment, through behavioral pattern correlation analysis under high-energy-consuming conditions, reveals that frequent swimming is highly correlated with foraging behavior in the morning, while rapid chasing is mostly associated with escape behavior in the evening. Establishing this mapping relationship helps to understand the driving forces behind the behavior and provides a reference for subsequent behavioral interventions or resource allocation.

[0098] This embodiment, through the aforementioned series of analytical methods, effectively elucidates the intrinsic relationship between behavior and energy consumption. This not only enhances the understanding of the characteristics of high-intensity activities but also provides data support for optimizing monitoring strategies and resource management. This multi-layered, multi-faceted analytical approach ensures the completeness and reliability of the process from data extraction to result evaluation.

[0099] Furthermore, the process of obtaining an enhanced, accurate dataset for behavior recognition includes:

[0100] By collecting and organizing energy consumption data, key indicators are extracted from the original records to form a preliminary energy consumption feature set;

[0101] Based on the image data of body color changes and fin spread, image processing techniques were used to separate the external features of the birth body to obtain a set of external feature descriptions;

[0102] Based on the set of shape feature descriptions and the set of energy consumption features, a correlation analysis model is constructed, a data mapping operation is performed, and the correspondence between shape features and energy consumption features is determined.

[0103] If the correspondence output by the correlation analysis model is lower than the preset threshold, then a second feature extraction is performed on the image data of body color change and fin spread to obtain a more refined subset of shape features.

[0104] Using known labeled data of a subset of physical features and behavioral states, a support vector machine model is used for classification training to determine the category of the behavioral state.

[0105] Based on the classification training results and combined with the logical rules of state analysis, an enhanced behavior recognition dataset is generated.

[0106] If the recognition accuracy still does not meet the preset standard, the dataset generation process will be iteratively optimized, and the feature extraction parameters will be readjusted until an accurate dataset for behavior recognition that meets the requirements is obtained.

[0107] For example, in the process of collecting and processing energy consumption data in this embodiment, key indicators such as heart rate and activity frequency are extracted by analyzing biological activity logs. Suppose that in a certain monitoring session, the record shows that an organism's heart rate remained 20% above the average for 30 minutes; this is then categorized as a high-energy-consumption period, forming a preliminary energy consumption feature set. This method helps to quickly filter out data that may be related to high-intensity activity, laying the foundation for subsequent analysis.

[0108] For example, in this embodiment, image segmentation technology is used to process image data related to body color changes and fin deployment, separating the organism's external features from the background. If, during a particular image acquisition, a change in body color brightness exceeds a preset value of 15%, and the fin deployment angle is greater than 60 degrees, these are marked as significant external features. This refined extraction helps to more accurately capture changes in the organism's state, providing a reliable basis for subsequent correlation analysis.

[0109] For example, in this embodiment, when constructing the association model between the set of external features and the set of energy consumption features, a mapping relationship can be trained using historical data. Suppose the model output shows that the correlation coefficient between body color change and energy consumption is 0.8, while the correlation coefficient for fin deployment is only 0.3, which does not reach the preset threshold of 0.5. Therefore, secondary feature extraction is required for the fin deployment data. This secondary extraction can focus on the frequency and duration of fin deployment, further refining the feature subset and improving the model's matching accuracy.

[0110] For example, this embodiment trains the classification of a subset of physical features and behavioral states using a support vector machine model on known labeled data. Suppose that an organism is labeled as being in a defensive state 80% of the time when its body color deepens and its fins frequently unfold; the model can then determine its behavioral category accordingly. This classification training helps transform complex physical features into recognizable behavioral patterns.

[0111] For example, in generating the enhanced behavior recognition dataset, this embodiment combines state analysis logic rules with temporal distribution features. If an organism exhibits a defensive state for 70% of the time within a specific time period, it can be identified as exhibiting highly vigilant behavior and included in the enhanced dataset. This approach effectively improves recognition accuracy and meets business requirements.

[0112] For example, if the recognition accuracy does not meet the preset standard, the feature extraction parameters can be adjusted through iterative optimization. Assuming the initial accuracy is 75% and has not reached the target of 85%, the threshold range for body color changes in image processing can be readjusted, decreasing from 15% to 10% to capture more subtle variations. This optimization method gradually approaches the target accuracy, ensuring the usability of the dataset.

[0113] Furthermore, the process of obtaining unified biological behavior analysis results includes:

[0114] Based on the enhanced behavior recognition accurate dataset, behavior type, timestamp and spatial coordinates are obtained to construct the initial dataset. Outliers are removed using preset cleaning rules to obtain the processed basic data set.

