Chicken disease symptom image recognition system and method based on deep learning

Abstract

This invention discloses a deep learning-based image recognition system and method for chicken disease symptoms, relating to the field of image recognition technology. The system includes: acquiring a set of spatial location images and a set of feeding images of target chickens within a preset time period; extracting a dataset of spatial location relationships between the target chickens and feed troughs, feeding frequency data, and a social network graph based on the spatial location image set and the feeding image set; correcting the feeding frequency data based on the spatial location relationship dataset to obtain corrected feeding behavior data; inputting the feeding behavior data and the social network graph into a preset deep learning model to generate a probability of behavioral abnormality; if the probability of abnormality exceeds a preset judgment threshold, the target chicken is judged to have suspected disease behavior. Its beneficial effects include: it can integrate spatial location relationship correction and social network graph analysis, effectively eliminating the interference of chicken house layout on behavioral characteristics, and has the advantages of improving the accuracy of disease identification and reducing misjudgments due to environmental factors.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a deep learning-based image recognition system and method for chicken disease symptoms. Background Technology

[0002] Chicken disease symptom image recognition is a technology that uses computer vision and deep learning to identify whether chickens are suffering from diseases by analyzing the features of chicken images. It is also an important research direction in the automation of livestock and poultry health management in recent years.

[0003] In existing technologies, some studies have proposed predicting whether chickens are sick by detecting behavioral characteristics strongly associated with disease, such as feeding behavior or "social withdrawal" behavior. This approach has certain positive significance in the early warning of diseases, especially in the detection of individuals with suspected abnormal behavior, where it has considerable sensitivity.

[0004] However, in actual farming scenarios, chicken behavior is easily affected by various non-pathological factors such as the layout of the chicken house and individual differences. For example, in cold environments, farms usually place centralized heat sources at one end or in the center of the chicken house to ensure the flock stays warm. Feed troughs, to prevent heat damage, are often placed on the side far from the heat source or at a higher position on the roof. In this case, weak or stressed chickens, due to their preference for warm areas, often gather near the heat source and stay away from the feed troughs. At the same time, because they hide in the warm corners of the house for a long time, their interaction with the group center is reduced. At this time, these chickens are easily misjudged as sick chickens. In fact, these chickens are neither sick nor are they really avoiding food or social interaction; it is just a pure "temperature preference" behavior. In such cases, predicting the health status of chickens solely based on behavioral characteristics strongly correlated with disease is prone to high false alarms.

[0005] Therefore, a deep learning-based image recognition system and method for chicken disease symptoms are proposed. Summary of the Invention

[0006] In view of the above-mentioned prior art, this application is hereby filed. Embodiments of this application provide a deep learning-based image recognition system and method for chicken disease symptoms, which can improve the accuracy of disease recognition and reduce misjudgments due to environmental factors.

[0007] According to one aspect of this application, a deep learning-based image recognition method for chicken disease symptoms is provided, comprising: acquiring a continuous spatial location image set and a feeding image set of a target chicken within a preset time period; extracting, based on the continuous spatial location image set and the feeding image set, a spatial location relationship dataset between the target chicken and the feed trough within the preset time period, feeding frequency data, and a social network graph representing the interaction intensity between the target chicken and neighboring chickens; correcting the feeding frequency data based on the spatial location relationship dataset to obtain feeding behavior data after eliminating the influence of spatial layout; inputting the feeding behavior data and the social network graph into a preset deep learning model to generate the probability of abnormal behavior of the target chicken within the preset time period; determining whether the probability of abnormal behavior exceeds a preset judgment threshold, and if so, determining that the target chicken has suspected disease behavior.

[0008] According to another aspect of this application, a deep learning-based image recognition system for chicken disease symptoms is provided, comprising: a data acquisition module for acquiring a continuous spatial location image set and a feeding image set of a target chicken within a preset time period; a data processing module for extracting, based on the continuous spatial location image set and the feeding image set, a dataset of the spatial location relationship between the target chicken and the feed trough within the preset time period, feeding frequency data, and a social network graph representing the interaction intensity between the target chicken and neighboring chickens; a deviation correction module for correcting the feeding frequency data based on the spatial location relationship dataset to obtain feeding behavior data after eliminating the influence of spatial layout; an anomaly probability prediction module for inputting the feeding behavior data and the social network graph into a preset deep learning model to generate the behavioral anomaly probability of the target chicken within the preset time period; and a decision module for determining whether the behavioral anomaly probability exceeds a preset judgment threshold, and if so, determining that the target chicken exhibits suspected disease behavior.

[0009] According to another aspect of this application, an electronic device is provided, including a memory and a processor, the memory being used to store computer-executable instructions, and the processor being used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method described above.

[0010] According to another aspect of this application, a computer storage medium is provided that stores computer-executable instructions thereon, which, when executed by a processor, implement the steps of the method described above.

