A model analysis-based behavior discrimination and culture method for non-molting macrobrachium rosenbergii

Abstract

The application discloses a model analysis-based behavior discrimination and breeding method for non-molting period macrobrachium rosenbergii, which comprises the following steps: setting a three-dimensional breeding model and monitoring points according to the breeding space and quantity; acquiring a breeding image set and water environment parameters through a monitoring unit, identifying a target group by using a yolov5 model, evaluating color feature differences, judging the molting period shrimp proportion, and setting a PIT marking scheme; analyzing target behavior characteristics, combining water environment parameters to generate characteristic data; introducing a self-encoder dimension reduction, dividing characteristics and cycle periods by using an SVN model binary classification; analyzing group activity probability according to period characteristic data, and setting path cost; finally, predicting molting period and non-molting period group activity migration by using a three-dimensional A* shortest path algorithm, and formulating a breeding isolation scheme, so that precise breeding management is realized.

Description

Technical Field

[0001] This invention relates to the field of giant freshwater prawn farming, and more specifically, to a model-based method for identifying and farming giant freshwater prawns during their non-molting period. Background Technology

[0002] In recent years, the giant freshwater prawn (Macrobrachium rosenbergii) has become an important economic shrimp species, and its aquaculture scale has continued to expand. However, the giant freshwater prawn's periodic molting biological characteristic poses a serious challenge to its aquaculture management. Molting is a critical stage in the giant freshwater prawn's life cycle. During this period, the individual's shell softens, its feeding ability weakens, and its activity level significantly decreases, making it extremely vulnerable to attacks and cannibalism from individuals not in the molting stage. This inter-individual fighting not only leads to reduced yields but also increases aquaculture costs, impacting aquaculture profitability.

[0003] Traditional giant freshwater prawn (Macrobrachium rosenbergii) farming management relies heavily on manual observation and experience, making it difficult to accurately grasp the behavioral patterns of individuals outside the molting period. Due to a lack of effective monitoring methods, farmers often struggle to analyze the behavioral characteristics of molting and non-molting populations, lack effective means to predict the behavioral patterns and activity trajectories of giant freshwater prawns outside the molting period, and fail to optimize regional farming programs, ultimately hindering the guarantee of farming yields and economic benefits. Summary of the Invention

[0004] This invention overcomes the shortcomings of existing technologies and proposes a model-based method for identifying and aquaculture the non-molting behavior of giant freshwater prawns.

[0005] The first aspect of this invention provides a model-based method for identifying and cultivating non-molting behavior of giant freshwater prawns, comprising:

[0006] S101: Based on the culture space and number of giant freshwater prawns, a three-dimensional culture model and multiple monitoring points are set up, with each monitoring point corresponding to an activity area;

[0007] S102: The monitoring unit acquires aquaculture image sets and water environment parameters during a preset time period. Based on the YOLOv5 model, the target group is identified in the aquaculture image set. Individual segmentation and color feature difference assessment are performed on the target group to determine the proportion of giant freshwater prawns in the molting period. A PIT tagging scheme is set for aquaculture monitoring.

[0008] S103: Based on PIT trajectory data and aquaculture image sets, periodically analyze target behavioral characteristics, and combine water environment parameters to generate behavioral characteristic data and environmental characteristic data;

[0009] S104: Set a fixed dimension M, introduce an autoencoder to reduce the dimensionality of behavioral feature data and environmental feature data, generate the first feature vector and the second feature vector for each period, fuse the two feature vectors as data points, perform binary classification on multiple data points through the SVN model, and divide all periods into the first time period and the second time period based on the classification results.

[0010] S105: Analyze the probability of group activity in each activity area based on the behavioral and environmental characteristic data corresponding to the first time period, and set the first path cost. Set the second path cost based on the second time period.

[0011] S106: Based on the first path cost and the second path cost of each activity area, the activity migration of the population during the molting and non-molting periods is predicted in the aquaculture model using the three-dimensional A* shortest path algorithm, and the aquaculture isolation scheme is formulated based on the prediction results.

[0012] In this solution, S101 specifically refers to:

[0013] The information on the size, area, depth, and water quality of the giant freshwater prawn farming area includes the following:

[0014] The number of farmed prawns includes the number of prawns at different growth stages;

[0015] A three-dimensional visualization-based aquaculture model is set up in the aquaculture area, and multiple monitoring points are divided in the aquaculture model based on the number of aquacultures.

