An anti-interference sound analysis system and method for pig farms

Abstract

This invention relates to the field of information processing technology, specifically disclosing an anti-interference sound analysis system and method for pig farms. The system includes: a sound acquisition module for real-time acquisition of sounds within the farm and monitoring whether the decibel value exceeds a sound threshold; a voiceprint analysis module for identifying and analyzing pig calls upon receiving a start command; and an image recognition module for analyzing acquired video images to identify the presence of dogs in the images; a logic control module for triggering the image recognition module to execute a barking recognition mode when a dog is present in the image and the sound decibel value exceeds the threshold; if the image recognition module determines that a dog is barking in barking recognition mode, the logic control module further generates a blocking command to pause the voiceprint analysis module's processing of the current sound data. The technical solution of this invention can effectively reduce interference and improve the reliability of pig sound recognition at a low cost.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and in particular to an anti-interference sound analysis system and method for pig farms. Background Technology

[0002] With the rapid development of large-scale pig farming technology, the informatization transformation of farms has become an industry trend. By deploying monitoring systems within farms to capture images of pigs and collect their audio data, and through image analysis and voiceprint recognition technology, scientific farming management can be achieved, including monitoring the health status and estrus cycle of pigs.

[0003] However, during actual system operation, it was found that the accuracy of sound analysis was highly susceptible to interference from the complex acoustic environment of farms. Farm environments are not silent; mechanical noise, wind noise, and the sounds of non-target animals all hinder the extraction of pig sounds. In such a multi-sounding scenario, the success rate of sound recognition often drops significantly. Frequent false alarms and missed alarms lead to the system issuing frequent invalid warnings, severely impacting user experience. This causes users to perceive the function as impractical and refuse to continue using it, increasing the difficulty of subsequent promotion and application.

[0004] Current solutions often focus on audio signal processing itself, considering the development of specialized filtering algorithms or source separation technologies. Relying solely on audio algorithms for separation not only significantly increases development costs and computing power requirements, but also results in low fault tolerance, long development and testing times, and missed market opportunities.

[0005] Therefore, how to utilize the existing multimodal sensing capabilities of farms to develop a pig sound recognition assistance technology that can effectively reduce interference, is low-cost, and highly reliable has become an urgent technical problem to be solved in the field of smart farming. Summary of the Invention

[0006] One of the objectives of this invention is to provide an anti-interference sound analysis system for pig farms, which can effectively reduce interference and improve the reliability of pig sound recognition at low cost.

[0007] To solve the above-mentioned technical problems, this application provides the following technical solution:

[0008] An anti-interference sound analysis system for pig farms includes:

[0009] The sound acquisition module is used to collect environmental sound data in the farm in real time.

[0010] The voiceprint analysis module is used to extract features from sound data after receiving a start command, and to identify and analyze the sounds of pigs.

[0011] The sound acquisition module is also used to monitor whether the decibel value of the sound exceeds the preset sound threshold;

[0012] Also includes:

[0013] The image acquisition module is used to acquire video images of the farm's interior and surrounding area.

[0014] The image recognition module is used to analyze the captured video images and identify whether there are dogs in the images;

[0015] The logic control module is used to trigger the image recognition module to execute the barking recognition mode when a dog is present in the image and the sound decibel value exceeds the threshold.

[0016] The image recognition module is also used to identify the dog's feeding status in barking recognition mode, including tethered and free-roaming status; if a dog is detected and it is tethered, it is determined whether the dog is barking; if the dog is determined to be barking, the logic control module is also used to generate a blocking command to pause the voiceprint analysis module's processing of the current sound data.

[0017] Furthermore, if the image recognition module detects the presence of a dog that is tethered, it extracts the dog's local features, including the head tilt position and the shape of the leash. If the dog's head is tilted and the leash is taut, it is determined that the dog is barking.

[0018] Furthermore, when the image recognition module detects the presence of a dog in a free-range state, it tracks its movement trajectory; if it detects that the dog moves from the center area of ​​the farm to the boundary area within a first preset time and its speed drops to 0, and then detects that its head is tilting up or down, it determines that the dog is barking.

