A robot for automatically inspecting poultry in three-dimensional cages
Through the automatic inspection robot, the environmental quality and poultry behavior in the three-dimensional cage mode is monitored in real time, and the neural network is used to process sound and image information, solving the problems of low inspection efficiency and health risks in the existing technology, achieving efficient and accurate Poultry house management.
Patent Information
- Application Number
- CN202310754688.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-26
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2043-06-26
AI Technical Summary
The existing technology is difficult to effectively monitor the environmental quality of poultry houses and poultry behavior in the three-dimensional multi-layer cage mode, resulting in low patrol efficiency, missed inspection, missed inspection and low egg laying efficiency, and long-term exposure of workers to complex environments affects health. .
Design a robot that automatically inspects poultry for three-dimensional cages, equipped with a motion mechanism, environmental data acquisition unit, positioning identification unit and poultry behavior detection unit. It uses BP neural network and convolutional neural network to process sound and image information and monitor environmental quality in real time and poultry status, quickly locate abnormal situations.
Efficient and accurate environmental quality and poultry behavior monitoring are achieved, inspection efficiency is improved, missed and missed inspections of manual inspections are reduced, abnormal situations are handled in a timely manner, and health risks of workers being exposed to harmful environments are avoided.
Smart Images

Figure CN116766226B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of poultry house environment detection, and in particular relates to a robot for automatically inspecting three-dimensional caged poultry. Background Art
[0002] With the continuous development of science and technology, the existing breeding industry has begun to upgrade to a three-dimensional multi-layer cage model, which has increased the scale of breeding on the basis of the original breeding area. However, the environmental quality in the poultry house of this breeding model is very complicated, and it will have a great impact on the respiratory system of the staff who patrol the poultry house for a long time. The traditional environmental quality detection device is a fixed detection. When the indoor area of the poultry house is large, multi-point detection needs to be set up. This detection method cannot fully describe the distribution of environmental quality in the entire poultry house.
[0003] At present, there are not many autonomous inspection technologies and equipment for poultry in three-dimensional multi-layer breeding mode. Most of them are manual inspections, which observe the physiological state of poultry with the naked eye and identify sick and dead poultry. In addition, in multi-layer breeding, workers are tired because they have to bend over constantly, which makes manual inspections inefficient and may cause missed inspections and false inspections. Because the number of breeding is quite large, the existing inspection equipment can only alarm for abnormal situations, but cannot quickly locate specific cages. Secondly, for egg-laying poultry, the manual inspection method cannot accurately identify poultry cages with low egg-laying efficiency, resulting in the problem that egg-laying efficiency cannot be improved in time.
[0004] Therefore, it is necessary to design a robot for automatically inspecting three-dimensional caged poultry to solve the above problems. Summary of the invention
[0005] The purpose of the present invention is to provide a robot for automatically inspecting poultry in three-dimensional cages to solve the above-mentioned problems, to achieve real-time monitoring of the behavior of poultry in the three-dimensional cage mode, and to effectively solve the problems encountered during manual inspections, thereby improving the breeding efficiency.
[0006] To achieve the above-mentioned purpose, the present invention provides the following scheme: a robot for automatically inspecting three-dimensional caged poultry, comprising a motion mechanism, a scissor-type lifting mechanism is fixedly connected to the top of the motion mechanism, an environmental data acquisition unit, a positioning identification unit, a poultry behavior detection unit, and a central data processing unit are fixedly connected to the top of the scissor-type lifting mechanism, the motion mechanism, the environmental data acquisition unit, the scissor-type lifting mechanism, the positioning identification unit, the poultry behavior detection unit and the central data processing unit are electrically connected, the poultry behavior detection unit is used to monitor the sound information, image information, and video information of poultry, the sound information, image information, and video information are transmitted to the central data processing unit, and the central data processing unit is used to process a BP neural network, a convolutional neural network, the sound information, image information, and video information.