[0011] Compared with the prior art, the deep learning-based chicken disease symptom image recognition system and method according to the embodiments of this application can integrate spatial location relationship correction and social network graph analysis, effectively eliminate the interference of chicken house layout on behavioral characteristics, and more accurately identify disease-related behavioral abnormalities by combining deep learning models. It has the advantages of improving the accuracy of disease recognition and reducing misjudgment due to environmental factors. Attached Figure Description

[0012] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0013] Figure 1 This is a flowchart of the deep learning-based image recognition method for chicken disease symptoms according to the present invention.

[0014] Figure 2 This is a block diagram of the chicken disease symptom image recognition system based on deep learning according to the present invention.

[0015] Figure 3 This is a block diagram of an electronic device according to the present invention. Detailed Implementation

[0016] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0017] Application Overview

[0018] In actual farming scenarios, non-pathological factors such as chicken house layout and temperature distribution significantly interfere with chicken behavior. Existing technologies rely on behavioral characteristics strongly correlated with disease, such as feeding and social withdrawal, as disease prediction indicators. In actual farming scenarios, chicken house environmental factors can easily interfere with the correlation between chicken behavior and disease status, directly causing the input data to contain non-pathological noise, resulting in a false alarm rate that exceeds the acceptable range.

[0019] If the above problems are not resolved, adaptive behaviors and pathological behaviors will be confused, the false alarm rate will fluctuate with the chicken house environment, the stability of disease warning will be undermined, the persistent misjudgments will lead to a decline in farmers' trust in the warnings, reduce the response efficiency of disease prevention and control measures, and more seriously, real diseased individuals may be missed due to alarm fatigue caused by system false alarms, delaying the best treatment time.

[0020] To address the aforementioned challenges, this application first considers how to eliminate the interference of chicken coop layout on behavioral feature extraction. It attempts to correct behavioral features using spatial location data. Further investigation reveals that relying solely on spatial location data may overlook the impact of group interaction on individual behavior. Therefore, a social network graph is introduced to capture dynamic interaction patterns among chickens. Ultimately, a dual mechanism integrating spatial location correction and group interaction analysis is determined to provide interference-resistant disease early warning.

[0021] Exemplary methods

[0022] Figure 1 The illustration shows a deep learning-based image recognition method for chicken disease symptoms according to an embodiment of this application, including: acquiring a continuous spatial location image set and a feeding image set of a target chicken within a preset time period; extracting a spatial location relationship dataset between the target chicken and the feed trough, feeding frequency data, and a social network graph representing the interaction intensity between the target chicken and neighboring chickens within the preset time period based on the continuous spatial location image set and the feeding image set; correcting the feeding frequency data based on the spatial location relationship dataset to obtain feeding behavior data after eliminating the influence of spatial layout; inputting the feeding behavior data and the social network graph into a preset deep learning model to generate the probability of abnormal behavior of the target chicken within the preset time period; determining whether the probability of abnormal behavior exceeds a preset judgment threshold, and if so, determining that the target chicken has suspected disease behavior.

[0023] The continuous spatial location image set refers to a collection of images capturing the spatial location of target chickens within a preset time period, acquired through image acquisition equipment. This can be achieved by periodically capturing images of chicken activity in different areas of the chicken house using a camera array or mobile monitoring equipment, recording the dynamic positional relationship between the chickens and the feed trough. The feeding image set refers to a collection of images capturing the feeding behavior of target chickens within a preset time period, acquired through image acquisition equipment. This can be achieved by using high-resolution cameras deployed near the feed trough to capture images of the chickens' heads or beaks contacting the feed, used to count the frequency of feeding. The spatial location relationship dataset refers to a set of quantitative data generated by analyzing the relative positions of target chickens and the feed trough in the continuous spatial location image set. This can be achieved by using image recognition algorithms to extract the chicken's outline coordinates and perform distance calculations and trajectory tracking with the feed trough coordinates, reflecting the spatial preference characteristics of chickens caused by environmental layout. Deep learning models refer to neural network models used to jointly analyze corrected eating behavior data and social network graphs. Specifically, they can be implemented by using convolutional neural networks to extract spatial features, graph neural networks to process social relationship data, and fully connected layers to perform probability prediction. They are used to identify abnormal patterns by combining individual behavior and group interaction features.

[0024] The core innovation of this application lies in constructing a disease identification model resistant to environmental interference by integrating multi-dimensional behavioral data and introducing a spatial bias correction mechanism. By simultaneously collecting spatial location and feeding behavior data, and correcting feeding frequency based on spatial relationships, it effectively distinguishes between passive behavioral abnormalities caused by environmental layout and active behavioral abnormalities caused by actual diseases. By combining social network graphs to capture group interaction features, it enhances the model's accuracy in identifying disease-marking behaviors such as "social withdrawal," thereby reducing the false alarm rate in complex aquaculture scenarios.

[0025] For example, the specific implementation of the solution in this application is as follows:

[0026] First, a continuous set of spatial location images and a set of feeding images of the target chickens were acquired over a 24-hour period. The spatial location images were acquired at a frequency of one frame every 10 seconds, while the feeding images were acquired when the chickens approached the feed trough.