[0016] In this solution, S102 specifically refers to:

[0017] The monitoring unit collects images and water environment parameters in each activity area to obtain a set of aquaculture images and water environment parameters.

[0018] Image denoising, enhancement, and standardization preprocessing were performed on the aquaculture image set;

[0019] Based on the YOLOv5 model, target group identification is performed on the aquaculture image set, and individual target identification and segmentation are performed in the target group to obtain individual images;

[0020] Color features of individual images are extracted using color histograms, the difference between color features and contrast features is calculated using Euclidean distance, and it is determined whether an individual image is in the molting stage.

[0021] By analyzing individual images, a comprehensive statistical analysis of the target group is performed to calculate the proportion of giant freshwater prawns in the molting stage.

[0022] If the proportion of giant freshwater prawns in the molting stage is greater than the preset proportion and the water environment parameters are under ideal conditions during the molting stage, then the PIT tagging scheme is set to capture and tag the prawns.

[0023] In this solution, S103 specifically refers to:

[0024] Based on PIT trajectory data, multi-regional behavioral analysis of the prawn population was conducted, and target population identification and secondary behavioral analysis were performed in combination with aquaculture image set to obtain behavioral characteristic data for each cycle.

[0025] Behavioral characteristics data include the swimming distance, swimming speed, regional shrimp population density changes, swimming range, and diurnal behavioral differences of the prawn population;

[0026] Water quality is monitored at monitoring points using monitoring units, generating environmental characteristic data for each cycle.

[0027] In this solution, S104 specifically refers to:

[0028] By setting a fixed dimension M, an autoencoder is introduced to vectorize behavioral feature data and environmental feature data and map them to a low-dimensional hidden space representation.

[0029] The low-dimensional hidden space is reconstructed using a multilayer perceptron, and the M-dimensional data is output to obtain the reconstructed feature data.

[0030] The difference between the feature data before and after reconstruction is evaluated by using cross-entropy loss and the reconstructed data is optimized. Finally, the dimensionality-reduced behavioral feature data and environmental feature data are obtained and labeled as the first feature vector and the second feature vector, respectively.

[0031] In this solution, S104 further includes:

[0032] The analysis target is a single cycle.

[0033] The first feature vector and the second feature vector are concatenated and fused to obtain the fused feature vector;

[0034] Generate a corresponding fusion feature vector for each cycle;

[0035] In the SVM model, a radial basis function kernel is set to construct a data space. The fused feature vector is used as the data point. The hyperparameters are optimized by grid search and the data is classified into two groups. The optimal hyperplane is searched to segment the data points. The classification results of the data points are mapped to each period to obtain two groups of periods. The two groups of period time periods are labeled as the first time period and the second time period, respectively.

[0036] In this solution, S105 specifically refers to:

[0037] Based on the behavioral characteristic data corresponding to the first time period, calculate the activity frequency of the target group in each activity area;

[0038] The environmental characteristic data corresponding to the first time period is compared with the expected environmental data to assess the environmental deviation;

[0039] Assess the probability of group activity based on activity frequency and environmental deviations, and set the first path cost for each activity area;

[0040] The second time period is analyzed, and the second path cost for each activity area is set.

[0041] In this solution, S106 specifically refers to:

[0042] The activity area with an environmental deviation less than the preset deviation is marked and used as the activity path point;

[0043] The path search order is generated by taking the activity area of ​​the target group in the most recent period as the starting point and combining the activity waypoints.

[0044] In the aquaculture model, a first path cost is set, and the shortest path is searched based on the path search order using the three-dimensional A* shortest path algorithm. The path search results are used as the first group migration prediction path.

[0045] The shortest path is searched based on the cost of the second path, and the predicted migration path of the second group is obtained.

[0046] By predicting the path and setting up isolation measures for aquaculture in the first and second time periods, an aquaculture isolation plan is generated.

[0047] A second aspect of the present invention also provides a model-based system for identifying and aquaculture non-molting behavior of giant freshwater prawns. The system includes a memory, a processor, and a data interface. The memory includes a model-based program for identifying and aquaculture non-molting behavior of giant freshwater prawns. When executed by the processor, the model-based program performs the following steps:

[0048] S101: Based on the culture space and number of giant freshwater prawns, a three-dimensional culture model and multiple monitoring points are set up, with each monitoring point corresponding to an activity area;

[0049] S102: The monitoring unit acquires aquaculture image sets and water environment parameters during a preset time period. Based on the YOLOv5 model, the target group is identified in the aquaculture image set. Individual segmentation and color feature difference assessment are performed on the target group to determine the proportion of giant freshwater prawns in the molting period. A PIT tagging scheme is set for aquaculture monitoring.

