A method and system for monitoring the breeding status of black-boned chickens

Abstract

This invention provides a method and system for monitoring the status of black-boned chicken farming, relating to the field of intelligent farming monitoring technology. The method includes: acquiring video monitoring data of black-boned chicken flocks in a farm using a camera device; identifying multiple monitoring areas and extracting image sets for each monitoring area to be analyzed; performing background modeling on the multiple monitoring areas based on the image sets to be analyzed, constructing a dynamic background model for each monitoring area and performing status detection to build a status feature set for each monitoring area; extracting a status sequence matrix of the monitoring areas from the status feature set and performing feature filtering to generate candidate status evolution matrices; extracting the behavior intensity change matrix and behavior rhythm change matrix for each monitoring area and performing black-boned chicken flock status anomaly detection for the monitoring areas, generating status monitoring results for different monitoring areas. This invention improves the accuracy of black-boned chicken flock status monitoring.

Description

Technical Field

[0001] This invention relates to the field of intelligent aquaculture monitoring technology, and in particular to a method and system for monitoring the status of black-boned chicken farming. Background Technology

[0002] In large-scale black-boned chicken farms, video surveillance is gradually becoming an important tool for managing the breeding status due to its wide coverage, low deployment cost, and traceability. Compared to methods relying on manual inspection or individual identification, group-level status monitoring can provide continuous data without interrupting production, thus having significant application value.

[0003] Common methods for monitoring flocks often use image density, regional activity levels, or heatmap distribution to characterize the spatial aggregation and overall activity level of chickens. These methods are relatively intuitive in perceiving overall trends. However, if the analysis only focuses on static statistics or single time windows without considering the behavioral processes within the farming environment, it can lead to misunderstandings. For example, an increase in density in the image could indicate normal concentrated feeding or overcrowding; the two scenarios may appear similar numerically, but their implications for farming management are entirely different. Similarly, a decrease in occupancy near the watering area could stem from orderly flock turnover or be related to abnormal watering facilities. If a comprehensive analysis of flock behavior is not conducted, taking into account both spatial distribution and temporal changes, it may be difficult to accurately distinguish some key states, resulting in lower accuracy in assessing the condition of the chicken flock in the farm. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a method and system for monitoring the status of black-boned chicken farming. By extracting and analyzing behavioral characteristics that reflect the group characteristics of black-boned chicken flocks, such as orderliness and crowding, from farming monitoring video data, this invention achieves status monitoring that is closer to actual farming management, thereby improving the accuracy of monitoring the status of black-boned chicken flocks.

[0005] The first aspect of this invention provides a method for monitoring the breeding status of black-boned chickens, comprising:

[0006] The status monitoring video of the black-bone chicken flock in the farm is collected by the camera device. Multiple breeding monitoring areas are determined according to the facility configuration data of the farm. The image set to be analyzed for each breeding monitoring area is extracted from the status monitoring video.

[0007] Background modeling is performed on multiple aquaculture monitoring areas based on the image set to be analyzed, and a dynamic background model is constructed for each aquaculture monitoring area. Based on the dynamic background model, state detection is performed on the image set to be analyzed, and a state feature set for each aquaculture monitoring area is constructed.

[0008] The state sequence matrix of the aquaculture monitoring area is extracted from the state feature set, including the state evolution sequence of multiple grid areas. The state sequence matrix is ​​then subjected to feature screening to generate candidate state evolution matrices.

[0009] Multiple candidate state evolution matrices are decomposed to extract the behavior intensity change matrix and behavior rhythm change matrix for each aquaculture monitoring area. Based on the behavior intensity change matrix and behavior rhythm change matrix, the state anomaly detection of black-bone chicken flocks in the aquaculture monitoring area is performed, and state monitoring results for different aquaculture monitoring areas are generated.

[0010] Preferably, background modeling is performed on multiple aquaculture monitoring areas based on the image set to be analyzed, and a dynamic background model is constructed for each aquaculture monitoring area, including:

[0011] Temporal difference detection is performed on the image set to be analyzed, including calculating the global difference energy parameter between any adjacent frame images, and constructing multiple candidate background images of the image set to be analyzed based on the global difference energy parameter;

[0012] Multiple candidate background images are processed by sliding window to construct multiple window background images of the image set to be analyzed. The initial background in the multiple window background images is determined and updated by neighbor diffusion, including local coverage detection of the window background images adjacent to the initial background, calculating the local coverage parameters of each grid region in the window background image, locally cutting the window background image and updating it to the initial background, generating multiple target background images after traversing multiple window background images, and constructing a dynamic background model of the aquaculture monitoring area based on the multiple target background images.