[0027] Then, based on the collected image set, a dataset of the spatial relationship between the target chickens and the feed trough was extracted. Specifically, the Euclidean distance between the target chicken and the center of the feed trough in each frame of the image was calculated. Simultaneously, the feeding frequency data of the target chickens was statistically analyzed, recording the number of feedings per hour. Furthermore, by analyzing the changes in the spatial distance between the target chicken and surrounding chickens, a social network graph representing the interaction intensity was constructed.

[0028] Then, the eating frequency data was corrected using the spatial location relationship dataset to obtain eating behavior data after eliminating the influence of spatial layout.

[0029] Next, the corrected eating behavior data and social network graph are input into a pre-trained convolutional neural network model. This model contains multiple convolutional and fully connected layers to extract behavioral features and generate probabilities of behavioral anomalies.

[0030] Finally, a threshold of 0.7 was set for the probability of abnormal behavior. When the probability value output by the model exceeds this threshold, the target chicken is judged to have suspected disease behavior.

[0031] Through the above scheme, this application can effectively eliminate the interference of non-pathological factors such as chicken house layout and chicken house environment on the extraction of chicken behavioral characteristics. Through the spatial deviation correction mechanism, it can accurately restore the chicken's true feeding intention, avoid misjudgment of behavior caused by environmental factors, and combine social network graph analysis to comprehensively capture individual and group behavioral characteristics, thereby improving the accuracy of disease identification.

[0032] This application provides two methods for correcting the frequency of eating:

[0033] In one example, correcting the feeding frequency involves inputting a spatial location relation dataset into a pre-defined probability distribution model of the target chickens' activity relative to the feed trough, to obtain the target chickens' activity frequency within a pre-defined time period. The internal locations are at various points relative to the feed trough. Activity probability Based on the probability of the activity Calculate within the preset time period Spatial deviation factor of the relative position of the target chickens and the feed trough :

[0034] ;

[0035] Determine the spatial deviation factor Is it greater than the preset maximum value of the spatial deviation factor? Or less than the preset minimum spatial deviation factor If so, then Set to zero; based on the spatial deviation factor Data on eating frequency Correction was performed to obtain corrected eating behavior data. , ,in, The preset spatial deviation adjustment coefficient is used to control the intensity of the influence of spatial deviation on the correction of feeding frequency.

[0036] Among them, the activity probability distribution model can be trained through historical location data, such as based on hidden Markov models or Gaussian mixture models, to quantify the passive position shift of chickens caused by individual preferences, etc.

[0037] Specifically, the activity probability distribution model can be trained using historical location data, such as based on Hidden Markov Models or Gaussian Mixture Models. This model quantifies passive position shifts in chickens due to individual preferences. It establishes a probability density function by analyzing the spatial relationship between the chickens' historical activity trajectories and the feed trough location. For example, by statistically analyzing the frequency of chickens' dwell time around the feed trough and in the area of ​​the disturbance source, the area near the disturbance source exhibits a lower probability value, reflecting passive distancing behavior. A threshold processing mechanism for the spatial deviation factor can eliminate abnormal deviations caused by sensor errors or temporary environmental disturbances, such as short-term startle. When the deviation factor exceeds a reasonable range, it is set to zero. During the correction process, an adjustment coefficient is introduced. The adjustment coefficient can be dynamically adjusted based on chicken house layout parameters through empirical models or rule settings. For example, the coefficient can be automatically reduced to prevent overcorrection when the disturbance source is close to the feed trough.

[0038] Through the above technical solution, this application can effectively distinguish the behavioral differences between chickens actively avoiding feed and passively moving away from feed troughs. The activity probability distribution model captures the positional shift of chickens caused by non-pathological factors such as heat source preference. The spatial deviation factor is calculated by the probability mean in the time dimension, which objectively reflects the limitation of the environmental layout on the activity range of chickens. The spatial deviation factor threshold is set to zero outliers, eliminating the noise impact caused by extreme environmental interference or data acquisition errors. Finally, the correction amplitude is dynamically controlled by the adjustment coefficient, so that the feeding frequency data retains the disease-related behavioral characteristics and eliminates the systematic bias caused by the environmental layout, which significantly improves the anti-interference ability and accuracy of the disease prediction model.

[0039] In another example, correcting feeding frequency involves inputting a spatial location relation dataset into a pre-defined path accessibility weight distribution model relative to feed troughs within the chicken coop to obtain the target chickens' feeding frequency within a pre-defined time period. The internal locations are at various points relative to the feed trough. Path reachability weights , According to each location point Path reachability weights Calculate the target chickens within a preset time period Internal path accessibility factor relative to the feed trough :

[0040] ;

[0041] Determine path reachability factors Is it greater than the preset maximum path reachability factor? Or less than the preset minimum path reachability factor If so, then Set to zero; based on path reachability factor Data on eating frequency Correction was performed to obtain corrected eating behavior data. , ,in, These are preset weighting coefficients used to control the strength of the impact of path accessibility on the correction of eating frequency.