[0050] S103: Based on PIT trajectory data and aquaculture image sets, periodically analyze target behavioral characteristics, and combine water environment parameters to generate behavioral characteristic data and environmental characteristic data;

[0051] S104: Set a fixed dimension M, introduce an autoencoder to reduce the dimensionality of behavioral feature data and environmental feature data, generate the first feature vector and the second feature vector for each period, fuse the two feature vectors as data points, perform binary classification on multiple data points through the SVN model, and divide all periods into the first time period and the second time period based on the classification results.

[0052] S105: Analyze the probability of group activity in each activity area based on the behavioral and environmental characteristic data corresponding to the first time period, and set the first path cost. Set the second path cost based on the second time period.

[0053] S106: Based on the first path cost and the second path cost of each activity area, the activity migration of the population during the molting and non-molting periods is predicted in the aquaculture model using the three-dimensional A* shortest path algorithm, and the aquaculture isolation scheme is formulated based on the prediction results.

[0054] A third aspect of the present invention also provides a computer-readable storage medium comprising a model-based analysis program for identifying and aquaculture non-molting behavior of giant freshwater prawns. When the model-based analysis program for identifying and aquaculture non-molting behavior of giant freshwater prawns is executed by a processor, it implements the steps of the model-based analysis method for identifying and aquaculture non-molting behavior of giant freshwater prawns as described in any of the preceding claims.

[0055] This invention discloses a model-based method for identifying and managing the behavior of giant freshwater prawns during the non-molting period. The method includes: setting up a three-dimensional aquaculture model and monitoring points based on the aquaculture space and quantity; acquiring aquaculture image sets and aquatic environment parameters through monitoring units; identifying target groups using a YOLOv5 model; assessing color feature differences; determining the proportion of molting prawns; and setting a PIT (Picking-In-the-Dots) labeling scheme; analyzing target behavioral characteristics and generating feature data based on aquatic environment parameters; introducing an autoencoder for dimensionality reduction; using an SVN (Single-Segmented Tree Nymph) model for binary classification to divide features and periodic time periods; analyzing the probability of group activity based on time period feature data and setting path costs; and finally, using a three-dimensional A* shortest path algorithm to predict the migration of group activities during and outside the molting period, and developing an aquaculture isolation plan to achieve precise aquaculture management. Attached Figure Description

[0056] Figure 1 The flowchart of a model-based method for identifying and aquaculture non-molting behavior of giant freshwater prawns according to the present invention is shown.

[0057] Figure 2 A simplified flowchart of the path prediction process of this invention is shown;

[0058] Figure 3 A block diagram of a model-based system for identifying and aquaculture behavior of giant freshwater prawns during the non-molting period is shown. Detailed Implementation

[0059] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0060] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0061] Figure 1 The flowchart illustrates a model-based method for identifying and aquaculture non-molting behavior of giant freshwater prawns according to the present invention.

[0062] like Figure 1 As shown, the first aspect of this invention provides a model-based method for identifying and cultivating non-molting behavior of giant freshwater prawns, comprising:

[0063] S101: Based on the culture space and number of giant freshwater prawns, a three-dimensional culture model and multiple monitoring points are set up, with each monitoring point corresponding to an activity area;

[0064] S102: The monitoring unit acquires aquaculture image sets and water environment parameters during a preset time period. Based on the YOLOv5 model, the target group is identified in the aquaculture image set. Individual segmentation and color feature difference assessment are performed on the target group to determine the proportion of giant freshwater prawns in the molting period. A PIT tagging scheme is set for aquaculture monitoring.

[0065] S103: Based on PIT trajectory data and aquaculture image sets, periodically analyze target behavioral characteristics, and combine water environment parameters to generate behavioral characteristic data and environmental characteristic data;

[0066] S104: Set a fixed dimension M, introduce an autoencoder to reduce the dimensionality of behavioral feature data and environmental feature data, generate the first feature vector and the second feature vector for each period, fuse the two feature vectors as data points, perform binary classification on multiple data points through the SVN model, and divide all periods into the first time period and the second time period based on the classification results.

