Intelligent robot for sow oestrus detection

Abstract

The invention discloses an intelligent robot for sow oestrus detection, belongs to the technical field of computers, and particularly relates to an intelligent robot for sow oestrus detection, which structurally and specifically comprises an image acquisition device for image acquisition and an intelligent mobile platform for autonomous inspection and obstacle avoidance, the bionic model device carried on the intelligent mobile platform is used for inducing the estrus of the sow and detecting the estrus of the sow, through mutual cooperation of the components, automatic inspection can be achieved, the detection work of the estrus state of the sow can be automatically completed, manual observation and intervention are not needed, and other detection technologies are not needed; the mental stress on the sow is effectively avoided. By adopting the intelligent robot, sow oestrus detection can be visually, effectively and quickly carried out, and the obtained detection result is high in precision and accuracy. Manpower and material resources of the pig farm are greatly saved, the intelligent robot can be repeatedly used, and the working cost is directly reduced.

Description

Technical Field

[0001] The invention belongs to the field of computer technology, and particularly relates to an intelligent robot for sow estrus detection. Background Art

[0002] The pig industry is an important part of my country's animal husbandry and one of the channels to ensure the source of meat we need for our lives. As the source of production power for pig farms, the number of weaned piglets produced annually by sows has become an important indicator to measure the reproductive capacity of sows, which is closely related to the benefits of pig farms. The technology of identifying sow estrus is not only directly related to the feeding plan of sows, but also affects whether sows can be bred normally and whether the production cycle can be started in time, which directly affects the number of piglets born by sows.

[0003] The inventors have found through research that in the prior art, the sow estrus detection method usually relies on manual identification, which consumes a lot of manpower and time, and has low detection accuracy and is prone to errors. Summary of the invention

[0004] In order to at least solve the above technical problems, the present invention provides an intelligent robot for sow estrus detection, comprising:

[0005] An image acquisition device, used for acquiring sow images in real time and sending the acquired sow images to a terminal that establishes a data connection with the image acquisition device;

[0006] The intelligent mobile platform is mounted under the bionic model device and is used for adaptive navigation and obstacle avoidance inspection according to the pig house environment;

[0007] The bionic model device is set to look like a boar, including an ultrasonic atomization module, a sound module, a mechanical arm and a data analysis module;

[0008] The ultrasonic atomization module is used to release a simulated boar odor;

[0009] The sound module is used to play aphrodisiac sounds;

[0010] The robot arm is used to link the bionic model device to rotate 360 ​​degrees and stay at any height when recognizing that the sow is in contact with the bionic model device, so as to imitate a real boar and cooperate with the sow to interact;

[0011] The data analysis module is used to use a deep neural network to identify the motion trajectory of the sow relative to the bionic model device according to the actual position of the sow and the actual position of the bionic model device in each frame of the sow image and the corresponding time, and to use a deep learning training method to identify the number of contacts between the sow and the bionic model device, generate a sow estrus detection result, and send it to a terminal connected to the bionic model device;

[0012] The terminal is used to receive the sow image collected by the image acquisition device and receive the sow estrus detection result sent by the bionic model device;

[0013] The terminal is also used to obtain subjective detection data input by the user, and to determine whether the sow estrus detection result is accurate based on the subjective detection data, and to mark it.

[0014] Further,

[0015] The ultrasonic atomization module is designed as a bionic pig nose, has the appearance of a pig nose, simulates the texture of a pig nose, and is specifically used to release at least one of saliva, urine and semen in the form of atomization to simulate the smell of a boar.

[0016] Further,

[0017] The data analysis module includes a data analysis submodule and a sending submodule;

[0018] The data analysis submodule is used to extract features within the point image frame, perform target detection on the pig nose of the sow, obtain its image position and record it as a rectangular frame A, then use the pre-calibrated image position of the target pig nose to record it as a rectangular frame B, determine the contact state of the single-frame image according to the distance between the rectangular frame A and the rectangular frame B, and use multiple frames of images to calculate the number of contacts. When the number of contacts exceeds a preset threshold, it is determined that the sow is in estrus, and a sow estrus detection result is generated;

[0019] The sending submodule is used to send the sow estrus detection result to the terminal connected to the bionic model device.