[0042] The path accessibility weight distribution model quantifies the ease with which different locations can reach the feed trough, transforming spatial relationships into calculable weight values. A weight value closer to 1 indicates a smoother path, while a lower value reflects an obstructed path. The model can be constructed based on the topological structure of the chicken house layout, such as by using a gridded chicken house plan, assigning accessibility weights to each grid cell. These weights can be determined based on the number of obstacles between the grid cell and the feed trough, or the curvature of the path.

[0043] Specifically, the path accessibility weight distribution model first divides the chicken coop space into multiple grid cells. The accessibility weight of each grid cell is calculated based on the path complexity between it and the feed trough. For example, a grid cell with no obstacles around the feed trough has a weight of 1, while the weight of a grid cell with fixed obstacles decreases according to the obstacle density, and the weight decreases by 0.1 for each additional turn in the path. Within a preset time period T, the location points of the chickens... It is mapped to the corresponding raster cell to obtain the real-time weight at that location. Path reachability factor By calculating 1 minus the average weight of all time points, for example, when chickens remain in low-weight areas for an extended period due to obstacles, The value increased significantly. (Based on judgment) Does it exceed or below This can eliminate abnormal data caused by sensor positioning drift, such as when chickens briefly enter an inaccessible area. It may momentarily exceed 0.8; in this case, set it to zero to avoid overcorrection. Weighting coefficients. The setting can be dynamically adjusted according to the chicken house layout. For example, in a complex layout chicken house, it can be set to 1.2 to effectively compensate for the underestimation of feeding frequency caused by path obstruction. As a result, the reduced feeding behavior of chickens due to path obstacles is accurately restored, avoiding the misjudgment of low feeding frequency caused by environmental factors as a disease symptom.

[0044] Through the aforementioned technical solution, this application quantifies the ease of reaching feed troughs from different locations within the chicken house by introducing a path accessibility weight distribution model. This more accurately reflects the actual feeding behavior of chickens. By calculating path accessibility factors and applying threshold judgments, interference from abnormal data is effectively eliminated. Finally, by dynamically adjusting feeding frequency data, the risk of misjudgment caused by path inaccessibility is eliminated, improving the accuracy of disease identification. This method is particularly suitable for chicken house environments with complex layouts, effectively distinguishing between reduced feeding due to path obstruction and reduced feeding due to disease, thereby reducing the false alarm rate and improving the reliability of chicken health monitoring.

[0045] The two methods for correcting feeding frequency described above target different types of interference sources. The activity probability distribution model is more suitable for identifying and correcting behavior-driven biases such as environmental preferences and individual behavioral patterns, while the path accessibility weight distribution model is more suitable for determining and correcting path-restricted biases in complex chicken house layouts and environments with multiple obstacles. Both methods quantify the "feeding opportunities" between chickens and feed troughs using a data-driven approach, correcting spatial biases in the original feeding frequency data from different angles. This effectively removes non-pathological influencing factors and retains behavioral signals that truly reflect disease tendencies. In practical applications, the two methods can be flexibly selected or combined based on the specific layout characteristics of the chicken house: for open chicken houses with uniform structures but environmental gradients, the activity probability model can be prioritized, while for intensive chicken houses with many partitions and dense obstacles, the path accessibility model is more adaptable. Furthermore, a hybrid correction mechanism can be designed to calculate the two bias factors in parallel during the preprocessing stage and generate more accurate correction parameters using a weighted fusion method, further improving the robustness and discriminative ability of the method in diverse environments.

[0046] In some of the solutions described above in this application, since individuals with different roles in a flock have inherent differences in their behavioral patterns under normal conditions, using a fixed judgment threshold may lead to dominant individuals being misjudged as abnormal due to aggressive behavior, or weak individuals being incorrectly identified as sick due to avoidance behavior, resulting in a decrease in the accuracy of disease identification.

[0047] In this regard, this application further proposes that before determining whether the probability of abnormal behavior exceeds a preset judgment threshold, the following steps are also included: obtaining the preset role tag of the target chicken in the flock; and adjusting the preset judgment threshold according to the role tag.

[0048] The acquisition of role tags can be achieved by calling a pre-stored role tag database to match the role of the target chicken in the flock. The role tag database can be constructed by analyzing the historical group interaction patterns of each chicken through image recognition technology. For example, by tracking the movement trajectory, pecking frequency, or space occupied by each chicken, it can be determined whether it belongs to the leader chicken, ordinary chicken, or weak chicken, and then the corresponding role tag is assigned.

[0049] This application provides two methods for adjusting the preset judgment threshold:

[0050] In one example, adjusting the preset judgment threshold includes: retrieving the preset character tag-judgment threshold mapping table based on the character tag to obtain the preset judgment threshold corresponding to the character tag.

[0051] Among them, the role tag-judgment threshold mapping table stores the judgment threshold parameters corresponding to different roles in advance through historical data or experimental calibration.

