Abnormal chicken cage identification method, device, system and robot

Abstract

The application provides an abnormal chicken cage recognition method, device, system and robot, and belongs to the technical field of livestock and poultry breeding. The method comprises the following steps: acquiring a sampling image of a target chicken cage; recognizing the sampling image to obtain a first recognition result and a second recognition result respectively; the first recognition result comprises the number of chicken heads, the number of eggs and the comb erecting state of each chicken in the target chicken cage; the second recognition result comprises the comb color information of each chicken; and the first recognition result and the second recognition result are input into a decision tree model to determine the category of the target chicken cage. The abnormal chicken cage recognition method, device, system and robot provided by the application are based on multi-sensor coordination, multi-source data fusion, the individual chickens in the chicken cage are subjected to comprehensive extraction of internal and external features, and then are sent into a decision tree judgment model for classification, so that the data is more reliable and the classification result is more accurate.

Description

Technical Field

[0001] This invention relates to the field of livestock and poultry breeding technology, and in particular to methods, devices, systems and robots for identifying abnormal chicken cages. Background Technology

[0002] Large-scale farming has become the main farming model in recent years. During the farming process, dead chickens and low-producing hens can seriously affect the farming efficiency of enterprises. Therefore, timely identification and removal of these dead chickens is of great significance.

[0003] Manually inspecting dead chickens and low-laying hens is labor-intensive, inaccurate, and slow to detect problems. Therefore, it is particularly important to replace manual labor with intelligent inspection equipment.

[0004] Currently, there are detection and identification devices for dead chickens and low-laying hens on the market, but most of them have the following shortcomings: limited classification of vital signs, low recognition rate, and lack of scalability. Therefore, developing efficient and accurate identification and location methods using new technologies has become an urgent need. Summary of the Invention

[0005] This invention provides a method, device, system, and robot for identifying abnormal chicken cages, which solves the problems of high labor intensity and low identification accuracy in the prior art that requires manual identification of dead chickens and low-producing chickens, and achieves accurate identification and location of chicken cages with dead chickens and low-producing chickens.

[0006] In a first aspect, the present invention provides a method for identifying abnormal chicken cages, comprising: acquiring a sampled image of the target chicken cage;

[0007] The sampled image is identified to obtain a first identification result and a second identification result respectively; the first identification result includes the number of chicken heads, the number of eggs, and the upright state of the combs of each chicken in the target chicken cage; the second identification result includes the comb color information of each chicken.

[0008] The first and second identification results are input into the decision tree model to determine the category of the target chicken cage;

[0009] The categories include low-laying hen cages, normal hen cages, and dead hen cages.

[0010] According to the present invention, a method for identifying abnormal chicken coops, identifying the sampled image and obtaining the first identification result includes:

[0011] The sampled image is input into a pre-trained chicken head detection network model to obtain multiple chicken head sub-images in the sampled image, and the number of chicken head sub-images is taken as the number of chicken heads in the target chicken cage;

[0012] The sampled image is input into a pre-trained egg detection network model to obtain multiple egg sub-images in the sampled image, and the number of the egg sub-images is taken as the number of eggs in the target chicken coop;

[0013] The sampled image is input into a pre-trained rooster comb detection network model to obtain multiple rooster comb sub-images in the sampled image, and the upright state of the rooster comb in each rooster comb sub-image is determined.

[0014] According to the present invention, a method for identifying abnormal chicken coops, comprising identifying the sampled image and obtaining the second identification result, specifically includes:

[0015] The sampled image is input into a pre-trained rooster comb segmentation network model to obtain at least one rooster comb sub-image segmented from the sampled image by the output of the rooster comb segmentation network model;

[0016] Based on the distribution of RGB values ​​of each chicken comb image, the comb color information of each chicken is determined.

[0017] The method for identifying abnormal chicken cages according to the present invention further includes: acquiring the sound information of each chicken in the target chicken cage;

[0018] Accordingly, the sound information, the first recognition result, and the second recognition result are input into the decision tree model to determine the category of the target chicken cage.

[0019] According to the present invention, a method for identifying abnormal chicken cages, wherein acquiring the sound information of each chicken in the target chicken cage includes:

[0020] Collect the sound signals of each chicken in the target chicken cage within a preset time period;

[0021] The characteristic parameters related to the sound signal are obtained. These characteristic parameters are linear predictive cepstral coefficients extracted after the start and end points of the sound signal are determined by analyzing the zero-crossing rate and energy of the sound signal.

[0022] The feature parameters are input into a pre-trained sound classification model to obtain the sound information output by the sound classification model;

[0023] The sound classification model is obtained by training a Hidden Markov Model as the initial model.

[0024] The method for identifying abnormal chicken cages according to the present invention further includes: acquiring the body temperature information of each chicken in the target chicken cage;

[0025] Accordingly, the sound information, the body temperature information, the first recognition result, and the second recognition result are input into the decision tree model to determine the category of the target chicken cage.