[0013] Preferably, the state sequence matrix of the aquaculture monitoring area is extracted from the state feature set, and the state sequence matrix is ​​subjected to feature screening to generate a candidate state evolution matrix, including:

[0014] For the construction of the state feature set, the multiple video frame images covered by each target background image in the dynamic background model are determined. Foreground detection is performed on each video frame image based on the multiple target background images to determine the foreground region in each video frame image. Multiple state feature images of the aquaculture monitoring area are generated and the state feature set is constructed.

[0015] The aquaculture monitoring area is divided into multiple grid areas. Local state features corresponding to multiple grid areas in each state feature map are extracted. A state evolution sequence for each grid area is constructed based on multiple local state features. A state sequence matrix for the aquaculture monitoring area is generated based on multiple state evolution sequences. The state sequence matrix is ​​then filtered to generate a candidate state evolution matrix.

[0016] Preferably, the detection of abnormal states of black-boned chicken flocks in the breeding monitoring area is based on the behavioral intensity change matrix and the behavioral rhythm change matrix, including:

[0017] Global behavioral coordination analysis is performed on the behavioral rhythm change matrix based on the behavioral intensity change matrix. This includes extracting the behavioral intensity sequence and behavioral rhythm sequence of the aquaculture monitoring area at multiple time points, fusing the behavioral rhythm sequence based on the behavioral intensity sequence, calculating the behavioral coordination parameters at multiple time points, and generating the behavioral coordination sequence of the aquaculture monitoring area.

[0018] Local behavioral difference analysis is performed on the behavioral rhythm change matrix based on the behavioral intensity change matrix. This includes identifying multiple neighboring grid pairs in the aquaculture monitoring area at each time point, calculating the neighborhood behavioral difference parameters of each neighboring grid pair based on the behavioral rhythm change matrix, fusing the multiple neighborhood behavioral difference parameters at each time point according to the behavioral intensity change matrix, calculating the behavioral difference parameters at multiple time points, and generating a behavioral difference sequence for the aquaculture monitoring area.

[0019] Based on behavioral coordination sequences and behavioral difference sequences, multiple aquaculture status labels are determined for the aquaculture monitoring area, and status monitoring results for the aquaculture monitoring area are generated.

[0020] Preferably, performing local coverage detection on the window background image adjacent to the initial background includes:

[0021] Pixel difference processing is performed on the initial background and window background images to determine multiple occupied points in each grid region of the window background image adjacent to the initial background. The ratio of the number of occupied points in the grid region to the total number of pixels is calculated as a local coverage parameter.

[0022] Preferably, the local difference fusion weight of each neighboring grid pair is determined according to the behavior intensity change matrix, and the multiple neighborhood behavior difference parameters at each time step are fused based on the local difference fusion weight to calculate the behavior difference parameters at multiple time steps.

[0023] A second aspect of the present invention provides a black-boned chicken farming status monitoring system for implementing the aforementioned black-boned chicken farming status monitoring method, comprising:

[0024] The aquaculture data acquisition module is used to collect monitoring videos of the status of the black-bone chicken flock in the farm through camera devices, determine multiple aquaculture monitoring areas based on the facility configuration data of the farm, and extract the image set to be analyzed for each aquaculture monitoring area from the status monitoring video;

[0025] The state feature analysis module is used to perform background modeling on multiple aquaculture monitoring areas based on the image set to be analyzed, construct a dynamic background model for each aquaculture monitoring area, perform state detection on the image set to be analyzed based on the dynamic background model, and construct a state feature set for each aquaculture monitoring area.

[0026] The state evolution analysis module is used to extract the state sequence matrix of the aquaculture monitoring area from the state feature set, including the state evolution sequences of multiple grid areas, and to perform feature screening on the state sequence matrix to generate candidate state evolution matrices.

[0027] The breeding status monitoring module is used to decompose multiple candidate state evolution matrices, extract the behavior intensity change matrix and behavior rhythm change matrix of each breeding monitoring area, and perform abnormal state detection of black-bone chicken flocks in the breeding monitoring area based on the behavior intensity change matrix and behavior rhythm change matrix, and generate status monitoring results for different breeding monitoring areas.