[0020] Further,

[0021] The data analysis module also includes a timer for counting the contact time between the bionic model device and the empty sow within a preset time.

[0022] Further,

[0023] The data analysis module is also used to count the length of time the sow's ears remain still. When it is monitored that on a certain day in the estrus cycle of an empty-stomach sow, the length of time the sow's ears remain still as counted by the timer is greater than or equal to a preset time domain value than on other days, that is, when the empty-stomach sow's interaction and contact time with the intelligent robot on a certain day is longer than on other days, and the longer time is greater than or equal to the preset time domain value, the sow estrus detection result is generated as an estrus state.

[0024] Further,

[0025] The data analysis submodule is specifically used to splice the collected sow images by randomly scaling, randomly cropping, and randomly arranging 4 pictures, and process the detection frame coordinates that exceed the boundaries to obtain a data set containing the width and height of all targets;

[0026] The data analysis submodule is also specifically used to adaptively calculate the best anchor box values ​​in different training sets according to different data sets before each training starts;

[0027] The data analysis submodule is also specifically used to compress and combine the information in the input feature map after obtaining the optimal anchor frame value, so as to extract higher-level features.

[0028] Further,

[0029] The data analysis submodule is also used to use a loss function to measure the overlap between the predicted bounding box and the true bounding box to determine whether the predicted box is correct. The specific formula used is: The calculation formula of DIOU is as follows:

[0030]

[0031] Where IoU(A, B) represents the IoU value between bounding boxes A and B, and d c (A, B) represents the center point distance between bounding boxes A and B, and ω represents the length of the image diagonal.

[0032] Further,

[0033] The ultrasonic atomization module is used to release at least one of the saliva, urine and semen of a middle-aged boar to simulate the odor of the boar.

[0034] Further,

[0035] The intelligent mobile platform is configured with a chassis capable of realizing four-wheel differential inspection, a radar module, a composition module, and an execution module, and is specifically used for adaptive navigation and obstacle avoidance inspection in a pig house environment according to a preset path.

[0036] Further,

[0037] The terminal is also used to prompt the user to input subjective detection data, wherein the subjective detection data includes back pressure reaction state, vulva state, arched back state, and bilateral ear standing state.

[0038] Beneficial effects of the present invention: The intelligent robot combines the morphological characteristics of the sow's estrus period with the principle of the intelligent robot's "contact window". By adopting a bionic simulation of boars, the robot releases simulated boar odors and plays aphrodisiac sounds, automatically and dynamically detecting the estrus behavior of non-pregnant sows in a non-contact and pressure-free manner, and the detection results are faster and more efficient. This provides a guarantee for improving the success rate of insemination during mating and reducing the use of high-quality boar semen.

[0039] By adopting the intelligent robot provided by the present invention, the best time for artificial insemination of individual non-pregnant sows can be determined in a timely and effective manner, laying a solid foundation for improving the reproductive level of sows in large-scale breeding. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0041] Figure 1 A schematic diagram of the structure of an intelligent robot for sow estrus detection provided by the present invention;

[0042] Figure 2 This is a schematic diagram of the working effect of an intelligent robot provided by the present invention. DETAILED DESCRIPTION

[0043] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0044] In order to more clearly illustrate the present invention, the present invention is further described below in conjunction with preferred embodiments and accompanying drawings. Similar components in the accompanying drawings are represented by the same reference numerals. It should be understood by those skilled in the art that the content specifically described below is illustrative rather than restrictive, and should not be used to limit the scope of protection of the present invention.

[0045] In one aspect of the present invention, an intelligent robot for detecting estrus of sows is provided. Figure 1 As shown, including:

[0046] The image acquisition device 101 is used to acquire sow images in real time and send the acquired sow images to a terminal 104 that establishes a data connection with the image acquisition device 101;

[0047] In the present invention, the image acquisition device 101 sets the acquisition frequency to 30fps and the resolution to 1280*720 to acquire sow images, and transmits the acquired sow images to the terminal 104 via a data line.