[0007] Preferably, the poultry behavior detection unit includes a high-definition camera, a dual-axis servo gimbal, an infrared thermal imager, and a sound auxiliary module; the high-definition camera, the dual-axis servo gimbal, the infrared thermal imager, and the sound auxiliary module are fixedly connected to the top of the scissor-type lifting mechanism; and the high-definition camera, the dual-axis servo gimbal, the infrared thermal imager, and the sound auxiliary module are electrically connected to the central data processing unit.
[0008] Preferably, the environmental data acquisition unit includes a temperature sensor, a carbon dioxide sensor, an ammonia sensor, a dust concentration sensor, and an oxygen concentration sensor, and the temperature sensor, carbon dioxide sensor, ammonia sensor, dust concentration sensor, and oxygen concentration sensor are fixedly connected to the top of the scissor-type lifting mechanism, and the temperature sensor, carbon dioxide sensor, ammonia sensor, dust concentration sensor, and oxygen concentration sensor are electrically connected to the central data processing unit.
[0009] Preferably, the positioning and identification unit includes a high-definition camera module and a fill light, and the high-definition camera module and the fill light are fixedly connected to the top of the scissor-type lifting mechanism, and the high-definition camera module and the fill light are electrically connected to the central data processing unit.
[0010] Preferably, the motion mechanism includes a base, a DC motor module is arranged at the bottom of the base, a plurality of infrared ranging sensor modules are fixedly connected to the side wall of the base, and the infrared ranging sensor module and the DC motor module are electrically connected to the central data processing unit.
[0011] Preferably, the scissor-type lifting mechanism includes a fixed base, the bottom end of the fixed base is fixedly connected to the top end of the base, the top end of the fixed base is movably connected to the bottom end of the scissor-type lifting bracket, the top end of the scissor-type lifting bracket is movably connected to the bottom end of the first lifting platform, the four corners of the top end of the first lifting platform are respectively fixedly connected to the bottom end of the intelligent adjustable lifting bracket, the four intelligent adjustable lifting brackets are parallel to each other and vertically arranged, and the second lifting platform is slidably connected between the four intelligent adjustable lifting brackets, the bottom end of the scissor-type lifting bracket is fixedly connected to the output end of the stepper motor slide module, the stepper motor slide module is fixedly connected to the top end of the base, the high-definition camera module, the fill light, and the central data processing unit are fixedly connected to the top end of the first lifting platform, and the high-definition camera, dual-axis servo gimbal, infrared thermal imager, sound auxiliary module, temperature sensor, carbon dioxide sensor, ammonia sensor, dust concentration sensor, and oxygen concentration sensor are fixedly connected to the top end of the second lifting platform.
[0012] Preferably, the process of the BP neural network processing sound information is:
[0013] S11, poultry sound data collection;
[0014] S12, poultry sound data preprocessing;
[0015] S13, poultry sound feature parameter extraction;
[0016] S14, calibration and establishment of training set;
[0017] S15, training BP neural network sound recognition model;
[0018] S16. Sound recognition model applied to poultry sound recognition.
[0019] Preferably, the process of the convolutional neural network processing image information is:
[0020] S21, poultry image data collection;
[0021] S22, calibration and establishment of training set;
[0022] S23, training convolutional neural network image recognition model;
[0023] S24. Image recognition model applied to poultry image recognition.
[0024] Compared with the prior art, the present invention has the following advantages and technical effects:
[0025] The present invention can efficiently and accurately collect three-dimensional environmental quality information, poultry behavior information and physiological status in large-scale three-dimensional multi-layer breeding poultry houses. After the collected information is processed by the big data of the background computer, the facilities in the poultry house can be timely regulated. After the sound, image and video information of the poultry are processed in real time by the central processing unit, the position of abnormal poultry or dead poultry can be timely located, and corresponding processing can be quickly made to prevent large-scale infection. The present invention can replace manual daily inspection work in high-frequency, long-term and full-coverage forms, greatly improve the inspection efficiency, and avoid the problem of the respiratory system of inspectors being affected by long-term exposure to complex environmental components. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative labor:
[0027] Figure 1 The structure of the present invention is schematically shown Figure 1 ;
[0028] Figure 2The structure of the present invention is schematically shown Figure 2 ;
[0029] Figure 3 It is a training flow chart of the BP neural network of the present invention;
[0030] Figure 4 This is a training flow chart of the convolutional neural network of the present invention.