[0028] The present invention has the following beneficial effects:

[0029] This invention utilizes dynamic background modeling of aquaculture monitoring video data, combined with the construction of behavioral intensity and rhythm change matrices, to comprehensively characterize and detect anomalies in the state characteristics of Silkie chicken flocks within the monitoring area, such as orderliness, crowding, and rhythm, without relying on individual-level identification. By generating behavioral coordination sequences and behavioral difference sequences, this invention can distinguish differences in flock states under different scenarios, thereby improving the accuracy and stability of state monitoring and making the monitoring results more closely aligned with actual aquaculture management needs. Attached Figure Description

[0030] Figure 1 This is a flowchart illustrating a method for monitoring the breeding status of black-boned chickens according to one embodiment of the present invention.

[0031] Figure 2 This is a schematic diagram of a black-boned chicken breeding status monitoring system provided in one embodiment of the present invention. Detailed Implementation

[0032] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention.

[0033] Please see Figure 1 The diagram illustrates a flowchart of a method for monitoring the breeding status of black-boned chickens according to one embodiment of the present invention. The method includes the following steps:

[0034] Step S1: Collect monitoring videos of the status of the black-boned chicken flock in the farm using a camera device. Determine multiple breeding monitoring areas based on the facility configuration data of the farm, and extract the image set to be analyzed for each breeding monitoring area from the status monitoring videos.

[0035] In this embodiment, video surveillance of the Silkie chicken flock can be collected by cameras deployed at different locations within the farm. Monitoring of the chickens' feeding and resting states can be combined with farm facility configuration data, such as location data of feed troughs, water lines, passageways, and roosting areas, to spatially divide the monitoring footage and determine multiple monitoring zones within the farm. Different monitoring zones can correspond to specific functional areas or spatial segments. After zone division, image sets for each monitoring zone are extracted from the monitoring videos to analyze the actual performance of the Silkie chicken flock under different farming conditions.

[0036] Step S2: Based on the image set to be analyzed, perform background modeling on multiple aquaculture monitoring areas, construct a dynamic background model for each aquaculture monitoring area, perform state detection on the image set to be analyzed based on the dynamic background model, and construct a state feature set for each aquaculture monitoring area.

[0037] In this embodiment, based on the image set to be analyzed, background modeling is performed for each breeding monitoring area to obtain a dynamic background model that can adapt to changes in ambient light and noise in the breeding scene. By comparing and analyzing different frame images with the background model, the foreground region in the image is extracted, thereby obtaining the state information of the black-boned chickens within the breeding monitoring area. A state feature set is constructed for each breeding monitoring area to describe the dynamic changes in the behavior of the black-boned chicken flock.

[0038] In some implementations, the specific methods for background modeling of multiple aquaculture monitoring areas based on the image set to be analyzed, and for constructing a dynamic background model for each aquaculture monitoring area, are as follows:

[0039] Temporal difference detection is performed on the image set to be analyzed, including calculating the global difference energy parameter between any adjacent frame images, and constructing multiple candidate background images of the image set to be analyzed based on the global difference energy parameter.

[0040] Specifically, the process of performing temporal differential detection on the image set to be analyzed involves performing differential processing on any adjacent frames in the image set in chronological order, and calculating the global differential energy parameter between adjacent frames. The global differential energy parameter can be obtained by statistically averaging the pixel differences across all frames, and is used to characterize the magnitude of the overall change between adjacent frames. By setting a global differential energy threshold, image frames with lower global differential energy are selected to construct multiple candidate background images, corresponding to scenes with less activity from the chicken flock, thus reflecting the background information of the monitored area.

[0041] Multiple candidate background images are processed by sliding window to construct multiple window background images of the image set to be analyzed. The initial background in the multiple window background images is determined and updated by neighbor diffusion to generate multiple target background images. Based on the multiple target background images, a dynamic background model of the aquaculture monitoring area is constructed.

[0042] Specifically, the image set to be analyzed is divided into multiple sliding windows in chronological order. Each window contains several candidate background images. The background image corresponding to each window is determined based on the multiple candidate background images within the window, for example, by averaging the multiple candidate background images and fusing them. Among the multiple window background images, one image is selected as the initial background image based on the principle of minimizing the difference energy. Specifically, the difference energy is calculated between any two window background images, and the window background image with the smallest sum of difference energy is selected as the initial background.

