Sow maternal behavior intelligent identification method based on computer vision and deep learning

Abstract

The invention provides a sow maternal behavior intelligent identification method based on computer vision and deep learning, and relates to the technical field of sow maternal behavior intelligent identification methods, which associates sow maternal behavior visual identification with factors such as health conditions, emotions and postures of sows, and establishes a dynamic prediction database by taking time as a node, so as to realize sow maternal behavior identification. The maternal behaviors are continuously predicted and trained, and the accuracy of maternal behavior recognition is further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent identification methods for sow maternal behavior, and in particular to a system and method for intelligent identification of sow maternal behavior based on computer vision and deep learning. Background Art

[0002] Traditional pig farming requires breeders to obtain real-time status information of pigs to monitor abnormal situations. The scale and number of pig farming in China are constantly expanding, and the informatization of pig farming is an important direction for the future supervision of pig farming.

[0003] The maternal behavior of sows refers to the behaviors that occur frequently before and after the birth of piglets, including carrying grass to build nests, calling piglets, breastfeeding, licking, etc.; it also includes negative behaviors such as infanticide; therefore, maternal behavior has a significant impact on the survival rate and growth rate of piglets during lactation.

[0004] Computer vision technology and deep learning have been widely used in pig behavior recognition, producing various algorithm models for pig behavior recognition.

[0005] However, in the prior art, the above algorithm model only performs single recognition on a certain behavior of a pig, and does not study the relationship between various maternal behaviors of pigs. Therefore, the recognition of the above individual behavior is often single and one-sided. Since various internal factors of pigs are not taken into account, there are misjudgments in image recognition, and it does not continuously reflect the maternal behavior of sows, and has no reference significance for subsequent recognition.

[0006] Therefore, the present application provides a sow maternal behavior intelligent identification system and method based on computer vision and deep learning to solve the above problems. Summary of the invention

[0007] The present invention aims to provide an intelligent recognition method for maternal behavior of sows based on computer vision and deep learning, so as to solve the technical problem that the existing maternal behavior recognition is one-sided, single, and isolated, and each behavior is extracted and identified separately, resulting in misjudgment of various behavior recognitions.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions: A method for intelligently identifying maternal behavior of sows based on computer vision and deep learning, the method comprising: Step S1: Set the target breeding area, put the sows to be born in the breeding pen, set up multiple cameras, video storage devices, servers, and thermal imaging devices; the server uses computer vision CV algorithm, machine learning ML algorithm and OpenPose algorithm to analyze the sow information, and upload the sow information to the server; the sow information includes the sow size, sow temperature sensory information, sow skin color, and recognize the sow posture, sow face, mouth, limbs and abdomen and each nipple of the sow according to the sow image; Step S2: When it is detected that the sow is in a continuous lying state and multiple piglets appear in the abdomen of the sow, a pig target detection algorithm based on lightweight YOLOv4 is quickly used to detect each piglet and number each piglet; Step S3: Identify the movement characteristics of each piglet, identify the mouth of the piglet, identify whether the mouth of the piglet is located on the nipple of the sow, identify whether the piglet is active, and identify whether there is liquid on the body of the piglet; cooperate with the thermal imaging equipment to identify the temperature characteristics of the piglet, and identify whether the piglet has a breast-pushing action based on the thermal imaging image and the OpenPose algorithm; Step S4: Identify the position of the sow's mouth and determine whether the tongue of the sow's mouth is located on the piglet's body; Step S5: When the tongue of the sow's mouth is located on the piglet's body, identify whether the piglet's body has body temperature, and compare the thermal imaging image of the piglet with the original image to determine whether there is a defect in the piglet's body image and whether there is liquid on the piglet's body; Step S6: When the piglet's body temperature is within the normal range, the piglet's body image is complete, and there is no liquid in the piglet's body, the piglet is judged to be in a normal state; when the sow's tongue is located on the piglet, the sow's behavior is judged to be licking behavior; when the piglet's mouth is located on the sow's nipple and the piglet's body temperature is within the normal range, the sow's behavior is judged to be feeding behavior; when the piglet's body part is not located on the sow's nipple, the piglet's body temperature is lower than the normal value, and the piglet is still for a long time, the piglet is judged to be in an inactive state; when the sow's tongue is located on the piglet, the piglet is in a normal state, the piglet has a normal body temperature, and the piglet's body image is missing compared to the initial image, it is judged that the sow has committed infanticide; The actual maternal behavior is compared with the maternal behavior identified by the above model algorithm. If the two are inconsistent, the model algorithm is corrected and trained until the trained model result is close to the logical judgment result.