[0067] S105: Analyze the probability of group activity in each activity area based on the behavioral and environmental characteristic data corresponding to the first time period, and set the first path cost. Set the second path cost based on the second time period.

[0068] S106: Based on the first path cost and the second path cost of each activity area, the activity migration of the population during the molting and non-molting periods is predicted in the aquaculture model using the three-dimensional A* shortest path algorithm, and the aquaculture isolation scheme is formulated based on the prediction results.

[0069] According to an embodiment of the present invention, S101 specifically includes:

[0070] The information on the size, area, depth, and water quality of the giant freshwater prawn farming area includes the following:

[0071] The number of farmed prawns includes the number of prawns at different growth stages;

[0072] A three-dimensional visualization-based aquaculture model is set up in the aquaculture area, and multiple monitoring points are divided in the aquaculture model based on the number of aquacultures.

[0073] It should be noted that the aquaculture model is a map model used for visual analysis of shrimp population migration and distribution. Each monitoring point includes at least one monitoring unit. The monitoring unit includes an underwater camera, a water quality parameter acquisition device, and a PIT (Potential Intake and Testing) device for information acquisition.

[0074] According to an embodiment of the present invention, step S102 specifically includes:

[0075] The monitoring unit collects images and water environment parameters in each activity area to obtain a set of aquaculture images and water environment parameters.

[0076] Image denoising, enhancement, and standardization preprocessing were performed on the aquaculture image set;

[0077] Based on the YOLOv5 model, target group identification is performed on the aquaculture image set, and individual target identification and segmentation are performed in the target group to obtain individual images;

[0078] Color features of individual images are extracted using color histograms, the difference between color features and contrast features is calculated using Euclidean distance, and it is determined whether an individual image is in the molting stage.

[0079] By analyzing individual images, a comprehensive statistical analysis of the target group is performed to calculate the proportion of giant freshwater prawns in the molting stage.

[0080] If the proportion of giant freshwater prawns in the molting stage is greater than the preset proportion and the water environment parameters are under ideal conditions during the molting stage, then the PIT tagging scheme is set to capture and tag the prawns.

[0081] It should be noted that the YOLOv5 model can be trained for target recognition and color feature training of giant freshwater prawns using a preset image library. Since there are significant differences in color features between prawns in the molting and non-molting stages, this invention uses histograms to represent color features for statistical analysis of the two groups: the molting group and the non-molting group. Image frames are introduced for approximate statistics, and subsequent PIT individual labeling is used for precise activity statistics of the first and second groups. Color feature analysis can quickly determine whether an individual is in the molting stage, and comprehensive statistics are performed based on the target group. The PIT labeling scheme generally sets the same number of individuals for the first and second groups. Ideally, the water environment parameters during the molting stage should be set to a water temperature maintained between 26 and 28°C, which allows the prawns to molt successfully.

[0082] The contrast features are generally idealized color features of the first or second group, used to determine whether an individual is in the molting stage.

[0083] According to an embodiment of the present invention, S103 specifically includes:

[0084] Based on PIT trajectory data, multi-regional behavioral analysis of the prawn population was conducted, and target population identification and secondary behavioral analysis were performed in combination with aquaculture image set to obtain behavioral characteristic data for each cycle.

[0085] Behavioral characteristics data include the swimming distance, swimming speed, regional shrimp population density changes, swimming range, and diurnal behavioral differences of the prawn population;

[0086] Water quality is monitored at monitoring points using monitoring units, generating environmental characteristic data for each cycle.

[0087] In this embodiment, the PIT trajectory is used to determine the movement trajectory of the mixed prawn population. Since the prawns in the molting stage and those in the non-molting stage have similar trajectories, it is necessary to use PIT to mark the first and second prawn populations in the same proportion for trajectory discrimination, so as to perform binary classification based on the SVN model in the subsequent process. The image analysis method is used to provide certain trajectory and behavioral feature references.

[0088] Behavioral characteristic data includes parameters such as swimming distance, swimming speed, regional shrimp population density changes, swimming range, and diurnal behavioral differences of the prawn population. Environmental characteristic data includes parameters such as dissolved oxygen, chemical oxygen demand, nutrient concentration, pH, and water temperature, all of which can be converted into feature vectors.