[0043] Furthermore, a neighbor diffusion update is performed on the initial background image to enhance the coverage and robustness of the background model. Specifically, for the window background image adjacent to the initial background image, local coverage detection is first performed on each grid region, and the local coverage parameter of each grid region is calculated. The local coverage parameter is used to measure whether the region is covered by chickens. When the local coverage parameter is lower than a preset coverage threshold, the region is considered usable as background. The grid regions in the window background image can be generated by dividing the aquaculture monitoring area. For the local coverage parameter, pixel difference can be performed between the initial background image and the window background image. Pixels with a difference value greater than a threshold, such as 15 gray levels, are marked as occupied points, indicating that they may be occupied by chickens. Then, the ratio of the number of occupied points in the grid region to the total number of pixels is calculated as the local coverage parameter. Grid regions with local coverage parameters not lower than the preset coverage threshold are removed from the window background image, and the image of that region in the initial background image is added to the window background image to complete the update of the window background image and mark it as the initial background. When the initial background image is not at the beginning or end of the window background image sequence, bidirectional neighbor diffusion update can be performed based on the initial background image. This completes the neighbor diffusion update process based on the initial background for multiple candidate background images, ultimately generating multiple target background images. Using multiple target background images as dynamic backgrounds under a continuous window, a dynamic background model of the aquaculture monitoring area is constructed, which is used to separate and extract foreground changes in the aquaculture scene.

[0044] In some implementations, the specific method for constructing the state feature set for each aquaculture monitoring area is as follows:

[0045] The process involves determining the multiple video frame images covered by each target background image in the dynamic background model, performing foreground detection on each video frame image based on the multiple target background images, identifying the foreground region in each video frame image, generating multiple state feature maps of the aquaculture monitoring area, and constructing a state feature set.

[0046] Specifically, in the dynamic background model, each target background image covers a different sliding window, thus determining the number of video frames covered by different target background images based on the video frame images contained in each sliding window. The video frame image to be detected can be pixel-level differentiated from the corresponding target background image to obtain a difference image. Then, thresholding is performed on the difference image to extract regions significantly different from the background as candidate foreground regions. To improve the accuracy of the detection results, morphological filtering such as erosion, dilation, opening, and closing operations can be further used to remove isolated noise points and fill local holes, as well as remove the influence of environmental noise such as dust and floating objects, thereby obtaining a coherent and complete foreground region. The constructed dynamic background model considers slow changes in illumination, such as sunlight angle and shadow movement, while also preventing some interfering information, such as chickens roosting for extended periods, from being learned into the background. Finally, a more accurate state feature set representing the state changes of the chicken flock can be constructed, including a state feature map in binary form corresponding to each video frame image.

[0047] Step S3: Extract the state sequence matrix of the aquaculture monitoring area from the state feature set, including the state evolution sequences of multiple grid areas, and perform feature screening on the state sequence matrix to generate candidate state evolution matrices.

[0048] In this embodiment, a state sequence matrix containing state evolution sequences of multiple grid regions is used to characterize the dynamic spatial distribution process of the population. To enhance the stability and robustness of subsequent analysis, the state sequence matrix is ​​subjected to feature screening, for example, by removing low-amplitude noise sequences to obtain candidate state evolution matrices, retaining the main evolution patterns related to the actual behavior of the population, and reducing misjudgments caused by occasional noise or environmental interference.

[0049] In some implementations, the specific method for extracting the state sequence matrix of the aquaculture monitoring area from the state feature set and generating candidate state evolution matrices by feature screening of the state sequence matrix is ​​as follows:

[0050] Local state features corresponding to multiple grid regions in each state feature map are extracted, and the state evolution sequence of each grid region is constructed based on multiple local state features.

[0051] Specifically, the state feature map contains the state information of the black-boned chicken flock in the breeding monitoring area. For each grid area, the proportion of pixels containing foreground information can be counted as local state features to characterize the coverage characteristics of black-boned chickens in the area. Multiple local state features of each grid area are spliced ​​together in time to construct a dynamic change in the coverage state of black-boned chickens in each grid area.

[0052] A state sequence matrix of the aquaculture monitoring area is generated based on multiple state evolution sequences, and the state sequence matrix is ​​filtered to generate a candidate state evolution matrix.