[0048] The intelligent mobile platform 102 is mounted below the bionic model device 103 and is used for adaptive navigation, obstacle avoidance and inspection according to the pig house environment.

[0049] In the present invention, the intelligent mobile platform 102 is installed on the bionic model device 103, and the intelligent mobile platform 102 is configured with a self-built chassis capable of realizing four-wheel differential inspection, including executing movement speed, position, and time.

[0050] Furthermore, the intelligent mobile platform 102, i.e. the inspection chassis, is equipped with a radar module, a composition module, an execution module and an inspection vehicle body. This enables the intelligent mobile platform 102 to automatically navigate and automatically patrol according to a preset path while avoiding obstacles. The inspection vehicle body is optimized and adjusted in terms of load capacity, speed, accuracy, stability and safety in combination with the pig house environment.

[0051] The present invention uses an intelligent mobile platform for inspection, so that in the complex environment of the pig house with many railings and feeding troughs, it can conduct automatic inspections by analyzing the pig house environment and the animal behavior of pigs in a targeted manner, thereby greatly saving manpower, improving work efficiency, realizing autonomous operation, and providing support for the development of precision animal husbandry.

[0052] The bionic model device 103 is configured to have a boar appearance, including an ultrasonic atomization module 1031, a sound module 1032, a mechanical arm 1033, and a data analysis module 1034;

[0053] Among them, the ultrasonic atomization module 1031 is used to release the simulated boar odor;

[0054] Furthermore, the ultrasonic atomization module 1031 is designed as a bionic pig nose, having the appearance of a pig nose and simulating the texture of a pig nose, and is specifically used to release at least one of saliva, urine and semen in the form of atomization to simulate the smell of a boar.

[0055] The ultrasonic atomization module 1031 can be specifically configured as a micro piezoelectric atomizer, and the bionic pig nose can be set to a silicone material and a pig nose appearance, so as to improve the authenticity of the bionic model device and further improve the interactivity between the sow to be tested and the bionic model device.

[0056] The sound module 1032 is used to play aphrodisiac sounds;

[0057] Furthermore, the sound module 1032 can be configured as an external amplifier of the playback device. The aphrodisiac sound can be the call of a boar.

[0058] Furthermore, in one embodiment of the present invention, combined with the influence of the boar age factor, the calls and odors of boars of different age groups are collected, the calls of boars of different age groups are configured in the sound module 1032, and the odors of boars of different age groups are configured in the ultrasonic atomization module 1031. In the present invention, the bionic model device 103 preferably releases the saliva of middle-aged boars and plays the calls of boars.

[0059] The robot arm 1033 is used to link the bionic model device to rotate 360 ​​degrees and stay at any height when recognizing that the sow is in contact with the bionic model device, so as to imitate a real boar and interact with the sow;

[0060] The data analysis module 1034 is used to use a deep neural network to identify the motion trajectory of the sow relative to the bionic model device according to the actual position of the sow and the actual position of the bionic model device in each frame of the sow image and the corresponding time, and to use a deep learning training method to identify the number of contacts between the sow and the bionic model device, generate a sow estrus detection result, and send it to the terminal connected to the bionic model device;

[0061] Furthermore, the data analysis module 1034 includes a data analysis submodule and a sending submodule.

[0062] Among them, the data analysis submodule is used to extract features within the point image frame, perform target detection on the sow's nose, obtain its image position and record it as rectangular frame A, then use the pre-calibrated image position of the target nose as rectangular frame B, determine the contact state of the single-frame image according to the distance between rectangular frame A and rectangular frame B, and use multiple frames of images to calculate the number of contacts. When the number of contacts exceeds the preset threshold, it is determined that the sow is in estrus, and the sow estrus detection result is generated.

[0063] Furthermore, the implementation of the data analysis submodule is elaborated in detail, including:

[0064] The data analysis submodule is specifically used to stitch the collected sow images using four pictures by random scaling, random cropping, and random arrangement, and to process the detection box coordinates that exceed the boundary to obtain a data set of width and height containing all targets, so that the model can identify targets in a smaller range.