[0009] Furthermore, it also includes an audio collection module, which includes a sow audio collection module and a piglet audio collection module, which respectively collect audio of the sow and different piglets, and combine image recognition to judge the maternal behavior of the sow.

[0010] Furthermore, the posture of the sow is identified through video, and the posture of a specific sow at different times after parturition is counted through the time module to establish a dynamic prediction database to assist computer vision in identifying the sow's lactation posture.

[0011] Furthermore, the frequency of spinal and abdominal line fluctuations, the breathing frequency of the pig's snout, the sneezing frequency and the excretion frequency of the pig are collected and monitored, and the above behaviors are counted using computer video recognition to establish a dynamic prediction database to assist computer vision in identifying the maternal behavior of the sow.

[0012] Furthermore, the sow's facial expressions, ear postures and white-of-the-eye areas are identified to determine the sow's emotional state, and the emotional data is matched with the sow's maternal behavior. The above information is stored in a database, the model is trained, and a dynamic prediction database is established to assist computer vision recognition in calibration.

[0013] Furthermore, for the sow's face, by identifying the position of the mouth and whites of the eyes, and the movement trajectory of the whites of the eyes, the sow's expression can be determined to see whether she is frightened or dull. By analyzing the ear posture, the frequency of ear swinging, the position of the ears, and the swing amplitude, it can be determined whether the sow is in an angry or relaxed emotional state, and a dynamic prediction database can be established to assist computer vision in identifying maternal behavior.

[0014] Furthermore, the number and gender of different piglets are counted, and the interaction behaviors between sows and all piglets at different time periods and times are counted to establish a dynamic prediction database to assist computer vision in identifying maternal behavior.

[0015] Compared with the prior art, the present invention has at least the following beneficial effects: In the scheme of the present application, the present invention adopts the method of monitoring and analyzing all piglets, identifying several maternal behaviors of the sow that exist simultaneously, combining the sow's behavior with the number of piglets, the gender of the piglets, the sow's own emotions, and health status, and at the same time, statistically analyzing the changes in the sow's licking, suckling and infanticide behaviors at different times in a timeline manner, establishing a dynamic database with time as a node, and further calibrating and predicting the computer vision recognition results, providing a reference basis for the subsequent identification of the sow's maternal behavior, and improving the accuracy and comprehensiveness of the identification of the sow's maternal behavior.

[0016] At the same time, for the identification of specific breastfeeding behavior, the present invention subdivides the sow's posture, takes time as a node, counts the changes in the sow's posture at each time, establishes a dynamic database, predicts the computer vision recognition results, and provides a reference basis for the subsequent identification of the sow's breastfeeding behavior.

[0017] In summary, the present invention has the technical effect of continuous recognition and improving the accuracy of computer vision recognition of sow maternal behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings, which are incorporated herein and constitute a part of the specification, illustrate embodiments of the invention and, together with the description, further serve to explain the principles of the invention and to enable those skilled in the relevant art to make and use the invention.

[0019] Figure 1 It is a logical schematic diagram of the intelligent identification method of sow maternal behavior based on computer vision and deep learning of the present invention; Figure 2 This is a logical schematic diagram of the present invention for identifying and collecting maternal behavior corresponding to sow postures and time nodes.

[0020] Figure 3 This is a logical schematic diagram of the sow health status and maternal behavior recognition and collection corresponding to the time nodes of the present invention.

[0021] Figure 4 This is a logical schematic diagram of the present invention for identifying and collecting maternal behavior corresponding to sow emotions, piglet numbers, genders and time nodes. DETAILED DESCRIPTION

[0022] The following is a detailed description of a method for intelligently identifying maternal behavior of sows based on computer vision and deep learning provided by the present invention in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternative methods to implement some known technologies; and the accompanying drawings are only for a more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0023] It should be noted that the references to "one embodiment", "embodiment", "exemplary embodiments", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0024] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0025] It will be understood that the meaning of “on,” “over,” and “above” in this disclosure should be interpreted in the broadest manner, so that “on” means not only “directly on” something, but also includes the meaning of being “on” something with intervening features or layers therebetween, and “on” or “over” means not only “on” or “above” something, but also includes the meaning of being “on” or “above” something with no intervening features or layers therebetween.