[0053] It's worth noting that traditional methods of monitoring flock behavior, such as density monitoring, primarily determine whether the breeding space for Silkie chickens is overcrowded, but lack sufficient differentiation in actual flock behavior. During the breeding process, Silkie chickens exhibit specific patterns under certain conditions. For example, during regular feeding, the flock will orderly rush to the feed trough, stay for a period to eat, and then gradually disperse. If overcrowding or blockage occurs, some chickens will try to squeeze in while others try to squeeze out, causing collisions and a relatively chaotic scene. Their activity patterns also show specific periods of activity, such as free foraging and resting, with overall activity remaining relatively regular. Multiple state evolution sequences represent the state change process of the Silkie chicken flock. In-depth analysis of multiple sequences can further detect whether there is disorder in the flock's behavioral state changes and uncover potential anomalies.

[0054] A state sequence matrix of a breeding monitoring area, composed of multiple state evolution sequences, can characterize abnormal group behavior of Silkie chickens under specific breeding conditions, such as feeding and resting, from the perspective of sequence waves. During the state analysis of multiple sequence waves, considering the brief shaking of chickens in the flock, such as swaying from side to side during walking, and the interference of local shadow flickering, the state sequence matrix is ​​filtered to remove low-frequency and high-frequency components, capturing short-term behaviors such as flocking in, staying, and dispersing. For example, a bandpass filter of [0.003, 0.03] Hz is applied to the state sequence matrix, allowing the candidate state evolution matrix to better characterize the rhythmic features of the Silkie chicken flock's group entry and exit from the feed trough area and their resting while feeding during the feeding process.

[0055] Step S4: Perform state decomposition on multiple candidate state evolution matrices respectively, extract the behavior intensity change matrix and behavior rhythm change matrix of each breeding monitoring area, and perform abnormal state detection of black-bone chicken flocks in the breeding monitoring area based on the behavior intensity change matrix and behavior rhythm change matrix to generate state monitoring results for different breeding monitoring areas.

[0056] In this embodiment, a time-series signal analysis approach is adopted to decompose the state sequence into a behavior intensity change matrix and a behavior rhythm change matrix. Specifically, a Hilbert transform is performed on the candidate state evolution matrix, decomposing the sequence matrix representing state changes into a behavior intensity change matrix and a behavior rhythm change matrix. The behavior intensity change matrix contains the instantaneous intensity information of the state sequence signals in different grid areas, used to describe the variation amplitude of the activity intensity and density of the black-boned chicken flock within the monitoring area. The behavior rhythm change matrix contains the phase information of the state sequence signals in different grid areas, used to describe the temporal rhythm and progress of the behavior changes of the black-boned chicken flock within the monitoring area. Based on the behavior intensity change matrix and the behavior rhythm change matrix, the dynamic change process of the black-boned chicken flock's behavior under different breeding scenarios is further analyzed to identify possible anomalies in the flock's group behavior patterns. Finally, the status monitoring results for each breeding monitoring area are generated, such as group orderliness, crowding level, rhythm stability, and potential anomaly warning information.

[0057] In some implementations, the specific method for detecting abnormal states of black-boned chicken flocks in a breeding monitoring area based on behavioral intensity change matrices and behavioral rhythm change matrices is as follows:

[0058] A global behavioral coordination analysis is performed on the behavioral intensity change matrix and the behavioral rhythm change matrix to generate a behavioral coordination sequence for the aquaculture monitoring area.

[0059] Specifically, for the global behavioral coordination analysis process, behavioral intensity sequences at multiple time points in the monitoring area are extracted from the behavioral intensity change matrix, and behavioral rhythm sequences at multiple time points are extracted from the behavioral rhythm change matrix. The behavioral rhythm sequences are then weighted and fused based on the behavioral intensity sequences. That is, in the process of characterizing the overall behavioral rhythm coordination level of the monitoring area through the behavioral rhythm sequences, grid areas with higher behavioral intensity are given greater weight. At each time point, a behavioral coordination parameter representing the global state is calculated, used to characterize the consistency of group behavior within the monitoring area at that time. The stronger the overall behavioral rhythm coordination, the higher the behavioral coordination parameter. If the behavior is chaotic, such as some areas entering while others leave, the behavioral rhythm is canceled out, resulting in a smaller behavioral coordination parameter, indicating lower synchronization. The behavioral coordination sequence of the monitoring area constructed based on temporal relationships depicts the trend of synchronization changes in the chicken flock over time. For example, in the feeding area, after feeding, the chicken flock enters and eats in a relatively orderly manner, and gradually leaves after finishing eating, indicating high overall coordination. However, if the feeding area becomes crowded or there is competition among the chickens, the consistency of overall behavior will decrease.