[0031] Among them, 1. Motion mechanism; 101. Infrared ranging sensor module; 102. DC motor module; 2. Environmental data acquisition unit; 201. Temperature sensor; 202. Carbon dioxide sensor; 203. Ammonia sensor; 204. Dust concentration sensor; 205. Oxygen concentration sensor; 3. Scissor-type lifting mechanism; 301. Fixed base; 302. Scissor-type lifting bracket; 303. First lifting platform; 304. Second lifting platform; 305. Intelligent adjustable lifting bracket; 306. Stepper motor slide module; 4. Positioning and identification unit; 401. High-definition camera module; 402. Fill light; 5. Poultry behavior detection unit; 501. High-definition camera; 502. Dual-axis servo gimbal; 503. Infrared thermal imager; 504. Sound auxiliary module; 6. Central data processing unit. DETAILED DESCRIPTION
[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0033] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0034] Reference Figure 1-Figure 4 The present invention provides a robot for automatically inspecting three-dimensional caged poultry, comprising a motion mechanism 1, a scissor-type lifting mechanism 3 is fixedly connected to the top of the motion mechanism 1, an environmental data acquisition unit 2, a positioning and identification unit 4, a poultry behavior detection unit 5, and a central data processing unit 6 are fixedly connected to the top of the scissor-type lifting mechanism 3, the motion mechanism 1, the environmental data acquisition unit 2, the scissor-type lifting mechanism 3, the positioning and identification unit 4, the poultry behavior detection unit 5 and the central data processing unit 6 are electrically connected, the poultry behavior detection unit 5 is used to monitor the sound information, image information, and video information of the poultry, and the sound information, image information, and video information are transmitted to the central data processing unit 6, and the central data processing unit 6 is used to process BP neural network, convolutional neural network, sound information, image information, and video information.
[0035] Further optimization scheme, the poultry behavior detection unit 5 includes a high-definition camera 501, a dual-axis steering gear pan-tilt 502, an infrared thermal imager 503, and a sound auxiliary module 504. The high-definition camera 501, the dual-axis steering gear pan-tilt 502, the infrared thermal imager 503, and the sound auxiliary module 504 are fixedly connected to the top of the scissor-type lifting mechanism 3, and the high-definition camera 501, the dual-axis steering gear pan-tilt 502, the infrared thermal imager 503, and the sound auxiliary module 504 are electrically connected to the central data processing unit 6.
[0036] The high-definition camera 501 and the sound auxiliary module 504 work together to solve the problem of image adhesion or obstruction of the poultry's body position during the shooting process of the high-definition camera 501.
[0037] The dual-axis servo gimbal 502 cooperates with the high-definition camera 501 to take photos and videos; the infrared thermal imager 503 can detect the body temperature characteristics of the poultry in the cage in real time.
[0038] The dual-axis servo gimbal 502 can autonomously adjust the shooting angle of the high-definition camera 501 to extract the poultry pictures in the cage more clearly and completely.
[0039] The sound auxiliary module 504 has two uses; on the one hand, it is used to record the sounds of poultry at night; on the other hand, after the pictures taken by the high-definition camera 501 in the poultry behavior detection unit 5 are processed by the central data processing unit 6 and the poultry is found to be blocked, the sound auxiliary module 504 can emit a sound of a certain frequency to stimulate the poultry to move in the cage, and the high-definition camera 501 can perform secondary photo recognition.
[0040] Further optimization scheme, the environmental data acquisition unit 2 includes a temperature sensor 201, a carbon dioxide sensor 202, an ammonia sensor 203, a dust concentration sensor 204, and an oxygen concentration sensor 205. The temperature sensor 201, the carbon dioxide sensor 202, the ammonia sensor 203, the dust concentration sensor 204, and the oxygen concentration sensor 205 are fixedly connected to the top of the scissor-type lifting mechanism 3, and the temperature sensor 201, the carbon dioxide sensor 202, the ammonia sensor 203, the dust concentration sensor 204, and the oxygen concentration sensor 205 are electrically connected to the central data processing unit 6.