[0065] Furthermore, the data analysis submodule is specifically used to randomly select the coordinates (xc, yc) of the image stitching reference point, randomly select four images, and after the four images are resized and scaled according to the reference points, they are placed in the upper left, upper right, lower left, and lower right positions of the large image of the specified size. According to the size transformation method of each image, the mapping relationship is mapped to the image label, and the large image is stitched according to the specified horizontal and vertical coordinates, and the coordinates of the detection box that exceeds the boundary are processed to obtain a data set of width and height containing all targets.

[0066] The data analysis submodule is also specifically used to adaptively calculate the optimal anchor box values ​​in different training sets based on different data sets before each training starts.

[0067] Furthermore, the data analysis submodule specifically scales the width and height of all targets in the data set to the preset size in a proportional manner, so that the maximum value of the width and height meets the preset size, and changes the bboxes from relative coordinates to absolute coordinates, specifically by multiplying the scaled width and height. Filter the bboxes and retain those with a width and height greater than or equal to two pixels. Use k-means clustering to obtain n anchors. (Use a genetic algorithm to randomly mutate the width and height of the anchors, and filter the mutation results, assign the expected mutation results to the anchors, and set the number of mutations to 1000 times.

[0068] The data analysis submodule is also specifically used to compress and combine the information in the input feature map after obtaining the optimal anchor frame value, so as to extract higher-level features.

[0069] Specifically, the input feature map is divided into two parts, one part is processed by a small convolutional network, and the other part is directly processed by the next layer. The two parts of the feature map are spliced ​​together as the input of the next layer to improve the feature extraction efficiency.

[0070] Furthermore, the data analysis submodule is specifically the first convolution layer in the network, which downsamples the input feature map to reduce the amount of calculation and the amount of parameters. This method can effectively reduce the amount of calculation and the amount of parameters, while retaining the important information in the input feature map, which helps to improve the feature extraction ability of the model. In addition, the input feature map is divided into two parts, one part is processed by the sub-network, and the other part is directly processed by the next layer. In the sub-network, a convolution layer is first used to compress the input feature map, and then a series of convolution operations are performed, and finally a convolution layer is used to expand to extract relatively few high-level features. In the next layer, the feature map processed by the sub-network is spliced ​​with the feature map directly processed, and then a series of convolution operations are performed to combine low-level detail features with high-level abstract features to improve the efficiency of feature extraction. This method can significantly reduce the parameters and amount of calculation of the network, improve the efficiency of feature extraction, and thus speed up the training and reasoning speed of the model.

[0071] The data analysis submodule detects and perceives the estrus behavior of sows based on the deep learning model algorithm. In the process of building the model, the loss function used is used to measure the difference between the predicted bounding box and the true bounding box. The loss function is specifically used to measure the overlap between the predicted bounding box and the true bounding box. In object detection, the overlap between the predicted box and the true box is evaluated to determine whether the predicted box is correct.

[0072] Specifically, for each predicted bounding box, we calculate its IoU value with all true bounding boxes, and then select the true bounding box with the largest IoU as its corresponding matching target, thereby calculating its IoU loss.

[0073] The specific formula is as follows:

[0074]

[0075] Where IoU(A, B) represents the IoU value between bounding boxes A and B, and d c (A, B) represents the center point distance between bounding boxes A and B, and ω represents the length of the image diagonal.

[0076] By introducing the center point distance, the impact of positioning error on target detection performance can be effectively alleviated.

[0077] The model structure used by the data analysis submodule mainly includes the following components: Input, which consists of 3 different output layers, responsible for detecting large, medium and small scale targets respectively. Backbone network, which has strong feature extraction capability and computational efficiency. Neck network is used to detect at different feature map levels, which can improve the performance of target detection and can fuse information from different feature map levels. At the output, a loss function is used to alleviate the problem of class imbalance in target detection and improve the performance of the model. Non-maximum suppression, after outputting the results, will process the overlapping target boxes to obtain the final detection results.

[0078] The sending submodule is used to send the sow estrus detection result to the terminal connected to the bionic model device.