[0052] Specifically, the role tag-judgment threshold mapping table can include the following role tags and their corresponding thresholds: leader chicken (0.7), ordinary chicken (0.8), and vulnerable chicken (0.9). When it is necessary to determine the disease of a target chicken, the role tag of the target chicken is first obtained, and then the corresponding preset judgment threshold is retrieved from the mapping table based on the role tag. Specifically, if the role tag of the target chicken is "leader chicken", the preset judgment threshold retrieved from the mapping table is 0.7; if the role tag is "ordinary chicken", the preset judgment threshold retrieved is 0.8; and if the role tag is "vulnerable chicken", the preset judgment threshold retrieved is 0.9. In this way, the judgment threshold can be dynamically adjusted according to the behavioral characteristics of chickens with different roles, thereby improving the accuracy of disease determination.

[0053] Through the above technical solution, this application can apply corresponding judgment thresholds to chickens in different roles, effectively adapting to behavioral differences among chickens in different roles. This method avoids judgment bias that may be caused by using a uniform threshold, and improves the sensitivity and specificity of disease judgment. For example, for weak chickens with a small activity range, using a higher judgment threshold can reduce the false alarm rate; while for leader chickens that are more active, a lower judgment threshold is used to improve detection sensitivity. Therefore, the technical solution of this application can more accurately identify abnormal behaviors of chickens in different roles, thereby improving the overall disease judgment effect.

[0054] In another example, adjusting the preset judgment threshold includes: obtaining the probability of abnormal behavior of the target chicken character tag. Probability of abnormal character behavior This represents the average probability that chickens corresponding to a role tag deviate from normal chicken behavior patterns over a historical period; it also retrieves the pre-defined overall probability of abnormal chicken behavior in the flock. Probability of abnormal behavior in chicken flocks This represents the average probability that the entire flock deviates from the normal chicken behavior pattern over a historical period; based on the probability of abnormal behavior of a role. Probability of abnormal behavior in chicken flocks Adjust the preset judgment threshold :

[0055] ;

[0056] in, This indicates the adjusted preset judgment threshold. This represents a preset weighting coefficient used to control the probability of abnormal character behavior. For the preset judgment threshold The adjustment range satisfies .

[0057] The probability of abnormal behavior for a specific role can be obtained by statistically analyzing the percentage of times chickens with the same role label were marked as exhibiting abnormal behavior within a historical period. For example, for chickens labeled "vulnerable individual," their historical probability of abnormal behavior might reach 15%. The probability of abnormal behavior in the flock can be calculated by averaging the abnormal behavior data of all chickens. For instance, when the flock's overall activity decreases due to a sudden drop in ambient temperature, this probability might rise to 8%. The weighting coefficient β can be set between 0.3 and 0.7, with the specific value adjusted based on the farm's actual false alarm tolerance. The formula for adjusting... / The ratio reflects the degree of deviation between an individual's role and the group benchmark. When the ratio is greater than 1, the adjustment range of the judgment threshold is related to... The values ​​are positively correlated.

[0058] Specifically, before determining whether the probability of abnormal behavior in a target chicken exceeds a threshold, historical abnormal data corresponding to the role tag to which the chicken belongs is first extracted from the database, while the overall abnormal baseline value of the current flock is obtained. A dynamically adapted threshold is generated by substituting the ratio of the role's abnormal probability to the flock's overall abnormal probability into an adjustment formula. For example, when the role's abnormal probability for a certain "active individual" is 5%, while the overall abnormal probability of the flock is 10%, / =0.5, at this point the judgment threshold will be increased. This effectively reduces the risk of misjudgment for chickens in this role. When a significant fluctuation in the probability of abnormal group behavior is detected, such as changes in group activity patterns due to seasonal changes, this adjustment mechanism can automatically correct the judgment thresholds for each role, avoiding individual misjudgments caused by changes in group behavior trends. This is achieved by introducing weighting coefficients. The sensitivity of threshold adjustment can be controlled, for example when When set to 0.5, the threshold adjustment range is limited to 50% of the original threshold to prevent threshold distortion in extreme cases. This dynamic adjustment process, combined with feeding behavior data corrected for spatial deviation, can eliminate the impact of individual behavioral deviations caused by chicken house layout on the final judgment.

[0059] Through the above technical solution, this application achieves dynamic threshold adjustment based on role tags and group behavior. This method considers both the inherent behavioral differences of chickens with different roles and uses the overall behavior of the group as a reference benchmark, thereby improving the accuracy of disease behavior judgment in complex breeding environments. By introducing the ratio of abnormal probability of role behavior to abnormal probability of flock behavior, the scheme can dynamically adapt to the correlation between changes in the overall behavior of the flock and individual roles, avoiding judgment bias that may be caused by simply relying on a fixed threshold. Simultaneously, weighting coefficients... The introduction of this mechanism allows for flexible control of the adjustment range, preventing misjudgments caused by over-correction. This dynamic adjustment mechanism is particularly suitable for complex and ever-changing aquaculture scenarios, effectively improving the sensitivity and specificity of abnormal behavior detection and providing a more reliable basis for early disease warning.