[0019] Furthermore, the image recognition module is also used to determine a non-interference item if it detects that a dog has moved from the center area of ​​the breeding farm to the boundary area within a first preset time until it disappears.

[0020] If the dog is determined to be neither barking nor disturbing, the logic control module is also used to control the image recognition module to determine whether there are people in the farm. If the recognition result is that there are people, a blocking command is generated; if the recognition result is that there are no people, a starting command is generated.

[0021] Furthermore, the image recognition module is also used to extract the rope area connecting the dog's collar and the fixing point, extract the outline of the rope through an edge detection algorithm, calculate the average curvature of the outline, and if the average curvature is less than a preset straightening threshold, it is marked as the rope is straightened; the image recognition module is also used to calculate the movement trajectory of the dog's head in the vertical direction, and if the key points of the head are detected to swing up and down repeatedly within a second preset time, and the amplitude of the swing exceeds a preset pitch threshold, it is determined that the dog's head is in a pitch state;

[0022] If the condition is met only when the head is tilted up but the rope is not taut, or only when the rope is taut but the head is not tilted up, it is considered a non-interference state.

[0023] Furthermore, it also includes a cloud server, and the image recognition module has a pre-stored list of non-disturbing dogs;

[0024] The image recognition module is also used to identify the breed of the dog after detecting its presence. If the dog is not in the preset list, the logic control module is also used to determine the current area of ​​the dog. If the dog is located in the entrance and exit area of ​​the farm, a request is sent to the cloud server to download or update the barking recognition algorithm package.

[0025] Furthermore, the logic control module is also used to collect operational data of the barking recognition mode. If the frequency of barking interference detected within a preset statistical period is lower than a preset active threshold, the local barking recognition mode algorithm package is unloaded or put into hibernation.

[0026] Furthermore, the logic control module is also used to count the time of each bark of the dog, determine whether the barking time exceeds a third preset time, and if it exceeds the preset time, obtain the recognition result of the image recognition module on the state of the pigs inside the farm.

[0027] The logic control module is also used to count the number of pigs with abnormal status based on the identification results. If the number of pigs with abnormal status is lower than the number threshold, the corresponding pigs with abnormal status are marked.

[0028] The logic control module is also used to obtain the recognition results of the pig status inside the farm from the image recognition module when the dog is not barking. If the pig with abnormal status is a marked pig, the priority of the sound data analysis at the time corresponding to the abnormal status of the pig is reduced when the voiceprint analysis module is analyzing.

[0029] The second objective of this invention is to provide an anti-interference sound analysis method for pig farms, comprising the following:

[0030] S1. Real-time acquisition of video images inside and around the farm, as well as environmental sound data within the farm;

[0031] S2. Analyze the captured video images to identify whether there are dogs in the images; if there are no dogs, or if the identified dogs belong to a preset list, proceed to step S7.

[0032] If a dog not on the preset list is identified, the area where the dog is currently located is identified. If the dog is located in the entrance and exit area of ​​the farm, the algorithm package for the barking recognition mode is downloaded or updated; and the process jumps to step S3; if the dog is not in the entrance and exit area of ​​the farm, the process jumps to step S7.

[0033] S3. Monitor whether the decibel value of the sound exceeds the preset sound threshold. If it does not exceed the sound threshold, proceed to step S7.

[0034] If the sound threshold is exceeded, the barking recognition mode is triggered, and the process proceeds to step S4.

[0035] S4. Identify whether the dog is barking. If yes, proceed to step S5; if not, proceed to step S6.

[0036] S5. Generate a blocking command to pause the processing of the current audio data;

[0037] S6. Identify whether there are people in the farm. If the identification result is that there are people, generate a blocking command; if the identification result is that there are no people, proceed to step S7.

[0038] S7. Recognize and analyze the currently collected sound data of pig calls.

[0039] Furthermore, step S4 specifically includes: identifying the dog's feeding status, which includes tethered and free-range;

[0040] When the dog is on a leash, extract local features of the dog, including head tilt and leash shape. Determine whether the dog is barking based on head tilt and leash shape. If the dog is not barking, proceed to step S5. If the dog is not barking, proceed to step S6.