[0089] The aquaculture image set can obtain information such as swimming distance, swimming speed, and changes in shrimp population density per unit time; PIT trajectory data can obtain information such as changes in shrimp population density, swimming range, and diurnal behavioral differences. Diurnal behavioral differences are calculated by analyzing the differences in activity range or distance through diurnal activity trajectories. Certain behavioral characteristic parameters can be obtained through analysis of PIT trajectory data and the aquaculture image set.

[0090] The swimming range can be set by analyzing the total area of ​​the swimming region or the maximum straight-line distance of the region.

[0091] According to an embodiment of the present invention, step S104 specifically includes:

[0092] By setting a fixed dimension M, an autoencoder is introduced to vectorize behavioral feature data and environmental feature data and map them to a low-dimensional hidden space representation.

[0093] The low-dimensional hidden space is reconstructed using a multilayer perceptron, and the M-dimensional data is output to obtain the reconstructed feature data.

[0094] The difference between the feature data before and after reconstruction is evaluated by using cross-entropy loss and the reconstructed data is optimized. Finally, the dimensionality-reduced behavioral feature data and environmental feature data are obtained and labeled as the first feature vector and the second feature vector, respectively.

[0095] It should be noted that in prawn farming, behavioral and environmental parameters generally involve multidimensional data. Furthermore, in the process of co-culturing prawns during and outside of the molting stage, existing technologies struggle to classify prawn behavior patterns. Combining environmental parameters with multidimensional analysis often leads to complex models, inconsistent results, and uncontrollable biases in predicting prawn behavior. Therefore, this invention introduces an autoencoder to reduce the dimensionality of the multidimensional feature data. This serves two purposes: firstly, it adapts to the low-dimensional data partitioning of the SVN model, improving the efficiency of data point stratification and classification; secondly, it simplifies data representation and allows for the extraction of effective behavioral patterns and environmental parameters from the multidimensional feature data for adaptability and behavioral model evaluation of the first and second populations. This enables the segmentation and isolation of the first and second populations (prawns during and outside of the molting stage). M can be set to 3-10.

[0096] Each period has a corresponding first eigenvector and a second eigenvector.

[0097] According to an embodiment of the present invention, step S104 further includes:

[0098] The analysis target is a single cycle.

[0099] The first feature vector and the second feature vector are concatenated and fused to obtain the fused feature vector;

[0100] Generate a corresponding fusion feature vector for each cycle;

[0101] In the SVM model, a radial basis function kernel is set to construct a data space. The fused feature vector is used as the data point. The hyperparameters are optimized by grid search and the data is classified into two groups. The optimal hyperplane is searched to segment the data points. The classification results of the data points are mapped to each period to obtain two groups of periods. The two groups of period time periods are labeled as the first time period and the second time period, respectively.

[0102] In this embodiment, an SVN model is introduced to perform binary classification of multi-period data points on behavioral and environmental characteristics, thereby identifying which time periods show higher activity levels for the first and second groups. The first and second time periods correspond to the active periods of the molting and non-molting groups, respectively. After dividing the time periods, a specific time period can be analyzed, combined with PIT trajectory data, to determine the proportion of the corresponding group in the molting period and label it as either the first or second time period. The radial basis function is the RBF kernel.

[0103] In existing technologies, it is difficult to effectively predict and accurately classify the activities of individuals or groups during the molting and non-molting stages in the polyculture process of giant freshwater prawns in complex environments. Furthermore, there are certain differences in the behavioral patterns of giant freshwater prawns during the molting and non-molting stages. This invention, however, integrates behavioral and environmental characteristics and uses the precise binary classification method of SVN (Support Vector Machine) to classify the integrated characteristics. It effectively predicts and classifies two activity patterns from existing cycles, corresponding to the activity cycles of the first and second groups, respectively. Based on the classification results, a first and a second time period are divided. In the first time period, the behavior pattern of the identified target group is biased towards the molting stage giant freshwater prawn group, with the molting stage giant freshwater prawn group accounting for a larger proportion. In the second time period, the behavior pattern of the identified target group is biased towards the non-molting stage giant freshwater prawn group, with the non-molting stage giant freshwater prawn group accounting for a larger proportion.

[0104] According to an embodiment of the present invention, step S105 specifically includes:

[0105] Based on the behavioral characteristic data corresponding to the first time period, calculate the activity frequency of the target group in each activity area;

[0106] The environmental characteristic data corresponding to the first time period is compared with the expected environmental data to assess the environmental deviation;

[0107] Assess the probability of group activity based on activity frequency and environmental deviations, and set the first path cost for each activity area;

[0108] The second time period is analyzed, and the second path cost for each activity area is set.