[0026] Additionally, spatially relative terms such as "under," "beneath," "lower," "above," "upper," and the like may be used herein for descriptive convenience to describe the relationship of one element or feature to another element or features, as shown in the accompanying drawings. Spatially relative terms are intended to encompass different orientations of the device in use or operation in addition to the orientation depicted in the accompanying drawings. The device may be oriented in other ways, and the spatially relative descriptors used herein may be similarly interpreted accordingly.

[0027] like Figure 1 As shown, an embodiment of the present invention provides a method for intelligently identifying maternal behavior of sows based on computer vision and deep learning, such as Figure 1 As shown, the method includes: Step S1: Set the target breeding area, put the sows in the breeding pen, set up multiple cameras, video storage devices, servers, and thermal imaging devices. The server uses computer vision CV algorithm, machine learning ML algorithm and OpenPose algorithm to analyze the sow information and upload the above information to the server.

[0028] The sow information includes sow size, sow temperature information, sow skin color, and identification of the sow's supine posture, standing posture, face, mouth, limbs and abdomen, and each nipple of the sow based on the sow image.

[0029] Step S2: When it is detected that the sow is in a continuous lying state and there are multiple piglets in the sow's abdomen, the pig target detection algorithm based on the lightweight YOLOv4 is quickly used to detect each piglet and number each piglet.

[0030] Step S3: Identify the movement characteristics of each piglet, identify the position of the piglet's mouth, identify whether the piglet's mouth is located on the sow's nipple, identify whether the piglet is active, and identify whether there is liquid on the piglet's body; cooperate with thermal imaging equipment to identify the temperature characteristics of the piglet, and identify whether the piglet has a breast-arching action based on the thermal imaging somatosensory map and the OpenPose algorithm.

[0031] Step S4: Identify the position of the sow's snout and determine whether the tongue of the sow's snout is located on the piglet's body.

[0032] Step S5: When the tongue of the sow's mouth is located on the piglet's body, identify whether the piglet's body has body temperature, and compare the thermal imaging image of the piglet with the original image to determine whether there is any defect in the piglet's body image and whether there is liquid on the piglet's body.

[0033] Step S6: When the piglet's body temperature is within the normal range, the piglet's body image is complete, and there is no liquid in the piglet's body, the piglet is judged to be in a normal state. The sow's tongue is on the piglet, that is, the sow's behavior is judged to be licking behavior; When the piglet's mouth is on the sow's nipple and the piglet's body temperature is within the normal range, the sow's behavior is judged to be feeding behavior.

[0034] When the piglet's body parts are not located on the sow's nipples, the piglet's body temperature is lower than normal, and the piglet remains motionless for a long time, the piglet is judged to be in an inactive state.

[0035] When the sow's snout is located on the piglet, the piglet is in an inactive state, and the piglet's body image is missing compared to the initial image, it is determined that the sow has eaten the inactive piglet.

[0036] When the sow's snout is located on the piglet, the piglet is in a normal state, the piglet has a normal body temperature, and the piglet's body image is missing compared to the initial image, it is determined that the sow has committed infanticide.

[0037] The actual maternal behavior is compared with the various model algorithms used, and the results obtained by mainstream algorithms such as YOLOv4, OpenPose, CV and machine learning ML algorithms are compared. If the two are inconsistent, the model algorithm is corrected and trained until the trained model results are close to the real results.

[0038] The present invention utilizes video image analysis in combination with thermal imaging images, performs logical judgment after comparison, and uses the result of the logical judgment to further modify the model algorithm for judging the maternal behavior of pigs, thereby improving the final accuracy of the model algorithm.

[0039] In addition, an audio acquisition module is introduced, which includes a sow audio acquisition module and a piglet audio acquisition module, which respectively collect audio of sows and different piglets, and upload them to the server after denoising to analyze the calls of sows and piglets at different stages. The decibels and frequencies of normal sow and piglet calls are stored as standards.

[0040] When the audio analysis of the piglet shows that its call is zero and there is no response for a long time, it is judged in the inactive state based on the image logic.

[0041] When the audio analysis of the piglet shows that its cry is sharp and the decibel value is higher than the standard, combined with the sound signal of the sow, if the audio analysis of the sow shows that its cry is abnormal and the decibel value is higher than the standard, a logical judgment is made combined with the image to determine that infanticide has occurred.