[0060] Based on the behavioral intensity change matrix, a local behavioral difference analysis was performed on the behavioral rhythm change matrix to generate a behavioral difference sequence for the aquaculture monitoring area.

[0061] Specifically, for the local behavioral difference analysis process, firstly, multiple neighboring grid pairs are identified within the aquaculture monitoring area, which can be vertically or horizontally adjacent grid pairs. Based on the behavioral rhythm change matrix, the neighborhood behavioral difference parameters of these neighboring grid pairs at each time step are calculated. This can be achieved by calculating the phase difference between two grids and processing it as an absolute value, reflecting the degree of difference in behavioral rhythm between adjacent areas. Further, combined with the behavioral intensity change matrix, these neighborhood behavioral difference parameters are weighted and fused. The fusion weight is set according to the mean of the intensity characteristics in the two grid areas, resulting in the local difference fusion weight for each neighboring grid pair. This ensures that neighborhoods with higher behavioral intensity have a greater weight in the overall difference calculation. Finally, the behavioral difference parameters characterizing the overall local difference state of the aquaculture monitoring area at each time step are calculated, and a behavioral difference sequence of the aquaculture monitoring area is constructed to reflect the local hedging, crowding, or disorder trends of the black-boned chicken flock over time. The larger the behavioral difference parameter, the more severe the local anomaly.

[0062] Based on behavioral coordination sequences and behavioral difference sequences, multiple aquaculture status labels are determined for the aquaculture monitoring area, and status monitoring results for the aquaculture monitoring area are generated.

[0063] Specifically, behavioral coordination parameters describing group behavior coordination and behavioral difference parameters describing local disorder characteristics can be divided into multiple intervals. For different combinations of these intervals, multiple breeding status labels are determined based on empirical knowledge. For example, when the behavioral coordination sequence significantly increases at a certain time period and the behavioral difference sequence is at a low level, it is judged as "orderly and concentrated feeding"; when both are at a low level, it is judged as "normal or resting". For the real-time behavioral coordination and behavioral difference sequences in the breeding monitoring area, the corresponding state intervals are determined and real-time breeding status labels are generated, thus generating the status monitoring results for the breeding monitoring area. This can be used for real-time early warning and breeding management decisions, enabling intelligent monitoring of the breeding status of Silkie chicken flocks, accurately identifying the real-time behavioral status of the flocks, and achieving intelligent management of Silkie chicken breeding.

[0064] Please see Figure 2 The diagram illustrates a structural schematic of a black-boned chicken farming status monitoring system provided in one embodiment of the present invention. The system includes:

[0065] The aquaculture data acquisition module is used to collect monitoring videos of the status of the black-bone chicken flock in the farm through camera devices, determine multiple aquaculture monitoring areas based on the facility configuration data of the farm, and extract the image set to be analyzed for each aquaculture monitoring area from the status monitoring video;

[0066] The state feature analysis module is used to perform background modeling on multiple aquaculture monitoring areas based on the image set to be analyzed, construct a dynamic background model for each aquaculture monitoring area, perform state detection on the image set to be analyzed based on the dynamic background model, and construct a state feature set for each aquaculture monitoring area.

[0067] The state evolution analysis module is used to extract the state sequence matrix of the aquaculture monitoring area from the state feature set, including the state evolution sequences of multiple grid areas, and to perform feature screening on the state sequence matrix to generate candidate state evolution matrices.

[0068] The breeding status monitoring module is used to decompose multiple candidate state evolution matrices, extract the behavior intensity change matrix and behavior rhythm change matrix of each breeding monitoring area, and perform abnormal state detection of black-bone chicken flocks in the breeding monitoring area based on the behavior intensity change matrix and behavior rhythm change matrix, and generate status monitoring results for different breeding monitoring areas.