[0060] The two preset threshold adjustment mechanisms described above, based on static role classification and dynamic behavioral deviation, form a highly complementary dual-adaptation mechanism. The role label-threshold mapping method is suitable for environments with stable labels and clear behavioral characteristics, enabling rapid assignment without real-time computation, thus improving deployment efficiency. The dynamic adjustment mechanism based on the ratio of abnormal behavior probabilities is more suitable for real-world scenarios with significant fluctuations in role behavior or frequent changes in flock status, responding in real-time to changes in the overall flock behavior and dynamically correcting the judgment criteria. In practical applications, these two methods can be used individually or in combination. For example, an initial judgment threshold can be obtained by looking up the role label table, and then dynamically fine-tuned based on the current period's group behavior data, forming a hierarchical judgment threshold generation process. This fusion mechanism not only considers the model's generalization ability and individual adaptability but also allows for setting weight preferences based on false alarm tolerance, forming differentiated judgment strategies for different management objectives.

[0061] In the above technical solutions of this application, the extraction of social network graphs includes: identifying the position coordinates of the target chicken and its neighboring chickens at each time node from a continuous spatial location image set; determining the spatial proximity relationship between the target chicken and its neighboring chickens within a preset time period based on the position coordinates of the target chicken and its neighboring chickens at each time node; calculating the interaction intensity between the target chicken and its neighboring chickens based on the spatial proximity relationship; and constructing a social network graph representing the interaction intensity between the target chicken and its neighboring chickens based on the interaction intensity.

[0062] Location coordinate identification can be achieved through target detection algorithms, such as using the YOLOv5 model to locate chickens in real time across consecutive image frames, outputting two-dimensional coordinate data for each time point. Determining spatial proximity requires time series analysis, such as counting the number of consecutive approaches between the target chicken and its neighbors within a preset time period. When the distance between adjacent chickens is less than 0.5 meters and the duration exceeds 30 seconds, it is considered a valid spatial proximity event. The calculation of interaction intensity can incorporate a time decay function; for example, the weight of a single proximity event can be set as the logarithm of the contact duration, resulting in differentiated contributions between short-term and long-term contacts. The construction of the social network graph can employ a graph database storage structure, where nodes represent individual chickens, and edge weights are determined by normalized interaction intensity values.

[0063] Through the above technical solution, this application effectively solves the problem of misjudgment of social behavior that may occur when calculating interaction intensity solely based on the location coordinates of discrete time points in situations with complex chicken coop layouts or environmental interference factors. By extracting location information from multiple time points from a continuous spatial location image set and combining it with the time dimension for analysis, the true social behavior patterns among chickens can be reflected more accurately. This dynamic spatiotemporal data analysis method can distinguish between accidental proximity and continuous social behavior, avoiding misjudgments caused by environmental factors. Furthermore, by combining spatial distance with temporal frequency to calculate interaction intensity, the actual frequency of interaction among chickens can be quantified more precisely. The final constructed social network graph, through the weighted relationship of interaction intensity, can accurately characterize the social activity of chickens, providing more reliable behavioral feature input for subsequent disease identification models, thereby improving the accuracy and reliability of disease identification.

[0064] Exemplary System

[0065] Figure 2 The illustration shows a deep learning-based image recognition system for chicken disease symptoms according to an embodiment of this application, comprising: a data acquisition module for acquiring a continuous spatial location image set and a feeding image set of a target chicken within a preset time period; a data processing module for extracting, based on the continuous spatial location image set and the feeding image set, a dataset of the spatial location relationship between the target chicken and the feed trough within the preset time period, feeding frequency data, and a social network graph representing the interaction intensity between the target chicken and neighboring chickens; a deviation correction module for correcting the feeding frequency data based on the spatial location relationship dataset to obtain feeding behavior data after eliminating the influence of spatial layout; an anomaly probability prediction module for inputting the feeding behavior data and the social network graph into a preset deep learning model to generate the behavioral anomaly probability of the target chicken within the preset time period; and a decision module for determining whether the behavioral anomaly probability exceeds a preset judgment threshold, and if so, determining that the target chicken exhibits suspected disease behavior.

[0066] In one example, the deviation correction module corrects the feeding frequency by: inputting a spatial location relationship dataset into a preset probability distribution model of the target chickens' activity relative to the feed trough, to obtain the target chickens' activity frequency within a preset time period. The internal locations are at various points relative to the feed trough. Activity probability Based on the probability of the activity Calculate within the preset time period Spatial deviation factor of the relative position of the target chickens and the feed trough :

[0067] ;

[0068] Determine the spatial deviation factor Is it greater than the preset maximum value of the spatial deviation factor? Or less than the preset minimum spatial deviation factor If so, then Set to zero; based on the spatial deviation factor Data on eating frequency Correction was performed to obtain corrected eating behavior data. , ,in, The preset spatial deviation adjustment coefficient is used to control the intensity of the influence of spatial deviation on the correction of feeding frequency.