[0079] In another embodiment of the present invention, the data analysis module further includes a timer for counting the contact time between the bionic model device and the empty sow within a preset time, and the preset time may be 3 minutes. The timer may also count the length of time the sow's ears remain still. By counting the length of time the intelligent robot interacts with the sow, it is possible to avoid the sow losing sexual interest due to long-term contact, thereby effectively improving the accuracy of detection.

[0080] Furthermore, the data analysis module is also used to generate an estrus detection result for the sow when it is monitored that the length of time the sow's ears are still on a certain day in the estrus cycle counted by the timer is greater than or equal to the length of time on other days, that is, when the empty-stomach sow's interaction and contact time with the intelligent robot is longer on a certain day than on other days, and the longer time is greater than or equal to the preset time domain value.

[0081] The terminal 104 is used to receive the sow image collected by the image acquisition device 101 and receive the sow estrus detection result sent by the bionic model device 103;

[0082] Furthermore, the terminal 104 is used to receive and display the sow images collected in real time by the image acquisition device 101, and can also receive and display the sow estrus detection results sent by the bionic model device 103, so that the user can observe the state of the sow at the terminal and obtain the sow estrus detection results fed back by the bionic model device.

[0083] Furthermore, the terminal can save sow images and sow estrus detection results for subsequent video analysis and other operations.

[0084] The terminal 104 is also used to obtain the subjective detection data input by the user, and determine whether the sow estrus detection result is accurate based on the subjective detection data, and mark it.

[0085] Furthermore, the terminal 104 is also used to prompt the user to input subjective detection data, wherein the subjective detection data includes back pressure reaction state, vulva redness and swelling state, arched back state, and bilateral ear still state.

[0086] In an embodiment corresponding to the first aspect of the present invention, when detecting estrus of sows, a model device is designed based on the behavioral characteristics of the dynamic response of empty-stomach sows to the feedback degree of boar's 'sex hormones', and the estrus response behavior is collected in a field test. The estrus response characteristics and reference thresholds of the intelligent robot's interaction duration, contact frequency, and bilateral ear stillness duration of empty-stomach sows are revealed, and a behavioral model for capturing the estrus characteristics of empty-stomach sows based on image recognition technology is established.

[0087] Construct an intelligent robot seduction and sow contact behavior intelligent and accurate recognition model using deep neural network to realize the judgment of pig nose contact in video images. Specifically,

[0088] Furthermore, data enhancement processing is performed, including: randomly selecting the coordinates (xc, yc) of the image stitching reference points, and randomly selecting four images. The four images are resized and scaled according to the reference points, and then placed at the upper left, upper right, lower left, and lower right positions of the large image of the specified size. According to the size transformation method of each image, the mapping relationship is mapped to the image label. According to the specified horizontal and vertical coordinates, the large image is stitched, and the coordinates of the detection box that exceeds the boundary are processed.

[0089] The present invention adopts the method of data enhancement, randomly uses 4 images, randomly scales and randomly splices them, adds many small targets, greatly increases the data diversity and makes the data set richer. By mixing four pictures with different semantic information, it is possible to detect targets beyond the conventional context, enhance the robustness of the model, and greatly improve the detection ability.

[0090] Adaptively process the data obtained by data augmentation, including: obtaining the width and height of all targets in the data set obtained by data augmentation. Scale each image to the preset size in proportion so that the maximum value of the width, height and width meets the preset size. Change the bboxes from relative coordinates to absolute coordinates, specifically by multiplying the scaled width and height. Filter bboxes and retain those with a width and height greater than or equal to two pixels. Use k-means clustering to obtain n anchors. (Use a genetic algorithm to randomly mutate the width and height of the anchors, and filter the mutation results, assign the expected mutation results to the anchors, and set the number of mutations to 1000.