[0041] The temperature sensor 201, the carbon dioxide sensor 202, the ammonia sensor 203, the dust concentration sensor 204, and the oxygen concentration sensor 205 continuously test the environmental information during the inspection process. When the environmental parameters exceed the set threshold, the inspection robot reports the exceeded parameters and positions to the central data processing unit 6 to implement environmental parameter early warning. The central data processing unit 6 is preferably a computer.
[0042] According to a further optimization scheme, the positioning and identification unit 4 includes a high-definition camera module 401 and a fill light 402. The high-definition camera module 401 and the fill light 402 are fixedly connected to the top of the scissor-type lifting mechanism 3. The high-definition camera module 401 and the fill light 402 are electrically connected to the central data processing unit 6.
[0043] The high-definition camera module 401 is used to capture the cage QR code information and the egg-laying status of the egg-laying poultry in the cage, and transmit it to the central data processing unit 6 .
[0044] The high-definition camera module 401 and the fill light 402 cooperate to shoot the QR code pasted on the three-dimensional breeding cage, and the central data processing unit 6 reads the cage position information in the QR code and stores the record. First, the cage position information in the QR code information is combined with the environmental quality information collected by the environmental data collection unit 2 to produce a three-dimensional model of the indoor environmental quality, so that the management personnel can better control the mechanical system in the poultry house. Secondly, the QR code can contain important information such as the number of poultry in the cage and the batch. Finally, the position of the QR code can enable the inspection robot to accurately locate.
[0045] According to a further optimization scheme, the motion mechanism 1 includes a base, a DC motor module 102 is arranged at the bottom of the base, a plurality of infrared ranging sensor modules 101 are fixedly connected to the side wall of the base, and the infrared ranging sensor module 101 and the DC motor module 102 are electrically connected to the central data processing unit 6.
[0046] The number of infrared distance sensor modules 101 is preferably three, which are respectively arranged in front and on both sides of the base. The infrared distance sensor modules 101 are used to detect the distance information between the robot and the surrounding obstacles, the cages on both sides and the walls in the poultry house to ensure that the robot can walk freely in the poultry house.
[0047] Further optimization scheme, the scissor-type lifting mechanism 3 includes a fixed base 301, the bottom end of the fixed base 301 is fixedly connected to the top end of the base, the top end of the fixed base 301 is movably connected to the bottom end of the scissor-type lifting bracket 302, the top end of the scissor-type lifting bracket 302 is movably connected to the bottom end of the first lifting platform 303, the top four corners of the first lifting platform 303 are respectively fixedly connected to the bottom end of the intelligent adjustable lifting bracket 305, the four intelligent adjustable lifting brackets 305 are parallel to each other and vertically arranged, and the second lifting platform 304 is slidably connected between the four intelligent adjustable lifting brackets 305. The bottom end of the bracket 302 is fixedly connected to the output end of the stepper motor slide module 306, the stepper motor slide module 306 is fixedly connected to the top of the base, the high-definition camera module 401, the fill light 402, and the central data processing unit 6 are fixedly connected to the top of the first lifting platform 303, and the high-definition camera 501, the dual-axis servo gimbal 502, the infrared thermal imager 503, the sound auxiliary module 504, the temperature sensor 201, the carbon dioxide sensor 202, the ammonia sensor 203, the dust concentration sensor 204, and the oxygen concentration sensor 205 are fixedly connected to the top of the second lifting platform 304.
[0048] The heights of the first lifting platform 303 and the second lifting platform 304 are infinitely adjusted by the stepper motor slide module 306 to realize inspection between each layer in the cage breeding mode.
[0049] To further optimize the solution, the process of BP neural network processing sound information is as follows:
[0050] S11, poultry sound data collection: the sounds made by poultry at night are obtained through the sound auxiliary module 504.
[0051] S12, poultry sound data preprocessing; for the poultry sound data obtained in S1, the quality of the poultry sound data is improved and the amount of sound data is reduced through pre-emphasis, frame windowing and voice endpoint detection preprocessing operations.