[0041] When the dog is free-range, its movement trajectory is tracked. If the dog is detected to move from the center of the farm to the boundary area within a first preset time and its speed drops to 0, and then its head is detected to be tilting up or down, it is determined that the dog is barking and the process proceeds to step S5. If the dog is detected to move from the center of the farm to the boundary area within a first preset time until it disappears, it is determined to be a non-interference item and the process proceeds to step S6.

[0042] Current solutions to the sound interference problem typically focus on audio signal processing itself, considering the development of specialized filtering algorithms or source separation techniques. However, sound analysis reveals that mechanical noise and wind noise have relatively low interference frequencies with pig sounds, while the sounds of non-target animals significantly interfere with pig sounds and pigs readily react to these sounds. Further analysis of farm surveillance videos and sound data shows that in many small and medium-sized farms, especially those in remote areas, dogs are often kept as guard dogs for security and protection. When strangers approach or other unusual situations occur, the dogs bark; pigs, upon hearing the barking, may react with stress or join in with grunting. In this case, the collected audio data becomes a mixture of barking and pig grunting, resulting in a very low signal-to-noise ratio. Alternatively, barking may be mistaken for pig grunting, hindering effective feature extraction and analysis, leading to false alarms. This solution identifies dog barking as the primary interference factor. Dog barking overlaps with pig grunting in certain frequency bands, and given the diverse breeds of dogs, relying solely on audio algorithms for separation significantly increases development costs and computational demands. Furthermore, it suffers from low fault tolerance, lengthy development and testing times, and may even lead to missed market opportunities due to the need for collaboration between different departments involved in image and sound processing. This invention takes a different approach, building upon existing image recognition functionality with new calibration to construct a sound-image linkage anti-interference mechanism. The system uses a sound acquisition module to monitor decibel levels and trigger image recognition. The logic control module generates a blocking command when a dog is detected barking, pausing voiceprint analysis. Utilizing the existing multimodal sensing capabilities of the farm avoids the development of complex blind source separation algorithms, resulting in lower development costs, higher recognition accuracy, and rapid deployment using existing hardware. This effectively solves the problem of frequent false alarms caused by environmental noise. Attached Figure Description

[0043] Figure 1 This is a logic block diagram of a second embodiment of an anti-interference sound analysis system for pig farms. Detailed Implementation

[0044] The following detailed description illustrates the specific implementation method:

[0045] Example 1

[0046] This embodiment provides an anti-interference sound analysis system for pig farms, including a sound acquisition module, an image acquisition module, an image recognition module, a logic control module, and a voiceprint analysis module.

[0047] The sound acquisition module is used to collect environmental sound data in the farm in real time and monitor whether the decibel value of the sound exceeds the preset sound threshold.

[0048] The image acquisition module is used to acquire video images of the farm's interior and surrounding areas; in this embodiment, the image acquisition module includes a visible light camera covering key areas of the farm.

[0049] The image recognition module is used to analyze the captured video images and identify whether there are dogs in the images;

[0050] If a dog is present, the logic control module triggers the image recognition module to execute a barking recognition mode when the sound collected by the sound acquisition module exceeds a sound threshold, including:

[0051] The image recognition module identifies the dog's feeding status and whether there are people inside the breeding farm. The feeding status includes tethered (with a leash) and free-range.

[0052] When the image recognition module detects the presence of a dog that is leashed, it extracts local features of the dog, including head tilt and the shape of the leash. The image recognition module then determines whether the dog is barking based on these features. In this embodiment, the determination is based on the following: if the dog's head is tilted forward and the leash is taut, the dog is determined to be barking.

[0053] When the image recognition module detects the presence of a dog in a free-range state, it tracks its movement trajectory. If the dog is detected to make a rapid movement (rush) from the center area of ​​the farm towards the boundary area within a first preset time (e.g., within 2 seconds), and its speed suddenly drops to 0 (sudden stop), and then its head is detected to be in a tilting state, it is determined that the dog is barking. If the dog is detected to make a rapid movement from the center area of ​​the farm towards the boundary area until it disappears within a first preset time, it is determined to be a non-interference item. In this embodiment, the center area refers to the area where pigs are raised.