[0042] When the audio analysis display of the piglet shows that the call is sharp and the decibel value is higher than the standard, combined with the sound signal of the sow, if the frequency of the sow's audio analysis display changes, and then the decibel value and frequency of the piglet's call tend to be abnormal, it means that the sow licked or changed its posture to avoid the piglet being crushed. Therefore, combined with image recognition, it is judged that the sow is licking or changing its posture.

[0043] The present invention further determines that the piglets are in an inactive state or a gnawing state through audio analysis combined with video image analysis, judges that the sow is in infanticide behavior, improves the accuracy of judgment, and further corrects the results of the model algorithm.

[0044] In addition, for lactating sows, they live in farrowing crates during lactation, and the size of the crates is Minimize the sow's behavioral activities. According to the body height from low to high, the sow's posture can be divided into: ① Lying posture: The posture of the sow resting and breastfeeding. According to the angle of the body tilt from small to large, it can be divided into standing, lying and side-lying. When the sow is breastfeeding, the body is lying on the left or right side, and the nipples are exposed to facilitate the piglets to find a sucking position. Standing is the posture of the sow refusing to breastfeed. The sow is not disturbed by the piglets' active sucking, and the sow can get a better rest; lying is a resting posture between lying on the side and lying on the standing. ② Standing posture: The most common posture of sow exercise. The typical movements of sows during lactation include eating, drinking water and excretion. The sow always maintains a standing posture throughout the process of these behaviors.

[0045] Therefore, the present invention recognizes the posture of the sow through video and determines the specific posture of the sow. If the image recognition result is side-lying, it indicates that the sow is in a breastfeeding posture; if the image recognition result is reclining, it is necessary to combine the image recognition of the piglet, thereby further improving the recognition accuracy of the sow's breastfeeding behavior.

[0046] If the image recognition result is a lying or standing posture, it means that the sow is not breastfeeding. At this time, combined with the piglet's behavior and thermal imaging, it is further determined whether the sow is eating, drinking or defecating, as well as other situations.

[0047] At the same time, during the lactation period, by setting up a time module, the duration and frequency of different postures of specific sows in the time periods of 0-7 days, 7-14 days, 15-21 days, and 22-28 days after parturition are counted, that is, the time periods, moments and frequencies of supine, reclining, side-lying, and standing are counted to form lactation posture data, and model training is carried out. The data is compared with the actual lactation behavior of the sow, and a posture prediction database for sows after parturition is established. Through statistics on different days and times, the probability of subsequent sows appearing in various postures is predicted, to assist computer vision in identifying the sow's lactation posture.

[0048] For example, when computer vision recognizes that the posture of a sow is different from that in the posture prediction database on a certain day and at a certain time, infrared thermal imaging is further used to identify the posture of the sow and piglets, and after judgment, it assists computer vision recognition in calibration.

[0049] In addition, the present invention uses embedded monitoring equipment installed in the excretion area of ​​the pig house to monitor the excretion behavior of a pen of sows around the clock, and identifies whether the sows are sick based on their excretion behavior. Video data of sows and piglets are obtained through cameras, and image data sets are obtained frame by frame. Deep convolutional neural networks are used for deep learning training, and a model that can more accurately distinguish the excretion postures, abnormal excrement, and pig back numbers of sows and piglets is obtained.

[0050] At the same time, the video is used to collect the fluctuation frequency of the pig's spine and abdomen, the breathing frequency of the pig's mouth, and the frequency of the pig's sneezing. By comparing the above characteristic values ​​with the standard of healthy pigs, it is judged whether the pig has a cough or a cold. The frequency of the pig's excretion is used to judge whether the pig has diarrhea.

[0051] Therefore, when the sow is in a sick state, the breastfeeding, licking and infanticide behaviors are monitored, and the above behaviors are counted using computer video recognition. The image recognition information of the above behaviors is stored in the original database to form a prediction database, and the model is trained to further improve the accuracy of subsequent recognition of maternal behaviors of sows in sick states.

[0052] When subsequent video detection finds that the sow is in a sick state, the probability of its breastfeeding, licking and infanticide behaviors can be predicted by collecting the frequency of fluctuations in the pig's spine and abdomen, the breathing frequency of the pig's mouth, the frequency of sneezing, and the frequency of diarrhea, thereby assisting computer vision in identifying the sow's maternal behavior.