[0109] It should be noted that the environmental deviation mentioned is a weighted average of deviations from various environmental parameters, used to reflect the suitability of the aquatic environment for shrimp farming. Expected environmental data includes optimal environmental parameters for both molting and non-molting periods. Higher activity frequency and smaller environmental deviation result in lower path costs. Activity frequency and environmental deviation effectively reflect the probability of group activity.

[0110] The first path cost and the second path cost are two different cost models, but the analysis process is consistent and used for path prediction between the first group and the second group.

[0111] According to an embodiment of the present invention, step S106 specifically includes:

[0112] The activity area with an environmental deviation less than the preset deviation is marked and used as the activity path point;

[0113] The path search order is generated by taking the activity area of ​​the target group in the most recent period as the starting point and combining the activity waypoints.

[0114] In the aquaculture model, a first path cost is set, and the shortest path is searched based on the path search order using the three-dimensional A* shortest path algorithm. The path search results are used as the first group migration prediction path.

[0115] The shortest path is searched based on the cost of the second path, and the predicted migration path of the second group is obtained.

[0116] By predicting the path and setting up isolation measures for aquaculture in the first and second time periods, an aquaculture isolation plan is generated.

[0117] It is understandable here that the location of the target group can be obtained through PIT trajectory data or image set identification. The shortest path search order of activity waypoints can be sorted from smallest to largest based on environmental deviation. In the migration prediction process, the analysis methods for the first and second groups are the same, but the activity waypoints, path search order, and path costs differ.

[0118] According to an embodiment of the present invention, the path search order further includes:

[0119] In the first or second time period, the activity frequency and environmental deviation of each activity area are sorted and linear fitting is performed respectively. Based on the fitting results, two goodness-of-fit values ​​based on activity frequency and environmental deviation are obtained.

[0120] Using goodness of fit as the weight, for each activity area, the activity frequency and environmental deviation are weighted and averaged to obtain the area priority;

[0121] The path search order is set according to the region priority.

[0122] In this embodiment, linear fitting can be achieved using a univariate linear equation: y = kx + b. The weighted average process of activity frequency and environmental deviation is specifically the sum of the activity frequency multiplied by its corresponding weight and the reciprocal of the environmental deviation multiplied by its corresponding weight. In the analysis of the goodness of fit between activity frequency and environmental deviation, a higher goodness of fit indicates a greater reference significance of the corresponding parameter in the activity pattern. Therefore, setting appropriate weights for regional priority settings can search for prediction paths that more closely approximate the actual situation.

[0123] Figure 2 A simplified flowchart of the path prediction process of the present invention is shown.

[0124] Figure 3 A block diagram of a model-based system for identifying and aquaculture behavior of giant freshwater prawns during the non-molting period is shown.

[0125] A second aspect of the present invention also provides a model-based system for identifying and aquaculture non-molting behavior of giant freshwater prawns. This system includes a memory 202, a processor 201, and a data interface 203. The data interface is used to send monitoring data, aquaculture data, and prediction results to a user terminal. The memory includes a model-based program for identifying and aquaculture non-molting behavior of giant freshwater prawns. When executed by the processor, the model-based program for identifying and aquaculture non-molting behavior of giant freshwater prawns performs the following steps:

[0126] S101: Based on the culture space and number of giant freshwater prawns, a three-dimensional culture model and multiple monitoring points are set up, with each monitoring point corresponding to an activity area;

[0127] S102: The monitoring unit acquires aquaculture image sets and water environment parameters during a preset time period. Based on the YOLOv5 model, the target group is identified in the aquaculture image set. Individual segmentation and color feature difference assessment are performed on the target group to determine the proportion of giant freshwater prawns in the molting period. A PIT tagging scheme is set for aquaculture monitoring.

[0128] S103: Based on PIT trajectory data and aquaculture image sets, periodically analyze target behavioral characteristics, and combine water environment parameters to generate behavioral characteristic data and environmental characteristic data;

[0129] S104: Set a fixed dimension M, introduce an autoencoder to reduce the dimensionality of behavioral feature data and environmental feature data, generate the first feature vector and the second feature vector for each period, fuse the two feature vectors as data points, perform binary classification on multiple data points through the SVN model, and divide all periods into the first time period and the second time period based on the classification results.