[0054] Specifically,

[0055] When determining the dog's status, the image recognition module extracts the rope area connecting the dog's collar and the anchor point, and extracts the rope's outline using an edge detection algorithm. The mean curvature of the outline is calculated; if the mean curvature is less than a preset straightening threshold, indicating the rope is straight, it is marked as straight. When identifying head tilt, the image recognition module first uses keypoint detection technology to locate the dog's nose tip and the center point of its head. The positional changes of these two keypoints are tracked in real-time across consecutive video frames, calculating the head's vertical trajectory. If the head keypoints are detected to exhibit rapid, continuous up-and-down oscillations within a second preset time (e.g., within 1 second) (i.e., the Y-axis coordinate shows alternating peaks and troughs), and the amplitude of the oscillation (displacement pixel value or pitch angle change rate) exceeds a preset pitch threshold, then the dog's head is determined to be in a tilt state.

[0056] The image recognition module only determines that a dog is in a state of intense barking when both conditions are met: the leash is taut and the head is tilted up or down. If only the head is tilted up but the leash is not taut (e.g., the dog is yawning), or only the leash is taut but the head is not tilted up or down (e.g., the dog is struggling but not barking), it is considered a non-interference state. In this case, the system continues to execute the subsequent personnel detection process.

[0057] If it is determined that the dog is barking, the logic control module is also used to generate a blocking command to pause the processing of the current sound data by the voiceprint analysis module in order to avoid false alarms. In this embodiment, the processing of the current sound data by the voiceprint analysis module is resumed after the barking ends and after a certain delay, such as 1 minute.

[0058] If the dog is determined to be neither barking nor causing disturbance, the logic control module also controls the image recognition module to determine whether there are people in the farm. If the recognition result indicates that there are people, a blocking command is generated; if the recognition result indicates that there are no people, a starting command is generated. In this embodiment, "farm" refers to personnel located in the pig farming area who can visually observe the pigs.

[0059] The voiceprint analysis module is used to extract features from sound data after receiving a start command, and to identify and analyze the sounds of pigs, such as whether they are abnormal sounds of pigs, including coughing and screaming.

[0060] This embodiment is specifically optimized for interference scenarios involving guard dogs in breeding farms, improving the practicality of sound anti-interference analysis. For the identification of leashed dogs, this embodiment abandons the conventional mouth recognition scheme and instead extracts the head tilt position. This is because monitoring cameras in breeding farms are usually installed at a high position, presenting a downward angle. From this downward angle, the detailed features of the dog's mouth are easily obscured or blurred, resulting in a low detection success rate; however, the overall head tilt movement is significant from a downward angle, leading to a high detection success rate. This embodiment also uses an edge detection algorithm to extract the leash contour and calculate its curvature. If a tilted head position is detected and the leash is taut, it is determined that the dog is barking, improving the accuracy of identification.

[0061] For free-roaming dogs, this embodiment utilizes behavioral trajectory analysis. When a dog is detected to move from the central area to the boundary area within a short period of time and then suddenly stop, followed by head tilting, it indicates that the dog has a tendency to move outwards to be driven away, and the likelihood of barking is extremely high, thus accurately determining the interference state. At the same time, if the dog moves but eventually disappears, its sound interferes with the data collection due to the increased distance, and the system determines it as a non-interference item.

[0062] This embodiment also introduces personnel detection as a secondary judgment logic. If it is determined that the dog is not barking or interfering, it further determines whether there are personnel in the farm. If the identification result is that there are people, a blocking command is generated, because at this time personnel can observe and understand the status of the pigs themselves without relying on the voice recognition system for early warning, thereby further avoiding invalid sound analysis and false alarms.

[0063] This embodiment can identify noise sources through image analysis, avoiding the need to deploy high-cost, high-computing-power audio AI filtering algorithms. This not only reduces hardware costs but also significantly lowers the false alarm rate caused by dog ​​barking, addressing the pain point of users abandoning the system due to frequent false alarms and improving product usability and user experience.

[0064] Example 2

[0065] like Figure 1 As shown, the difference between this embodiment and Embodiment 1 is that this embodiment also includes a cloud server, and the image recognition module, logic control module and voiceprint analysis module are integrated in a local edge processing box.

[0066] The image recognition module also has a pre-stored list of non-disruptive dog breeds, such as companion dogs that are not prone to barking.