[0069] In another example, the deviation correction module corrects the feeding frequency by inputting a spatial location relationship dataset into a preset path accessibility weight distribution model relative to the feed trough within the chicken house, to obtain the target chickens' feeding frequency within a preset time period. The internal locations are at various points relative to the feed trough. Path reachability weights , According to each location point Path reachability weights Calculate the target chickens within a preset time period Internal path accessibility factor relative to the feed trough :

[0070] ;

[0071] Determine path reachability factors Is it greater than the preset maximum path reachability factor? Or less than the preset minimum path reachability factor If so, then Set to zero; based on path reachability factor Data on eating frequency Correction was performed to obtain corrected eating behavior data. , ,in, These are preset weighting coefficients used to control the strength of the impact of path accessibility on the correction of eating frequency.

[0072] In one example, before the decision module determines whether the probability of abnormal behavior exceeds a preset threshold, it also includes: obtaining the preset role label of the target chicken in the flock; and adjusting the preset threshold based on the role label.

[0073] In one example, the decision module adjusts the preset judgment threshold by: retrieving the preset role tag-judgment threshold mapping table based on the role tag to obtain the preset judgment threshold corresponding to the role tag.

[0074] In another example, the decision module adjusts the preset judgment threshold by: obtaining the probability of abnormal behavior of the target chicken character tag. Probability of abnormal character behavior This represents the average probability that chickens corresponding to a role tag deviate from normal chicken behavior patterns over a historical period; it also retrieves the pre-defined overall probability of abnormal chicken behavior in the flock. Probability of abnormal behavior in chicken flocks This represents the average probability that the entire flock deviates from the normal chicken behavior pattern over a historical period; based on the probability of abnormal behavior of a role. Probability of abnormal behavior in chicken flocks Adjust the preset judgment threshold :

[0075] ;

[0076] in, This indicates the adjusted preset judgment threshold. This represents a preset weighting coefficient used to control the probability of abnormal character behavior. For the preset judgment threshold The adjustment range satisfies .

[0077] In one example, the data processing module constructs a social network graph by: identifying the location coordinates of the target chicken and its neighboring chickens at each time point from a continuous set of spatial location images; determining the spatial proximity relationship between the target chicken and its neighboring chickens within a preset time period based on the location coordinates of the target chicken and its neighboring chickens at each time point; calculating the interaction strength between the target chicken and its neighboring chickens based on the spatial proximity relationship; and constructing a social network graph representing the interaction strength between the target chicken and its neighboring chickens based on the interaction strength.

[0078] Exemplary electronic devices

[0079] Figure 3 An electronic device according to an embodiment of this application is illustrated. The electronic device may be the mobile device itself, or a standalone device independent of it, which may communicate with the mobile device to receive collected input signals from it and send selected target driving behaviors to it.

[0080] Figure 3 A block diagram of an electronic device according to an embodiment of this application is illustrated.

[0081] like Figure 3 As shown, the electronic device includes one or more processors and memory.

[0082] A processor can be a central processing unit (CPU) or other form of processing unit with data processing and / or instruction execution capabilities, and can control other components in an electronic device to perform desired functions.

[0083] The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and a processor may execute the program instructions to implement the driving behavior decision-making methods of the various embodiments of this application described above, and / or other desired functions.

[0084] In one example, the electronic device may also include input devices and output devices, which are interconnected via a bus system and / or other forms of connection mechanism (not shown).

[0085] Of course, for the sake of simplicity, Figure 3 Only some of the components of the electronic device relevant to this application are shown in this illustration; components such as buses, input / output interfaces, etc., are omitted. In addition, the electronic device may include any other suitable components depending on the specific application.

[0086] Exemplary computer-readable media

[0087] Embodiments of this application may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps in the driving behavior decision-making methods according to various embodiments of this application described in the "Exemplary Methods" section above.

[0088] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0089] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the application to the necessity of employing the aforementioned specific details for implementation.

[0090] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0091] It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of this application.