[0130] S105: Analyze the probability of group activity in each activity area based on the behavioral and environmental characteristic data corresponding to the first time period, and set the first path cost. Set the second path cost based on the second time period.

[0131] S106: Based on the first path cost and the second path cost of each activity area, the activity migration of the population during the molting and non-molting periods is predicted in the aquaculture model using the three-dimensional A* shortest path algorithm, and the aquaculture isolation scheme is formulated based on the prediction results.

[0132] When the system is running, it can implement the steps of model analysis-based non-molting behavior identification and aquaculture method of giant freshwater prawn as described in any of the above items.

[0133] A third aspect of the present invention also provides a computer-readable storage medium comprising a model-based analysis program for identifying and aquaculture non-molting behavior of giant freshwater prawns. When the model-based analysis program for identifying and aquaculture non-molting behavior of giant freshwater prawns is executed by a processor, it implements the steps of the model-based analysis method for identifying and aquaculture non-molting behavior of giant freshwater prawns as described in any of the preceding claims.

[0134] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, the processes or functions described in the embodiments of this application can be generated, in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic cable, data subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital universal optical disc), or a semiconductor medium (e.g., solid-state drive). In the various embodiments of this application, unless otherwise specified or logically conflicting, the terminology and / or descriptions between different embodiments are consistent and can be mutually referenced. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0135] In the embodiments of this application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone, where A and B can be singular or plural. In the textual description of the embodiments of this application, the character " / " generally indicates that the preceding and following related objects have an "or" relationship. In this application, "first," "second," and various numerical designations are only for the convenience of description and are not used to limit the scope of the embodiments of this application. For example, they are used to distinguish different messages, rather than to describe a specific order or sequence.

[0136] It is understood that the various numerical designations used in the embodiments of this application are merely for descriptive convenience and are not intended to limit the scope of the embodiments of this application. The order of the process numbers does not imply the order of execution; the execution order of each process should be determined by its function and internal logic.

[0137] Finally, it should be noted that the above description is only a specific implementation of this application, but the protection scope of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be covered within the protection scope of this application.

Claims

1. A model-based method for identifying and culturing the non-molting behavior of giant freshwater prawns, characterized in that, include: S101: Based on the culture space and number of giant freshwater prawns, a three-dimensional culture model and multiple monitoring points are set up, with each monitoring point corresponding to an activity area; S102: The monitoring unit acquires aquaculture image sets and water environment parameters during a preset time period. Based on the YOLOv5 model, the target group is identified in the aquaculture image set. Individual segmentation and color feature difference assessment are performed on the target group to determine the proportion of giant freshwater prawns in the molting period. A PIT tagging scheme is set for aquaculture monitoring. S103: Based on PIT trajectory data and aquaculture image sets, periodically analyze target behavioral characteristics, and combine water environment parameters to generate behavioral characteristic data and environmental characteristic data; S104: Set a fixed dimension M, introduce an autoencoder to reduce the dimensionality of behavioral feature data and environmental feature data, generate the first feature vector and the second feature vector for each period, fuse the two feature vectors as data points, perform binary classification on multiple data points through an SVM model, and divide all periods into the first time period and the second time period based on the classification results. S105: Analyze the probability of group activity in each activity area based on the behavioral and environmental characteristic data corresponding to the first time period, and set the first path cost. Set the second path cost based on the second time period. S106: Based on the first path cost and the second path cost of each activity area, the three-dimensional A* shortest path algorithm is used to predict the activity migration of the population during the molting and non-molting periods in the breeding model, and the breeding isolation scheme is based on the prediction results. Specifically, S105 is: Based on the behavioral characteristic data corresponding to the first time period, calculate the activity frequency of the target group in each activity area; The environmental characteristic data corresponding to the first time period is compared with the expected environmental data to assess the environmental deviation; Assess the probability of group activity based on activity frequency and environmental deviations, and set the first path cost for each activity area; Analyze the second time period and set the second path cost for each activity area; Specifically, S106 is as follows: The activity area with an environmental deviation less than the preset deviation is marked and used as the activity path point; The path search order is generated by taking the activity area of ​​the target group in the most recent period as the starting point and combining the activity waypoints. In the aquaculture model, a first path cost is set, and the shortest path is searched based on the path search order using the three-dimensional A* shortest path algorithm. The path search results are used as the first group migration prediction path. The shortest path is searched based on the cost of the second path, and the predicted migration path of the second group is obtained. By predicting the path and setting up isolation measures for aquaculture in the first and second time periods, an aquaculture isolation plan is generated.