[0053] The mood of the sow is closely related to the maternal behavior of the sow. When the sow is frightened and dull, the frequency of its refusal to breastfeed will increase. When the sow is angry, the frequency of its infanticide behavior will increase.

[0054] When the sow shows various emotions, the present invention identifies the sow's facial expression, ear posture and white of the eye area. For the sow's face, the sow's expression is determined by identifying the position of the mouth and the white of the eye, and the movement trajectory of the white of the eye to determine whether there is fear or dullness; through the ear posture, the ear shaking frequency, ear position, and swing amplitude are analyzed to identify whether the sow is in an angry or relaxed emotional state.

[0055] By collecting a large amount of sows' facial expressions and ear postures and comparing them with the facial expressions and ear postures in a normal state, we can determine the sow's emotional state and form an emotional database.

[0056] After the sow gives birth, the facial expression and ear posture of the sow are collected to identify the sow's emotions and the correspondence with the sow's maternal behavior.

[0057] Therefore, when the above emotions are detected in sows, the image recognition information of the maternal behavior identified at this time is stored in the original database, and the model is trained to form a prediction database, so as to further improve the accuracy of subsequent recognition of maternal behavior of sows in different emotional states.

[0058] When subsequent video recognition finds that the sow is in the above emotional state, the probability of her suckling, licking and infanticide behaviors can be predicted, assisting computer vision in identifying the sow's maternal behavior.

[0059] In addition, the interactive behaviors of the sow and all the piglets are counted at the same time, each piglet is numbered, the interactive behaviors of the sow and piglets are tested separately, and the various maternal behaviors of the sow and each piglet are identified.

[0060] First, the piglets are identified and numbered, and the number N of piglets and the sex ratio of piglets are counted.

[0061] When the number of piglets N was in the range of 5-15, the changes in maternal behavior of sows with different litter sizes were statistically analyzed.

[0062] Specifically, after the sow gives birth, the interaction between the sow and all the piglets N is counted, the number of occurrences of suckling, licking and infanticide behaviors are identified, and the genders of the piglets that exhibit the interactive behaviors are further divided.

[0063] For example, when N=10, the sow exhibited lactation and licking behaviors with 6 of the sow piglets, licking behaviors with 3 of the male piglets, and infanticide behaviors with 1 of the male piglets.

[0064] Statistics are collected on the interaction between the sow and all the male and female piglets N at different time periods after the sow gives birth, and the timetable is stored in the database.

[0065] When the number of piglets is too small, the sow has the appropriate energy to take care of each piglet; but when the number of piglets is too large, the sow may not be able to allocate enough energy to take care of all the piglets, and may abandon her offspring or even kill her infants.

[0066] When there are a large number of male piglets among piglets, they are lively and aggressive by nature, and their behavior of rooting for the sow's breast is more frequent, which consumes the sow's physical strength, stimulates the sow and reduces the breastfeeding behavior.

[0067] However, due to the different numbers and genders of piglets, the sow may be affected in the process of nursing the same litter of piglets, and multiple maternal behaviors may occur simultaneously over time, which are intertwined with each other.

[0068] Therefore, compared with the prior art that single-handedly and isolatedly counts a certain behavior, the present invention monitors and analyzes all piglets, identifies several maternal behaviors of sows that exist simultaneously, combines the sow's behavior with the number of piglets, the gender of the piglets, the sow's own emotions, and health status, and at the same time, counts the changes in the sow's licking, suckling and infanticide behaviors at different times in a timeline manner to form a dynamic database, further calibrates and predicts the computer vision recognition results, provides a reference basis for the subsequent recognition of the sow's maternal behavior, and improves the accuracy and comprehensiveness of the recognition of the sow's maternal behavior.

[0069] The present invention identifies different maternal behaviors as the age of piglets increases during the process of sow breeding, based on the different numbers and genders of piglets, and stores them in a database. When subsequent video detection is performed, the number and gender of piglets are detected in advance, and the maternal behaviors that will occur at different times and moments can be predicted based on the number and gender of piglets, thereby assisting computer vision in identifying the maternal behaviors of sows and further improving the accuracy of computer vision recognition.