[0091] The features are extracted, specifically used to compress and combine the information in the input feature map, so as to extract higher-level feature representations. Specifically, the first convolutional layer in the network downsamples the input feature map to reduce the amount of calculation and the amount of parameters. This method can effectively reduce the amount of calculation and the amount of parameters, while retaining the important information in the input feature map, which helps to improve the feature extraction ability of the model. In addition, the input feature map is divided into two parts, one part is processed by the sub-network, and the other part is directly processed by the next layer. In the sub-network, a convolutional layer is first used to compress the input feature map, and then a series of convolution operations are performed, and finally a convolutional layer is used to expand to extract relatively few high-level features. In the next layer, the feature map processed by the sub-network is spliced ​​with the feature map directly processed, and then a series of convolution operations are performed to combine low-level detail features with high-level abstract features to improve the efficiency of feature extraction. This method can significantly reduce the parameters and amount of calculation of the network, improve the efficiency of feature extraction, and thus speed up the training and reasoning of the model.

[0092] The deep learning-based model algorithm detects and perceives the estrus behavior of sows. In the process of building the model, the loss function used is used to measure the difference between the predicted bounding box and the true bounding box. The loss function is specifically used to measure the degree of overlap between the predicted bounding box and the true bounding box. In object detection, the overlap between the predicted box and the true box is evaluated to determine whether the predicted box is correct.

[0093] Specifically, for each predicted bounding box, we calculate its IoU value with all true bounding boxes, and then select the true bounding box with the largest IoU as its corresponding matching target, thereby calculating its IoU loss.

[0094] The specific formula is as follows:

[0095]

[0096] Where IoU(A, B) represents the IoU value between bounding boxes A and B, and d c (A, B) represents the center point distance between bounding boxes A and B, and ω represents the length of the image diagonal.

[0097] By introducing the center point distance, the impact of positioning error on target detection performance can be effectively alleviated.

[0098] The model structure mainly includes the following components: Input, which consists of 3 different output layers, responsible for detecting large, medium and small scale targets respectively. Backbone network, which has strong feature extraction capability and computational efficiency. Neck network is used for detection at different feature map levels, which can improve the performance of target detection and integrate information from different feature map levels. At the output, a loss function is used to alleviate the problem of class imbalance in target detection and improve the performance of the model. Non-maximum suppression, after outputting the result, will process the overlapping target boxes to obtain the final detection result. Mish activation function is used instead of ReLU activation function to improve the performance of the model.

[0099] In general, the above method can effectively improve the performance and robustness of the model, making the model accuracy as high as 99.99%.

[0100] The intelligent robot provided by the present invention can cooperate with an empty-stomach sow that actively contacts the bionic model device, collect images of the sow, identify the states of the sow such as standing for a long time and with both ears still, and further detect the estrus state of the sow.

[0101] In summary, the present invention is based on the biological characteristics of using an intelligent robot for estrus induction, and performs sow-to-intelligent robot contact identification. Using a deep neural network, the position of the bionic and actual sow in each frame is determined. By combining the time relationship, the motion trajectory of the detected sow relative to the intelligent robot can be obtained. Through deep learning training, the number of contacts between the sow and the intelligent robot can be identified, providing data support for the estrus determination of the sow, the video is segmented into frames, the target (sow) in the frame is segmented, and the pig skeleton is identified. The detection speed and detection accuracy can be greatly improved. Accurate estrus detection can realize scientific sow production. In addition, these technologies mainly break the limitations of traditional sow estrus detection, which is of great significance in reducing work pressure and maintaining scientific detectability.

[0102] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0103] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as herein.

[0104] It should be understood that the detailed description of the technical solutions of the present invention by means of the preferred embodiments is illustrative rather than restrictive. A person skilled in the art may modify the technical solutions described in the embodiments, or replace some of the technical features by equivalents, based on reading the specification of the present invention; and these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent robot for sow estrus detection, characterized in that: include: An image acquisition device, used for acquiring sow images in real time and sending the acquired sow images to a terminal that establishes a data connection with the image acquisition device; The intelligent mobile platform is mounted under the bionic model device and is used for adaptive navigation and obstacle avoidance inspection according to the pig house environment; The bionic model device is set to look like a boar, including an ultrasonic atomization module, a sound module, a mechanical arm and a data analysis module; The ultrasonic atomization module is used to release a simulated boar odor; The sound module is used to play aphrodisiac sounds; The robot arm is used to link the bionic model device to rotate 360 ​​degrees and stay at any height when recognizing that the sow is in contact with the bionic model device, so as to imitate a real boar and cooperate with the sow to interact; The data analysis module is used to use a deep neural network to identify the motion trajectory of the sow relative to the bionic model device according to the actual position of the sow and the actual position of the bionic model device in each frame of the sow image and the corresponding time, and to use a deep learning training method to identify the number of contacts between the sow and the bionic model device, generate a sow estrus detection result, and send it to a terminal connected to the bionic model device; The terminal is used to receive the sow image collected by the image acquisition device and receive the sow estrus detection result sent by the bionic model device; The terminal is also used to obtain subjective detection data input by the user, and to determine whether the sow estrus detection result is accurate based on the subjective detection data, and to mark it.