[0067] After the image recognition module detects the presence of a dog, it uses a lightweight classification network to identify the dog's breed. If the identified dog belongs to a breed within a preset list, it is assumed that the dog will not cause significant acoustic interference. If a dog not on the preset list is identified, the logic control module obtains the camera's installation location metadata to determine the dog's current location. If the dog is located in the entrance / exit area of ​​the breeding farm (where dogs are easily disturbed by external factors), a request is sent to the cloud server to download or update the barking recognition algorithm package.

[0068] The barking recognition algorithm package in this embodiment is a dedicated program module integrated into the image recognition module. It is used to execute the barking state recognition logic described in Embodiment 1, including: recognizing the dog's feeding state, extracting the head tilt state and leash shape features in the tethered state to determine whether it is barking, tracking the movement trajectory in the free-roaming state and combining it with head movements to determine whether it is barking, and determining non-interference items.

[0069] The logic control module is also used to collect operational data on the barking recognition mode. If the frequency of dog barking interference is lower than the preset activity threshold within a preset statistical period (e.g., within one month), it indicates that the dogs in the area are relatively quiet. In this case, the local barking recognition mode algorithm package is unloaded or put into hibernation to release computing resources and reduce system power consumption.

[0070] The logic control module is also used to count the duration of each dog bark and determine whether the barking time exceeds a third preset time (e.g., 5 seconds). If it exceeds the preset time, the module obtains the recognition results of the pigs' status inside the farm from the image recognition module. In this embodiment, if the pigs are agitated or squealing, it is considered an abnormal state. In this embodiment, the judgment of the pigs' squealing state uses the same head pitch state recognition as that for dogs. When recognizing agitation, the center point of the pig's body contour is extracted, and the displacement vector of the center point is calculated between consecutive frames. If the detected movement speed of the center point exceeds a preset agitation speed threshold, and the number of times the direction of the displacement vector changes exceeds a preset frequency threshold within a fourth preset time (i.e., exhibiting a disorderly and chaotic state), then the pig is determined to be in an agitated state. The use of centroid tracking + speed / direction determination, a non-complex motion capture algorithm, can save edge computing power.

[0071] The logic control module is also used to count the number of pigs with abnormal status based on the identification results. If the number of pigs with abnormal status is lower than the number threshold (2 in this embodiment, and in other embodiments, it can be set proportionally according to the total number of pigs raised), the corresponding pigs with abnormal status are marked.

[0072] The logic control module is also used to, when the dog is not barking, obtain the recognition results of the pig status inside the farm from the image recognition module. If the abnormal pig is a marked pig, the priority of the sound data analysis at the time corresponding to the abnormal pig status is reduced during the voiceprint analysis module analysis. In this embodiment, the priority is reduced to the lowest level.

[0073] In this embodiment, a list of non-interference dogs is pre-stored locally. The algorithm package for the barking recognition mode is only requested from the cloud server to be downloaded or updated when a dog not on the list is detected and is located in an area of ​​the farm entrance or exit that is susceptible to external interference. This ensures that updates are only initiated when the farm actually houses breeds prone to barking and is in an active area, avoiding resource consumption and interference in other scenarios where dogs are not kept or where quiet breeds are kept. Secondly, the logic control module collects operational data on the barking recognition mode. If the frequency of barking interference is detected to be below the active threshold within a preset period, the local algorithm package is automatically uninstalled or put into hibernation. Timely suspension of related functions when the interference frequency is too low avoids false interference with normal sound acquisition and effectively reduces system recognition latency, freeing up computing resources. Regarding the potential stress response of pigs upon hearing barking, this embodiment distinguishes between individual stress and group abnormalities by analyzing the behavioral consistency of the pig herd. When only a few pigs exhibit restlessness or barking, they are identified as susceptible to individual stress, marked, and their data is automatically downgraded in the voiceprint analysis. Due to cost constraints and limited edge computing resources, dynamic allocation of computing power can avoid in-depth processing of invalid interference signals, reduce the load on edge computing nodes, and improve the overall operating efficiency and recognition accuracy of the system.