[0052] S13, poultry sound feature parameter extraction; for the poultry sound data in S2, after Fourier transform, the power spectrum of the poultry sound is obtained, Mel filter is performed, and the filtered logarithm is discrete cosine transformed to obtain the static and dynamic characteristics of the poultry sound, and finally the Mel frequency cepstrum coefficients of the poultry sound are obtained.
[0053] S14, calibration and establishment of training set; experienced breeding personnel calibrate the poultry sound data recorded in S1, judge whether the poultry is in a state of heat stress, hunger or normal state by listening to the sounds of the poultry, classify and calibrate the poultry sound fragments, and establish a training set for the BP neural network.
[0054] S15, training a BP neural network sound recognition model; training a BP neural network sound recognition model using the training set of poultry sound data obtained after S1 to S4.
[0055] S16. The sound recognition model is applied to poultry sound recognition. The trained BP neural network sound recognition model is used to recognize the sounds of poultry at night in a three-dimensional cage breeding mode.
[0056] The number of neurons in the input layer, output layer, and hidden layer of the BP neural network are 3, 2, and 4 respectively; the weights from the first layer of the input layer to the second layer of the hidden layer are set to [100, 100, 100], and the weights from the second layer of the input layer to the output layer are set to [3, 3, 3]; the original Sigmoid activation function is changed to the Swish activation function to optimize the accuracy of the BP neural network; the initial learning rate is set to 0.001 using the adaptive learning rate rule; the number of iterations is set to 10000 using the strategy of automatically saving the optimal model during training.
[0057] To further optimize the solution, the process of convolutional neural network processing image information is as follows:
[0058] S21, poultry image data collection: the sound auxiliary module 504 emits a specific frequency sound to the poultry in the three-dimensional cage, so that the poultry in the cage produces stress behavior, and then the high-definition camera 501 cooperates with the dual-axis steering gear pan / tilt 502 to collect images of the poultry producing stress behavior, and adjusts the size of the target image to n*n pixels;
[0059] S22, calibration and establishment of training set; manually framing and calibrating the stressed poultry pictures obtained in S21 as training set for convolutional neural network; convolutional neural network YOLOv5 network;
[0060] S23, train the convolutional neural network image recognition model; train the YOLOv5 network using the image training set of S22;
[0061] S24. Image recognition model is applied to poultry image recognition. The trained YOLOv5 network image recognition model is used to detect sick and dead poultry in a three-dimensional cage model.
[0062] The convolutional neural network is mainly composed of CSPDarknet and PANet. The image with a size of 640×640×3 is subjected to multiple 3×3 and 1×1 convolution operations, pooling dimensionality reduction, residual operation, and fully connected classification to obtain the feature results. The Adam optimizer is used for back propagation of weights. Its loss function includes three parts: target classification loss, bounding box coordinate loss, and target confidence loss. The loss function calculation value is the sum of the above three types of loss values.
[0063] The robot for automatically inspecting poultry in three-dimensional cages of the present invention monitors the environmental quality of poultry houses in three-dimensional cage mode in real time by building multiple environmental sensors. A sound recognition model based on BP neural network is used to judge whether poultry has stress behavior in a relatively quiet night, and an image recognition model based on YOLOv5 neural network is used to detect sick and dead poultry in the cage. The combination of the sound recognition model and the image recognition model can achieve real-time monitoring of the behavior of poultry in three-dimensional cage mode, effectively solve the problems encountered during manual inspection, and improve breeding efficiency.
[0064] In the description of the present invention, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside" and "outside" etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.
[0065] The embodiments described above are only descriptions of the preferred modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.