2. The method for identifying and cultivating non-molting behavior of giant freshwater prawns based on model analysis according to claim 1, characterized in that, Specifically, S101 is as follows: The information on the size, area, depth, and water quality of the giant freshwater prawn farming area includes the following: The number of farmed prawns includes the number of prawns at different growth stages; A three-dimensional visualization-based aquaculture model is set up in the aquaculture area, and multiple monitoring points are divided in the aquaculture model based on the number of aquacultures.

3. The method for identifying and cultivating non-molting behavior of giant freshwater prawns based on model analysis according to claim 1, characterized in that, Specifically, S102 is as follows: The monitoring unit collects images and water environment parameters in each activity area to obtain a set of aquaculture images and water environment parameters. Image denoising, enhancement, and standardization preprocessing were performed on the aquaculture image set; Based on the YOLOv5 model, target group identification is performed on the aquaculture image set, and individual target identification and segmentation are performed in the target group to obtain individual images; Color features of individual images are extracted using color histograms, the difference between color features and contrast features is calculated using Euclidean distance, and it is determined whether an individual image is in the molting stage. By analyzing individual images, a comprehensive statistical analysis of the target group is performed to calculate the proportion of giant freshwater prawns in the molting stage. If the proportion of giant freshwater prawns in the molting stage is greater than the preset proportion and the water environment parameters are under ideal conditions during the molting stage, then the PIT tagging scheme is set to capture and tag the prawns.

4. The method for identifying and cultivating non-molting behavior of giant freshwater prawns based on model analysis according to claim 1, characterized in that, Specifically, S103 is: Based on PIT trajectory data, multi-regional behavioral analysis of the prawn population was conducted, and target population identification and secondary behavioral analysis were performed in combination with aquaculture image set to obtain behavioral characteristic data for each cycle. Behavioral characteristics data include the swimming distance, swimming speed, regional shrimp population density changes, swimming range, and diurnal behavioral differences of the prawn population; Water quality is monitored at monitoring points using monitoring units, generating environmental characteristic data for each cycle.

5. The method for identifying and cultivating non-molting behavior of giant freshwater prawns based on model analysis according to claim 1, characterized in that, Specifically, S104 is: By setting a fixed dimension M, an autoencoder is introduced to vectorize behavioral feature data and environmental feature data and map them to a low-dimensional hidden space representation. The low-dimensional hidden space is reconstructed using a multilayer perceptron, and the M-dimensional data is output to obtain the reconstructed feature data. The difference between the feature data before and after reconstruction is evaluated by using cross-entropy loss and the reconstructed data is optimized. Finally, the dimensionality-reduced behavioral feature data and environmental feature data are obtained and labeled as the first feature vector and the second feature vector, respectively.

6. The method for identifying and cultivating non-molting behavior of giant freshwater prawns based on model analysis according to claim 1, characterized in that, S104 further includes: The analysis target is a single cycle. The first feature vector and the second feature vector are concatenated and fused to obtain the fused feature vector; Generate a corresponding fusion feature vector for each cycle; In the SVM model, a radial basis function kernel is set to construct a data space. The fused feature vector is used as the data point. The hyperparameters are optimized by grid search and the data is classified into two groups. The optimal hyperplane is searched to segment the data points. The classification results of the data points are mapped to each period to obtain two groups of periods. The two groups of period time periods are labeled as the first time period and the second time period, respectively.

7. A model-based system for identifying and aquaculture behavior of giant freshwater prawns during the non-molting period, characterized in that, The system includes: a memory, a processor, and a data interface. The memory includes a model-based analysis program for identifying and cultivating non-molting behavior of giant freshwater prawns. When the processor executes the model-based analysis program for identifying and cultivating non-molting behavior of giant freshwater prawns, it implements the steps of the model-based analysis method for identifying and cultivating non-molting behavior of giant freshwater prawns as described in claim 1.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a model-based analysis program for identifying and aquaculture non-molting behavior of giant freshwater prawns. When the model-based analysis program for identifying and aquaculture non-molting behavior of giant freshwater prawns is executed by a processor, it implements the steps of the model-based analysis method for identifying and aquaculture non-molting behavior of giant freshwater prawns as described in any one of claims 1 to 6.

Images