[0074] Example 3

[0075] The difference between this embodiment and Embodiments 1 and 2 is that this embodiment provides an anti-interference sound analysis method for pig farms, including the following:

[0076] S1. Real-time acquisition of video images inside and around the farm, as well as environmental sound data within the farm;

[0077] S2. Analyze the captured video images to identify whether there are dogs in the images; if there are no dogs, or if the identified dogs belong to a preset list, proceed to step S7.

[0078] If a dog not on the preset list is identified, the camera's installation location metadata is obtained, and the dog's current location is determined. If the dog is located in the entrance / exit area of ​​the farm, the cloud server sends a request to download or update the barking recognition algorithm package; and proceeds to step S3; if the dog is not in the entrance / exit area of ​​the farm, proceeds to step S7.

[0079] S3. Monitor whether the decibel value of the sound exceeds the preset sound threshold. If it does not exceed the sound threshold, proceed to step S7.

[0080] If the sound threshold is exceeded, the barking recognition mode is triggered, and the process proceeds to step S4.

[0081] S4. Identify whether the dog is barking. If yes, proceed to step S5; if not, proceed to step S6.

[0082] Specifically, it includes:

[0083] Identify the dog's living status, including whether it is kept on a leash or allowed to roam freely;

[0084] When a dog is detected to be in a tethered state, local features of the dog are extracted, including head tilt and tether shape. Based on head tilt and tether shape, it is determined whether the dog is barking. If the dog is not barking, proceed to step S5. If the dog is not barking, proceed to step S6.

[0085] When a dog is detected to be free-roaming, its movement trajectory is tracked. If a dog is detected to move rapidly from the center of the farm towards the boundary within a first preset time period, and its speed suddenly drops to 0, and then its head is detected to be tilting up or down, it is determined that the dog is barking, and the process proceeds to step S5. If a dog is detected to move rapidly from the center of the farm towards the boundary within a first preset time period until it disappears, it is determined to be a non-interference item, and the process proceeds to step S6.

[0086] S5. Generate a blocking command to pause the processing of the current audio data;

[0087] S6. Identify whether there are people in the farm. If the identification result is that there are people, generate a blocking command; if the identification result is that there are no people, proceed to step S7.

[0088] S7. Recognize and analyze the currently collected sound data of pig calls.

[0089] The above are merely embodiments of the present invention. The invention is not limited to the fields covered by these embodiments. Commonly known structures and characteristics in the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are able to access all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. An anti-interference sound analysis system for pig farms, comprising: The sound acquisition module is used to collect environmental sound data in the farm in real time. The voiceprint analysis module is used to extract features from sound data after receiving a start command, and to identify and analyze the sounds of pigs. Its distinguishing feature is that the sound acquisition module is also used to monitor whether the decibel value of the sound exceeds a preset sound threshold; Also includes: The image acquisition module is used to acquire video images of the farm's interior and surrounding area. The image recognition module is used to analyze the captured video images and identify whether there are dogs in the images; The logic control module is used to trigger the image recognition module to execute the barking recognition mode when a dog is present in the image and the sound decibel value exceeds the threshold. The image recognition module is also used to identify the dog's feeding status in the barking recognition mode, including tethered and free-roaming; if a dog is detected and it is tethered, it is determined whether the dog is barking; if the dog is determined to be barking, the logic control module is also used to generate a blocking command to pause the voiceprint analysis module's processing of the current sound data. The image recognition module is also used to determine a non-interference item if it detects that a dog has moved from the center area of ​​the breeding farm to the boundary area within a first preset time until it disappears. If the dog is determined to be neither barking nor disturbing, the logic control module is also used to control the image recognition module to determine whether there are people in the farm. If the recognition result is that there are people, a blocking command is generated; if the recognition result is that there are no people, a starting command is generated. It also includes a cloud server, and the image recognition module has a pre-stored list of non-disruptive dogs; The image recognition module is also used to identify the breed of the dog after detecting its presence. If the dog is not in the preset list, the logic control module is also used to determine the current area of ​​the dog. If the dog is located in the entrance and exit area of ​​the farm, a request is sent to the cloud server to download or update the barking recognition algorithm package.

2. The anti-interference sound analysis system for pig farms according to claim 1, characterized in that: If the image recognition module detects the presence of a dog that is tethered, it extracts the dog's local features, including the head tilt position and the shape of the leash. If the dog's head is tilted and the leash is taut, it is determined that the dog is barking.