[0115] Based on the processed basic data set, the behavior type, timestamp, and spatial coordinates are formatted using standard communication protocols to generate data packets that conform to the specifications and determine structured transmission units.

[0116] For structured transmission units, data is distributed through communication interfaces between devices, timestamps and spatial coordinate changes during transmission are recorded, and the distributed data records are obtained.

[0117] If the distributed data records contain missing timestamps or spatial coordinates, the missing parts are filled in using interpolation methods to obtain a complete data sequence.

[0118] Based on the complete data sequence, the support vector machine model is applied to classify the behavior types. The accuracy of the behavior classification is judged by combining the contextual information of timestamps and spatial coordinates.

[0119] The categorized behavioral data are integrated to generate a unified biological behavior analysis output. Visualization tools are used to present the distribution of behavior types in time and space, resulting in the final analysis view.

[0120] For example, in constructing the initial dataset, this embodiment extracts fish behavior types, timestamps, and spatial coordinate data from underwater monitoring equipment. Suppose a monitoring session recorded fish swimming behavior with a timestamp of 2023-10-01 08:00:00 and spatial coordinates of X = 10.5, Y = 20.3, and Z = 5.2 meters. Using preset cleaning rules, data with abnormal timestamps or coordinates outside a reasonable range are removed. For instance, a record with a future timestamp is considered an outlier and removed. This ensures the reliability of the basic data set, laying the foundation for subsequent analysis.

[0121] For example, during the data formatting stage, for the processed basic data sets, standard communication protocols are used to unify the behavior type, timestamp, and spatial coordinates into JSON format data packets. Suppose a data packet contains the behavior type "fast movement," the timestamp "2023-10-01 08:01:00," and the spatial coordinates "X = 11.2, Y = 21.0, Z = 5.0 meters," a structured transmission unit is generated through formatting. This method facilitates data exchange between devices and ensures data consistency.

[0122] For example, in the data distribution process, data packets are transmitted to multiple analysis terminals via a communication interface, and changes in timestamps and spatial coordinates during transmission are recorded. Suppose that when a data packet is transmitted to terminal A, its timestamp is updated to "2023-10-01 08:02:00", while its coordinates remain unchanged. Recording this process facilitates subsequent tracing of the data flow path and improves the transparency of data management.

[0123] For example, if a missing timestamp is found in the distributed data records, such as a record lacking the time point "08:01:30", the missing value can be estimated based on the preceding and following timestamps using linear interpolation, and then filled in as "X=11.0, Y=20.8, Z=5.1 meters". This method can effectively maintain the continuity of the data sequence and provide a complete basis for subsequent classification.

[0124] For example, in the behavior classification stage, a support vector machine model is applied to classify behavior types. Combined with contextual information such as timestamps and spatial coordinates, it can be determined whether the fish behavior is "foraging" or "escape." Suppose a data sequence shows drastic changes in fish coordinates within a short period; combined with timestamp analysis indicating an active morning period, this can be inferred as "foraging" behavior. This contextual approach improves classification accuracy.

[0125] For example, the final integrated and categorized behavioral data generates a unified output for biological behavior analysis, and visualization tools are used to show the temporal and spatial distribution of behavioral types. Suppose a view shows that fish exhibit "foraging" behavior primarily between 8:00 and 9:00 AM, with spatial coordinates concentrated in the X = 10 to 15 meters range. This intuitive presentation helps researchers quickly grasp behavioral patterns and improves analytical efficiency.

[0126] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for analyzing the behavior of marine ranch organisms, characterized in that, include: Raw behavioral image data of marine organisms are collected, and deep learning algorithms are used to extract fine-grained features from the image sequences to obtain a quantitative description of the organisms' minute movements. Based on the fine-grained features, acoustic positioning points are set in a specific preset area and combined with visual recognition to perform point tracking of biological individuals, record behavioral change patterns and generate time series data. The behavior execution speed index is obtained from the time series data, and the numerical distribution of the execution speed index is calculated using a rate quantization algorithm to construct a correlation model between behavior rate and environmental conditions. The complex behavioral sequences in the association model are decomposed into independent units using a transaction processing mechanism. Each unit contains pattern recognition and anomaly detection operations, resulting in a processed behavioral transaction dataset. High-intensity activity features are extracted from the behavioral transaction dataset. The frequency and duration of the high-intensity activity features are analyzed using a high-energy consumption monitoring method. If the frequency exceeds a preset threshold, it is marked as a high-energy consumption state, and the energy consumption assessment result is obtained. Based on the energy consumption assessment results and combined with changes in biological morphology, an association analysis model is used to map the relationship between morphological features and behavioral states, resulting in an enhanced and accurate dataset for behavior recognition. The enhanced behavior recognition accurate dataset is encoded with behavior type, timestamp, and spatial coordinates. Data packets are formatted using communication protocol standards for seamless transmission and sharing between different devices, resulting in unified biological behavior analysis results.