2. The intelligent robot according to claim 1, characterized in that: The ultrasonic atomization module is designed as a bionic pig nose, has the appearance of a pig nose, simulates the texture of a pig nose, and is specifically used to release at least one of saliva, urine and semen in the form of atomization to simulate the smell of a boar.

3. The intelligent robot according to claim 2, characterized in that: The data analysis module includes a data analysis submodule and a sending submodule; The data analysis submodule is used to extract features within the point image frame, perform target detection on the pig nose of the sow, obtain its image position and record it as a rectangular frame A, then use the pre-calibrated image position of the target pig nose to record it as a rectangular frame B, determine the contact state of the single-frame image according to the distance between the rectangular frame A and the rectangular frame B, and use multiple frames of images to calculate the number of contacts. When the number of contacts exceeds a preset threshold, it is determined that the sow is in estrus, and a sow estrus detection result is generated; The sending submodule is used to send the sow estrus detection result to the terminal connected to the bionic model device.

4. The intelligent robot according to claim 3, characterized in that: The data analysis module also includes a timer for counting the contact time between the bionic model device and the empty sow within a preset time.

5. The intelligent robot according to claim 4, characterized in that: The data analysis module is also used to count the length of time the sow's ears remain still. When it is monitored that on a certain day in the estrus cycle of an empty-stomach sow, the length of time the sow's ears remain still as counted by the timer is greater than or equal to a preset time domain value than on other days, that is, when the empty-stomach sow's interaction and contact time with the intelligent robot on a certain day is longer than on other days, and the longer time is greater than or equal to the preset time domain value, the sow estrus detection result is generated as an estrus state.

6. The intelligent robot according to claim 5, characterized in that: The data analysis submodule is specifically used to splice the collected sow images by randomly scaling, randomly cropping, and randomly arranging 4 pictures, and process the detection frame coordinates that exceed the boundaries to obtain a data set containing the width and height of all targets; The data analysis submodule is also specifically used to adaptively calculate the best anchor box values ​​in different training sets according to different data sets before each training starts; The data analysis submodule is also specifically used to compress and combine the information in the input feature map after obtaining the optimal anchor frame value, so as to extract higher-level features.

7. The intelligent robot according to claim 6, characterized in that: The data analysis submodule is also used to use a loss function to measure the overlap between the predicted bounding box and the true bounding box to determine whether the predicted box is correct. The specific formula used is: The calculation formula of DIOU is as follows: Where IoU(A, B) represents the IoU value between bounding boxes A and B, and d c (A, B) represents the center point distance between bounding boxes A and B, and ω represents the length of the image diagonal.

8. The intelligent robot according to claim 7, characterized in that: The ultrasonic atomization module is used to release at least one of the saliva, urine and semen of a middle-aged boar to simulate the odor of the boar.

9. The intelligent robot according to claim 8, characterized in that: The intelligent mobile platform is configured with a chassis capable of realizing four-wheel differential inspection, a radar module, a composition module, and an execution module, and is specifically used for adaptive navigation and obstacle avoidance inspection in a pig house environment according to a preset path.

10. The intelligent robot according to claim 9, characterized in that: include: The terminal is also used to prompt the user to input subjective detection data, wherein the subjective detection data includes back pressure reaction state, vulva state, arched back state, and bilateral ear standing state.

Images