[0069] The above are merely specific embodiments of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A method for monitoring the breeding status of black-boned chickens, characterized in that, include: The status monitoring video of the black-bone chicken flock in the farm is collected by the camera device. Multiple breeding monitoring areas are determined according to the facility configuration data of the farm. The image set to be analyzed for each breeding monitoring area is extracted from the status monitoring video. Temporal difference detection is performed on the image set to be analyzed, including calculating the global difference energy parameter between any adjacent frame images, and constructing multiple candidate background images of the image set to be analyzed based on the global difference energy parameter; Multiple candidate background images are processed by sliding window to construct multiple window background images of the image set to be analyzed. The initial background in the multiple window background images is determined and updated by neighbor diffusion, including local coverage detection of the window background images adjacent to the initial background, calculating the local coverage parameters of each grid region in the window background image, locally cutting the window background image and updating it to the initial background, generating multiple target background images after traversing multiple window background images, constructing a dynamic background model of the aquaculture monitoring area based on the multiple target background images, and performing state detection on the image set to be analyzed based on the dynamic background model to construct a state feature set for each aquaculture monitoring area. For the construction of the state feature set, the multiple video frame images covered by each target background image in the dynamic background model are determined. Foreground detection is performed on each video frame image based on the multiple target background images to determine the foreground region in each video frame image. Multiple state feature images of the aquaculture monitoring area are generated and the state feature set is constructed. The aquaculture monitoring area is divided into multiple grid areas. Local state features corresponding to multiple grid areas in each state feature map are extracted. A state evolution sequence for each grid area is constructed based on multiple local state features. A state sequence matrix of the aquaculture monitoring area is generated based on multiple state evolution sequences. The state sequence matrix is ​​then filtered to generate a candidate state evolution matrix. State decomposition was performed on multiple candidate state evolution matrices to extract the behavior intensity change matrix and behavior rhythm change matrix for each aquaculture monitoring area. Global behavioral coordination analysis is performed on the behavioral rhythm change matrix based on the behavioral intensity change matrix. This includes extracting the behavioral intensity sequence and behavioral rhythm sequence of the aquaculture monitoring area at multiple time points, fusing the behavioral rhythm sequence based on the behavioral intensity sequence, calculating the behavioral coordination parameters at multiple time points, and generating the behavioral coordination sequence of the aquaculture monitoring area. Local behavioral difference analysis is performed on the behavioral rhythm change matrix based on the behavioral intensity change matrix. This includes identifying multiple neighboring grid pairs in the aquaculture monitoring area at each time point, calculating the neighborhood behavioral difference parameters of each neighboring grid pair based on the behavioral rhythm change matrix, fusing the multiple neighborhood behavioral difference parameters at each time point according to the behavioral intensity change matrix, calculating the behavioral difference parameters at multiple time points, and generating a behavioral difference sequence for the aquaculture monitoring area. Based on behavioral coordination sequences and behavioral difference sequences, multiple aquaculture status labels are determined for aquaculture monitoring areas, generating status monitoring results for different aquaculture monitoring areas.

2. The method for monitoring the breeding status of black-boned chickens according to claim 1, characterized in that, Local cover detection of the window background image adjacent to the initial background includes: Pixel difference processing is performed on the initial background and window background images to determine multiple occupied points in each grid region of the window background image adjacent to the initial background. The ratio of the number of occupied points in the grid region to the total number of pixels is calculated as a local coverage parameter.

3. The method for monitoring the breeding status of black-boned chickens according to claim 1, characterized in that, The local difference fusion weight of each neighboring grid pair is determined based on the behavior intensity change matrix. Based on the local difference fusion weight, multiple neighborhood behavior difference parameters at each time step are fused to calculate the behavior difference parameters at multiple time steps.

4. A monitoring system for the breeding status of black-boned chickens, characterized in that, The system is used to implement the method for monitoring the breeding status of black-boned chickens as described in any one of claims 1-3, including: The aquaculture data acquisition module is used to collect monitoring videos of the status of the black-bone chicken flock in the farm through camera devices, determine multiple aquaculture monitoring areas based on the facility configuration data of the farm, and extract the image set to be analyzed for each aquaculture monitoring area from the status monitoring video; The state feature analysis module is used to perform background modeling on multiple aquaculture monitoring areas based on the image set to be analyzed, construct a dynamic background model for each aquaculture monitoring area, perform state detection on the image set to be analyzed based on the dynamic background model, and construct a state feature set for each aquaculture monitoring area. The state evolution analysis module is used to extract the state sequence matrix of the aquaculture monitoring area from the state feature set, including the state evolution sequences of multiple grid areas, and to perform feature screening on the state sequence matrix to generate candidate state evolution matrices. The breeding status monitoring module is used to decompose multiple candidate state evolution matrices, extract the behavior intensity change matrix and behavior rhythm change matrix of each breeding monitoring area, and perform abnormal state detection of black-bone chicken flocks in the breeding monitoring area based on the behavior intensity change matrix and behavior rhythm change matrix, and generate status monitoring results for different breeding monitoring areas.

Images