2. The method according to claim 1, characterized in that, The process of obtaining a quantitative description of the minute movements of organisms includes: Raw image data is collected using underwater camera equipment to obtain image sequences containing data on marine biological behavior, thus obtaining preliminary visual records; The image sequence is processed using a pre-trained deep learning model to extract fine-grained features from each frame of the image, thereby obtaining the subtle details of the organism's movements. Based on the minute movement details of organisms, temporal analysis is performed on consecutive frame images to obtain the change patterns of swimming posture and to determine the dynamic performance of posture features. If an abnormal frame is detected in the swimming posture change pattern, a second feature extraction is performed on the abnormal frame to obtain motion description information; By analyzing the characteristic data related to mouth opening and closing, the frequency and amplitude of mouth movements are obtained, and quantitative indicators of mouth behavior are determined. Based on the quantitative indicators of swimming posture and mouth behavior, and combined with time-series data, a comprehensive comparison is made to determine whether the organism's behavior pattern conforms to the preset normal range. If the behavior pattern exceeds the preset range, the relevant image sequences are marked to obtain detailed feature records of the abnormal behavior.

3. The method according to claim 1, characterized in that, The process of recording behavioral patterns and generating time series data includes: Within a specific preset area, an area layout method is used to set acoustic positioning points and combine them with visual recognition technology to initially locate biological individuals, obtain spatial location information, and determine the initial tracking range. Based on the initial tracking range, a fixed-point tracking method is used to continuously monitor key organisms, obtain real-time location signals from acoustic positioning points, and extract fine-grained features by combining visual recognition techniques to obtain a preliminary record of behavioral change patterns. Based on the preliminary record of the behavioral change pattern, the movement path of key organisms is continuously collected by tracking and recording method to obtain path change data and determine whether the activity pattern conforms to the preset pattern. If the path change data deviates from the preset pattern, high-frequency signals are collected from the abnormal location through acoustic positioning points, and images of the abnormal behavior are captured by visual recognition technology to obtain behavior change information. Based on the behavioral change information, a time series data generation method is used to associate the behavioral change pattern with the time dimension to obtain continuous time series data and determine long-term behavioral trends. The time series data is segmented and processed using a data generation method. The behavioral change patterns of each time period are stored independently to obtain structured behavioral record data. If an abnormal trend is detected in the structured behavioral record data, a pre-established convolutional neural network model is used to perform in-depth analysis of the data during the abnormal time period to obtain potential behavioral change patterns and determine whether further attention is needed.

4. The method according to claim 1, characterized in that, The process of constructing a correlation model between behavioral rate and environmental conditions includes: Behavioral indicators are obtained from the time series data, and data processing tools are used to clean and format the raw data to obtain a preliminarily organized behavioral dataset. Based on the preliminary organized behavioral dataset, the rate quantization algorithm was applied to calculate the numerical distribution of swimming speed and feeding frequency in the execution speed index, and the range of change of each index in different time periods was determined. If the range of change exceeds a preset threshold, the numerical distribution is segmented to obtain the time points of abnormal fluctuations and the corresponding behavioral indicator values, and to determine whether there is a rate anomaly. By comparing the time points and behavioral indicator values ​​of abnormal fluctuations, data records related to environmental conditions are obtained, and the specific manifestations of environmental conditions at the time points of abnormal fluctuations are determined. Based on the specific manifestations of environmental conditions, a regression analysis algorithm is used to construct a correlation model between behavioral rate and environmental conditions.