[0092] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0093] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A chicken disease symptom image recognition method based on deep learning, characterized in that, include: Acquire a continuous set of spatial location images and a set of feeding images of the target chickens within a preset time period; Based on the continuous spatial location image set and the feeding image set, extract the spatial location relationship data between the target chicken and the feed trough, the feeding frequency data, and the social network graph representing the interaction intensity between the target chicken and neighboring chickens within the preset time period. Based on the spatial location relationship dataset, the eating frequency data is corrected to obtain eating behavior data after eliminating the influence of spatial layout; The feeding behavior data and the social network graph are input into a preset deep learning model to generate the probability of abnormal behavior of the target chickens within the preset time period. Determine whether the probability of the abnormal behavior exceeds a preset threshold. If so, determine that the target chicken has suspected disease behavior. The correction of the eating frequency includes: The spatial location relationship dataset is input into a preset path reachability weight distribution model relative to the feed trough within the chicken house to obtain the target chickens' reach within the preset time period. The internal locations are at various points relative to the feed trough. Path reachability weights , The path reachability weight distribution model divides the chicken house space into multiple grid units, and the reachability weight of each grid unit is calculated based on the path complexity between it and the feed trough. Based on the aforementioned location points Path reachability weights Calculate the target chickens during the preset time period Internal path accessibility factor relative to the feed trough , ; Determine the path reachability factor Is it greater than the preset maximum path reachability factor? Or less than the preset minimum path reachability factor If so, then Set to zero; According to the path reachability factor Regarding the frequency of eating data Correction was performed to obtain corrected eating behavior data. , ,in, These are preset weighting coefficients used to control the strength of the impact of path accessibility on the correction of eating frequency.

2. The deep learning-based image recognition method for chicken disease symptoms according to claim 1, characterized in that, Before determining whether the probability of the abnormal behavior exceeds a preset threshold, the following steps are also included: Obtain the preset role tags of target chickens within the flock; The preset judgment threshold is adjusted based on the character tag.

3. The deep learning-based image recognition method for chicken disease symptoms according to claim 2, characterized in that, The adjustment of the preset judgment threshold includes: Based on the character tag, a preset character tag-judgment threshold mapping table is retrieved to obtain the preset judgment threshold corresponding to the character tag.

4. The deep learning-based image recognition method for chicken disease symptoms according to claim 2, characterized in that, The adjustment of the preset judgment threshold includes: Obtain the probability of abnormal behavior of the target chickens corresponding to their predefined target chicken character tags. The probability of abnormal behavior of the character This represents the average probability that the chickens corresponding to the aforementioned role tag deviate from the normal chicken behavior pattern during a historical period; Obtain the preset probability of abnormal behavior in the entire flock of chickens. The probability of abnormal behavior in the chicken flock This represents the average probability that the entire flock deviates from the normal chicken behavior pattern during a historical period. Based on the aforementioned abnormal behavior probability Probability of abnormal behavior in chicken flocks Adjust the preset judgment threshold : ； in, This indicates the adjusted preset judgment threshold, the This represents a preset weighting coefficient used to control the probability of abnormal behavior of the character. For the preset judgment threshold The adjustment range satisfies .

5. The deep learning-based image recognition method for chicken disease symptoms according to claim 1, characterized in that, The extraction of the social network graph includes: Identify the position coordinates of the target chicken and its neighboring chickens at each time point from the continuous spatial location image set; Based on the location coordinates of the target chicken and its neighboring chickens at each time point, the spatial proximity relationship between the target chicken and its neighboring chickens within the preset time period is determined. The interaction intensity between the target chicken and each of its neighboring chickens is calculated based on the spatial proximity relationship. Based on the interaction strength, a social network graph representing the interaction strength between the target chicken and each of its neighboring chickens is constructed.

6. A deep learning-based image recognition system for chicken disease symptoms, characterized in that, include: The data acquisition module is used to acquire a continuous set of spatial location images and a set of feeding images of the target chickens within a preset time period; The data processing module is used to extract, based on the continuous spatial location image set and the feeding image set, a dataset of the spatial location relationship between the target chicken and the feed trough within the preset time period, feeding frequency data, and a social network graph representing the interaction intensity between the target chicken and neighboring chickens. The deviation correction module is used to correct the eating frequency data according to the spatial location relationship dataset to obtain eating behavior data after eliminating the influence of spatial layout. An anomaly probability prediction module is used to input the feeding behavior data and the social network graph into a preset deep learning model to generate the anomaly probability of the target chicken's behavior within the preset time period. The decision module is used to determine whether the probability of the abnormal behavior exceeds a preset judgment threshold. If so, it is determined that the target chicken has suspected disease behavior. The correction of the eating frequency includes: The spatial location relationship dataset is input into a preset path reachability weight distribution model relative to the feed trough within the chicken house to obtain the target chickens' reach within the preset time period. The internal locations are at various points relative to the feed trough. Path reachability weights , The path reachability weight distribution model divides the chicken house space into multiple grid units, and the reachability weight of each grid unit is calculated based on the path complexity between it and the feed trough. Based on the aforementioned location points Path reachability weights Calculate the target chickens during the preset time period Internal path accessibility factor relative to the feed trough , ; Determine the path reachability factor Is it greater than the preset maximum path reachability factor? Or less than the preset minimum path reachability factor If so, then Set to zero; According to the path reachability factor Regarding the frequency of eating data Correction was performed to obtain corrected eating behavior data. , ,in, These are preset weighting coefficients used to control the strength of the impact of path accessibility on the correction of eating frequency.

7. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method as described in any one of claims 1 to 5.

8. A computer storage medium storing computer-executable instructions thereon, characterized in that: When the computer-executable instructions are executed by a processor, they implement the steps of the method as described in any one of claims 1 to 5.

Images