[0070] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0071] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for intelligently identifying maternal behavior of sows based on computer vision and deep learning, characterized in that: The method comprises: Step S1: Set the target breeding area, put the sows to be born in the breeding pen, set up multiple cameras, video storage devices, servers, and thermal imaging devices; the server uses computer vision CV algorithm, machine learning ML algorithm and OpenPose algorithm to analyze the sow information, and upload the sow information to the server; the sow information includes the sow size, sow temperature sensory information, sow skin color, and recognize the sow posture, sow face, mouth, limbs and abdomen and each nipple of the sow according to the sow image; Step S2: When it is detected that the sow is in a continuous lying state and multiple piglets appear in the abdomen of the sow, a pig target detection algorithm based on lightweight YOLOv4 is quickly used to detect each piglet and number each piglet; Step S3: Identify the movement characteristics of each piglet, identify the mouth of the piglet, identify whether the mouth of the piglet is located on the nipple of the sow, identify whether the piglet is active, and identify whether there is liquid on the body of the piglet; cooperate with the thermal imaging equipment to identify the temperature characteristics of the piglet, and identify whether the piglet has a breast-pushing action based on the thermal imaging image and the OpenPose algorithm; Step S4: Identify the position of the sow's mouth and determine whether the tongue of the sow's mouth is located on the piglet's body; Step S5: When the tongue of the sow's mouth is located on the piglet's body, identify whether the piglet's body has body temperature, and compare the thermal imaging image of the piglet with the original image to determine whether there is a defect in the piglet's body image and whether there is liquid on the piglet's body; Step S6: When the piglet's body temperature is within the normal range, the piglet's body image is complete, and there is no liquid in the piglet's body, the piglet is judged to be in a normal state; when the sow's tongue is located on the piglet, the sow's behavior is judged to be licking behavior; when the piglet's mouth is located on the sow's nipple and the piglet's body temperature is within the normal range, the sow's behavior is judged to be feeding behavior; when the piglet's body part is not located on the sow's nipple, the piglet's body temperature is lower than the normal value, and the piglet is still for a long time, the piglet is judged to be in an inactive state; when the sow's mouth is located on the piglet, the piglet is in a normal state, the piglet has a normal body temperature, and the piglet's body image is missing compared to the initial image, it is judged that the sow has committed infanticide; The actual maternal behavior is compared with the maternal behavior identified by the above model algorithm. If the two are inconsistent, the model algorithm is corrected and trained until the trained model result is close to the logical judgment result.

2. According to claim 1, a method for intelligently identifying maternal behavior of sows based on computer vision and deep learning is characterized in that: It also includes an audio acquisition module, which includes a sow audio acquisition module and a piglet audio acquisition module, which respectively collect audio of the sow and different piglets, and combine image recognition to judge the maternal behavior of the sow.

3. The method for intelligently identifying maternal behavior of sows based on computer vision and deep learning according to claim 2, characterized in that: The sow's posture is identified through video, and the posture of a specific sow at different times after giving birth is counted through the time module. A dynamic prediction database is established to assist computer vision in identifying the sow's lactation posture.

4. The method for intelligently identifying maternal behavior of sows based on computer vision and deep learning according to claim 3, characterized in that: The frequency of sow's spine and abdomen fluctuation, the breathing frequency of the pig's mouth, the sneezing frequency and the excretion frequency are collected and monitored, and the above behaviors are counted using computer video recognition to establish a dynamic prediction database to assist computer vision in identifying the sow's maternal behavior.

5. The method for intelligently identifying maternal behavior of sows based on computer vision and deep learning according to claim 4, characterized in that: The sow's facial expression, ear posture and white of the eye area are identified to determine the sow's emotional state, and the emotional data is matched with the sow's maternal behavior. The above information is stored in the database, the model is trained, and a dynamic prediction database is established to assist computer vision in identifying the sow's maternal behavior.

6. The method for intelligently identifying maternal behavior of sows based on computer vision and deep learning according to claim 5, characterized in that: For the sow's face, the position of the mouth and whites of the eyes, and the movement trajectory of the whites of the eyes can be identified to determine the sow's expression and whether she is frightened or dull. Through ear posture, the frequency, position, and amplitude of ear swinging can be analyzed to identify whether the sow is in an angry or relaxed emotional state, and a dynamic prediction database can be established to assist computer vision in identifying maternal behavior.

7. The method for intelligently identifying maternal behavior of sows based on computer vision and deep learning according to claim 6, characterized in that: Count the number and gender of different piglets, count the interaction between sows and all piglets at different time periods and times, establish a dynamic prediction database, and assist computer vision in identifying maternal behavior.

Images