5. The method according to claim 1, characterized in that, The process of obtaining the processed behavioral transaction dataset includes: The complex behavior is broken down into multiple sub-sequence segments using a segmentation technique, resulting in a preliminary set of decomposed sequences; Based on the preliminarily decomposed sequence set, a pre-established pattern recognition rule base is applied to each sub-sequence segment to determine whether there is a pattern that conforms to the rules, and to determine the pattern classification result of each segment. If the pattern classification result shows that a certain subsequence segment does not match any rule, it is marked as a potential abnormal segment, and the marked abnormal candidate set is obtained; The support vector machine algorithm is used to extract features and classify each potential abnormal segment in the marked abnormal candidate set to determine whether it is a real abnormality, thereby obtaining an abnormal confirmation set. Based on the aforementioned abnormal confirmation set, and in conjunction with the transaction processing mechanism, abnormal segments and normal segments are encapsulated into independent units to generate a structured set of behavioral transaction units. By integrating the behavioral transaction unit set and using a unified formatted storage method, a complete behavioral transaction dataset is constructed. Each cell in the behavioral transaction dataset is validated. If the validation finds missing data or inconsistent format, an automatic completion and correction process is triggered to obtain a complete and consistent behavioral transaction dataset.

6. The method according to claim 1, characterized in that, The process of obtaining energy consumption assessment results includes: Raw records are obtained from behavioral transaction data. Information related to high-intensity activities is filtered to separate activity segments that involve frequent swimming and rapid chasing, resulting in a preliminary classification of the data set. For the dataset of the preliminary classification, a time window segmentation method is used to calculate the frequency and duration of frequent swimming and fast chasing, and to determine the quantitative indicators of each activity segment; Based on the quantitative indicators, a comparison is made with a preset threshold. If the frequency or duration exceeds the preset threshold, the corresponding activity segment is marked as a high-energy-consuming state, and a set of marked activity states is obtained. Using the marked set of activity states, a high-energy-consumption monitoring method is employed to analyze the distribution of high-energy-consumption states in the overall data, and to obtain the proportion and temporal distribution characteristics of high-energy-consumption segments. Based on the proportion and temporal distribution characteristics of high-energy-consuming segments, the energy consumption is predicted using a support vector machine model to determine the energy consumption level for each time period. By combining the predicted energy consumption levels with the duration and frequency of activity segments, energy consumption assessment results for each activity segment are generated, and the final energy consumption distribution is determined. Based on the final energy consumption distribution, correlation analysis is performed on the activity segments in the high-energy-consuming state to obtain the behavioral patterns of frequent swimming and rapid chasing in the high-energy-consuming state, and to obtain the mapping relationship between behavior and energy consumption.

7. The method according to claim 1, characterized in that, The process of obtaining an enhanced, accurate dataset for behavior recognition includes: By collecting and organizing energy consumption data, key indicators are extracted from the original records to form a preliminary energy consumption feature set; Based on the image data of body color changes and fin spread, image processing techniques were used to separate the external features of the birth body to obtain a set of external feature descriptions; Based on the set of shape features and the set of energy consumption features, a correlation analysis model is constructed, a data mapping operation is performed, and the correspondence between shape features and energy consumption features is determined. If the correspondence output by the correlation analysis model is lower than the preset threshold, then a second feature extraction is performed on the image data of body color change and fin unfolding to obtain a more refined subset of shape features; Using the known labeled data of the aforementioned subset of external features and behavioral states, a support vector machine model is used for classification training to determine the category of the behavioral state. Based on the classification training results and combined with the logical rules of state analysis, an enhanced behavior recognition dataset is generated. If the recognition accuracy still does not meet the preset standard, the dataset generation process will be iteratively optimized, and the feature extraction parameters will be readjusted until an accurate dataset for behavior recognition that meets the requirements is obtained.

8. The method according to claim 1, characterized in that, The process of obtaining unified biological behavior analysis results includes: Based on the enhanced behavior recognition accurate dataset, behavior type, timestamp and spatial coordinates are obtained, an initial dataset is constructed, and outliers are removed using preset cleaning rules to obtain the processed basic data set; Based on the processed basic data set, the behavior type, timestamp, and spatial coordinates are formatted using standard communication protocols to generate data packets that conform to the specifications and determine structured transmission units. For the structured transmission unit, data is distributed through the communication interface between devices, timestamps and spatial coordinate changes during transmission are recorded, and the distributed data records are obtained. If the distributed data records contain missing timestamps or spatial coordinates, the missing parts are filled in using interpolation methods to obtain a complete data sequence. Based on the complete data sequence, a support vector machine model is applied to classify the behavior types, and the accuracy of the behavior classification is judged by combining the contextual information of timestamps and spatial coordinates. The categorized behavioral data are integrated to generate a unified biological behavior analysis output. Visualization tools are used to present the distribution of behavior types in time and space, resulting in the final analysis view.

Images