3. The anti-interference sound analysis system for pig farms according to claim 2, characterized in that: When the image recognition module detects the presence of a dog in a free-range state, it tracks the dog's movement trajectory. If it detects that the dog moves from the center of the farm towards the boundary area within a first preset time and its speed drops to 0, and then detects that its head is tilting up or down, it determines that the dog is barking.

4. The anti-interference sound analysis system for pig farms according to claim 3, characterized in that: The image recognition module is also used to extract the rope area connecting the dog collar and the fixed point, extract the outline of the rope through the edge detection algorithm, calculate the average curvature of the outline, and if the average curvature is less than the preset straightening threshold, it is marked as the rope is straightened; the image recognition module is also used to calculate the movement trajectory of the dog's head in the vertical direction, and if the key points of the head are detected to swing up and down repeatedly within a second preset time, and the amplitude of the swing exceeds the preset pitch threshold, it is determined that the dog's head is in a pitch state; If the condition is met only when the head is tilted up but the rope is not taut, or only when the rope is taut but the head is not tilted up, it is considered a non-interference state.

5. The anti-interference sound analysis system for pig farms according to claim 4, characterized in that: The logic control module is also used to collect statistics on the operation data of the barking recognition mode. If the frequency of dog barking interference is detected to be lower than the preset activity threshold within the preset statistical period, the local barking recognition mode algorithm package is unloaded or put into hibernation.

6. The anti-interference sound analysis system for pig farms according to claim 5, characterized in that: The logic control module is also used to count the time of each bark of the dog, determine whether the barking time exceeds a third preset time, and if it exceeds the preset time, obtain the recognition result of the image recognition module on the state of the pigs inside the farm. The logic control module is also used to count the number of pigs with abnormal status based on the identification results. If the number of pigs with abnormal status is lower than the number threshold, the corresponding pigs with abnormal status are marked. The logic control module is also used to obtain the recognition results of the pig status inside the farm from the image recognition module when the dog is not barking. If the pig with abnormal status is a marked pig, the priority of the sound data analysis at the time corresponding to the abnormal status of the pig is reduced when the voiceprint analysis module is analyzing.

7. A method for analyzing interference-resistant sound in pig farms, characterized in that, Includes the following: S1. Real-time acquisition of video images inside and around the farm, as well as environmental sound data within the farm; S2. Analyze the captured video images to identify whether there are dogs in the images; if there are no dogs, or if the identified dogs belong to a preset list, proceed to step S7. If a dog not on the preset list is identified, identify the area where the dog is currently located. If the dog is located in the entrance or exit area of ​​the farm, download or update the barking recognition algorithm package. Then proceed to step S3; If not in the entrance / exit area of ​​the farm, proceed to step S7; S3. Monitor whether the decibel value of the sound exceeds the preset sound threshold. If it does not exceed the sound threshold, proceed to step S7. If the sound threshold is exceeded, the barking recognition mode is triggered, and the process proceeds to step S4. S4. Identify whether the dog is barking. If so, proceed to step S5. If the dog is not barking, proceed to step S6; S5. Generate a blocking command to pause the processing of the current audio data; S6. Identify whether there are people in the farm. If the identification result is that there are people, generate a blocking command; if the identification result is that there are no people, proceed to step S7. S7. Recognize and analyze the currently collected sound data of pig calls.

8. The anti-interference sound analysis method for pig farms according to claim 7, characterized in that: Step S4 specifically includes: identifying the dog's feeding status, which includes tethered and free-range; When the dog is tethered, local features of the dog are extracted, including head tilt and leash shape. The dog is judged to be barking based on head tilt and leash shape. If the dog is not barking, proceed to step S5. If the dog is not barking, proceed to step S6. When the dog is free-roaming, its movement trajectory is tracked. If the dog is detected to move from the center of the farm to the boundary within a first preset time and its speed drops to 0, and then its head tilts, it is determined that the dog is barking, and proceed to step S5. If the dog is detected to move from the center of the farm to the boundary within a first preset time until it disappears, it is determined to be a non-interference item, and proceed to step S6.

Images