Claims
1. A robot for automatically inspecting poultry in three-dimensional cages, characterized in that: The invention comprises a motion mechanism (1), the top of which is fixedly connected to a scissor-type lifting mechanism (3), the top of which is fixedly connected to an environmental data acquisition unit (2), a positioning and identification unit (4), a poultry behavior detection unit (5), and a central data processing unit (6), the motion mechanism (1), the environmental data acquisition unit (2), the scissor-type lifting mechanism (3), the positioning and identification unit (4), the poultry behavior detection unit (5), and the central data processing unit (6), the poultry behavior detection unit (5) being electrically connected to the central data processing unit (6), the poultry behavior detection unit (5) being used to monitor the sound information, image information, and video information of the poultry, the sound information, image information, and video information being transmitted to the central data processing unit (6), and the central data processing unit (6) being used to process a BP neural network, a convolutional neural network, the sound information, image information, and video information; The process of the BP neural network processing sound information is as follows: S11, poultry sound data collection; Acquiring the sounds made by poultry at night through a sound auxiliary module (504); S12, poultry sound data preprocessing; for the poultry sound data obtained in S11, the quality of the poultry sound data is improved and the amount of sound data is reduced by pre-emphasis, frame windowing and voice endpoint detection preprocessing operations; S13, poultry sound feature parameter extraction; after the poultry sound data in S12 is subjected to Fourier transform, the power spectrum of the poultry sound is obtained, and the Mel filter is filtered, and the obtained logarithm after filtering is subjected to discrete cosine transform to obtain the static and dynamic characteristics of the poultry sound, and finally the Mel frequency cepstrum coefficient of the poultry sound is obtained therefrom S14, calibration and establishment of training set: experienced breeding personnel calibrate the poultry sound data recorded in S11, judge whether the poultry is in a heat stress state, a hungry state or a normal state by listening to the poultry sounds, classify and calibrate the poultry sound segments, and establish a training set of BP neural network; S15, training a BP neural network sound recognition model; training a BP neural network sound recognition model using the training set of poultry sound data obtained after S11 to S14; S16, the sound recognition model is applied to poultry sound recognition; the trained BP neural network sound recognition model is used to recognize the nighttime poultry sounds in a three-dimensional cage mode; The number of neurons in the input layer, output layer, and hidden layer of the BP neural network is 3, 2, and 4 respectively; the weights from the first layer of the input layer to the second layer of the hidden layer are set to [100, 100, 100], and the weights from the second layer of the input layer to the output layer are set to [3, 3, 3]; the original Sigmoid activation function is changed to the Swish activation function to optimize the accuracy of the BP neural network; the initial learning rate is set to 0.001 using the adaptive learning rate rule; the number of iterations is set to 10000 using the strategy of automatically saving the optimal model during training; The process of the convolutional neural network processing image information is as follows: S21, poultry image data collection; using a sound auxiliary module (504) to emit a specific frequency sound to the poultry in the three-dimensional cage, so that the poultry in the cage produces stress behavior, and then using a high-definition camera (501) in conjunction with a dual-axis steering gear pan / tilt (502) to collect images of the poultry producing stress behavior, and adjusting the size of the target image to n*n pixels; S22, calibration and establishment of training set; manually framing and calibrating the stressed poultry pictures obtained in S21 as training set for convolutional neural network; convolutional neural network YOLOv5 network; S23, train the convolutional neural network image recognition model; train the YOLOv5 network using the image training set of S22; S24. Image recognition model is applied to poultry image recognition; the trained YOLOv5 network image recognition model is used for the detection of sick and dead poultry in three-dimensional cage mode; The convolutional neural network is mainly composed of CSPDarknet and PANet. The image with a size of 640×640×3 is subjected to multiple 3×3 and 1×1 convolution operations, pooling dimensionality reduction, residual operation, and full connection classification to obtain the feature results; The Adam optimizer is used for back propagation of weights; Its loss function consists of three parts: target classification loss, bounding box coordinate loss, and target confidence loss. The loss function calculation value is the sum of the above three types of loss values. The sound recognition model based on BP neural network is used to judge the stress behavior of poultry in the relatively quiet environment at night, and the image recognition model based on YOLOv5 neural network is used to detect sick and dead poultry in cages; By combining the sound recognition model with the image recognition model, the behavior of poultry in the three-dimensional cage mode can be monitored in real time. The poultry behavior detection unit (5) comprises a high-definition camera (501), a dual-axis steering platform (502), an infrared thermal imager (503), and a sound auxiliary module (504); the high-definition camera (501), the dual-axis steering platform (502), the infrared thermal imager (503), and the sound auxiliary module (504) are fixedly connected to the top of the scissor-type lifting mechanism (3); and the high-definition camera (501), the dual-axis steering platform (502), the infrared thermal imager (503), and the sound auxiliary module (504) are electrically connected to the central data processing unit (6); The sound auxiliary module (504) has two purposes: on the one hand, it is used to record the sounds of poultry at night; on the other hand, after the pictures taken by the high-definition camera (501) in the poultry behavior detection unit (5) are processed by the central data processing unit (6) and the poultry is found to be blocking, the sound auxiliary module (504) can emit a sound of a certain frequency to stimulate the poultry to move in the cage, and the high-definition camera (501) can perform secondary photo recognition; The environmental data acquisition unit (2) comprises a temperature sensor (201), a carbon dioxide sensor (202), an ammonia sensor (203), a dust concentration sensor (204), and an oxygen concentration sensor (205); the temperature sensor (201), the carbon dioxide sensor (202), the ammonia sensor (203), the dust concentration sensor (204), and the oxygen concentration sensor (205) are fixedly connected to the top of the scissor-type lifting mechanism (3); and the temperature sensor (201), the carbon dioxide sensor (202), the ammonia sensor (203), the dust concentration sensor (204), and the oxygen concentration sensor (205) are electrically connected to the central data processing unit (6); The positioning identification unit (4) comprises a high-definition camera module (401) and a fill light (402), wherein the high-definition camera module (401) and the fill light (402) are fixedly connected to the top of the scissor-type lifting mechanism (3), and the high-definition camera module (401) and the fill light (402) are electrically connected to the central data processing unit (6); The high-definition camera module (401) is used to capture the cage position QR code information and the egg-laying status of the egg-laying poultry in the cage, and transmit the information to the central data processing unit (6); The high-definition camera module (401) and the fill light (402) cooperate to shoot the QR code pasted on the three-dimensional breeding cage, and the central data processing unit (6) reads the cage position information in the QR code and stores the record; firstly, the cage position information in the QR code information is combined with the environmental quality information collected by the environmental data collection unit (2) to produce a three-dimensional model of the environmental quality in the house, so that the management personnel can better control the mechanical system in the poultry house; secondly, the QR code can contain the number of poultry in the cage and important batch information; finally, the position of the QR code enables the inspection robot to accurately locate; The motion mechanism (1) comprises a base, a DC motor module (102) is arranged at the bottom of the base, a plurality of infrared distance measuring sensor modules (101) are fixedly connected to the side wall of the base, and the infrared distance measuring sensor module (101) and the DC motor module (102) are electrically connected to the central data processing unit (6); The scissor-type lifting mechanism (3) comprises a fixed base (301), the bottom end of the fixed base (301) is fixedly connected to the top end of the base, the top end of the fixed base (301) is movably connected to the bottom end of the scissor-type lifting bracket (302), the top end of the scissor-type lifting bracket (302) is movably connected to the bottom end of the first lifting platform (303), the top four corners of the first lifting platform (303) are respectively fixedly connected to the bottom end of the intelligent adjustable lifting bracket (305), the four intelligent adjustable lifting brackets (305) are parallel to each other and vertically arranged, the four intelligent adjustable lifting brackets (305) are slidably connected to the second lifting platform (304), the scissor-type lifting brackets (302 ) is fixedly connected to the output end of the stepper motor slide module (306), the stepper motor slide module (306) is fixedly connected to the top of the base, the high-definition camera module (401), the fill light (402), and the central data processing unit (6) are fixedly connected to the top of the first lifting platform (303), and the high-definition camera (501), the dual-axis servo gimbal (502), the infrared thermal imager (503), the sound auxiliary module (504), the temperature sensor (201), the carbon dioxide sensor (202), the ammonia sensor (203), the dust concentration sensor (204), and the oxygen concentration sensor (205) are fixedly connected to the top of the second lifting platform (304).