[0026] According to a method for identifying abnormal chicken cages provided by the present invention, the step of obtaining the body temperature information of each chicken in the target chicken cage includes:

[0027] The body surface temperature of each chicken was collected at multiple sampling times;

[0028] Based on the least squares method, a binary linear regression was performed on the body surface temperature of each chicken at the multiple sampling times to obtain the body temperature information of each chicken.

[0029] According to a method for identifying abnormal chicken cages provided by the present invention, the step of inputting the sound information, the body temperature information, the first identification result, and the second identification result into a decision tree model to determine the category of the target chicken cage includes:

[0030] Based on a pre-constructed feature data table for determining the category of chicken cages, the feature data table records the mapping relationship between sound information, body temperature information, matching degree of chicken head number, matching degree of egg number, chicken comb upright status and chicken comb color information and the category of chicken cages.

[0031] Using the feature data table, a confusion matrix table is generated based on the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state, and chicken comb color information related to the target chicken cage, so as to calculate the probability that the target chicken cage belongs to any category based on the confusion matrix table;

[0032] Based on the probability that the target chicken cage belongs to any category, the information entropy corresponding to the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state and chicken comb color information are determined respectively.

[0033] Based on the information entropy corresponding to the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state, and chicken comb color information, the information gain corresponding to the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state, and chicken comb color information is determined respectively.

[0034] The information gains corresponding to the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state, and chicken comb color information are normalized respectively to determine the weights corresponding to the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state, and chicken comb color information.

[0035] Based on the matching degree of the number of chicken heads and its corresponding weight, and the body temperature information and its corresponding weight, calculate the confidence level that the target chicken cage is a dead chicken cage;

[0036] Based on the matching degree of the number of eggs and its corresponding weight, the upright state of the rooster comb and its corresponding weight, the rooster comb color information and its corresponding weight, the body temperature information and its corresponding weight, and the sound information and its corresponding weight, the confidence level of classifying the target chicken cage as a low-laying chicken cage is calculated.

[0037] The category of the target chicken cage is determined based on the confidence level that the target chicken cage is a dead chicken cage and the confidence level that the target chicken cage is a low-laying chicken cage.

[0038] The chicken head count matching degree is determined by comparing the number of chicken heads identified in the target chicken cage with the standard number of chicken heads in the target chicken cage. The egg count matching degree is determined by comparing the number of eggs identified in the target chicken cage with the standard number of eggs in the target chicken cage.

[0039] According to the present invention, an abnormal chicken cage identification method is provided, wherein each chicken cage in the chicken house is provided with a QR code of the same shape in the upper left corner, and each QR code is used to store the location information of the chicken cage.

[0040] The acquisition of the sampled image of the target chicken cage includes:

[0041] Acquire any image of a chicken coop captured by the sampling camera;

[0042] The coordinate information of two adjacent QR codes in the chicken coop image is obtained to determine the coordinate information of the target chicken cage in the chicken coop image; the target chicken cage is the chicken cage included in the area defined by the same side of the two adjacent QR codes in the chicken coop image.

[0043] The sampled image of the target chicken cage is cropped from the chicken coop image based on the coordinate information of the chicken cage.

[0044] According to the present invention, a method for identifying abnormal chicken coops includes obtaining the coordinate information of two adjacent QR codes in a chicken coop image to determine the coordinate information of the target chicken coop in the chicken coop image, comprising:

[0045] Based on the coordinate information of each QR code in the chicken coop image, combined with the optical center coordinates of the target chicken cage from the sampling camera, the intrinsic parameters of the sampling camera, and the focusing distance of the target chicken cage, the world coordinate information of each QR code is determined;

[0046] Based on the world coordinate information of the two QR codes, the coordinate information of the target chicken coop is determined.

[0047] According to the present invention, a method for identifying abnormal chicken cages, after determining the category of the target chicken cage, further includes:

[0048] Locate the virtual area of ​​the target chicken coop in the virtual chicken coop map, and mark the category of the target chicken coop within the virtual area.

[0049] Secondly, the present invention also provides an identification device for abnormal chicken cages, comprising:

[0050] Image acquisition unit, used to acquire sampled images of the target chicken cage;

[0051] An image detection unit is used to identify the sampled image to obtain a first identification result and a second identification result respectively; the first identification result includes the number of chicken heads, the number of eggs, and the upright state of the combs of each chicken in the target chicken cage; the second identification result includes the comb color information of each chicken.

[0052] The chicken cage classification unit is used to input the first identification result and the second identification result into the decision tree model to determine the category of the target chicken cage;

[0053] The categories include low-laying hen cages, normal hen cages, and dead hen cages.

[0054] Thirdly, the present invention provides a robot including a main control unit, which, when running, implements the abnormal chicken cage identification method as described in any of the first aspects.

[0055] Fourthly, the present invention provides an abnormal chicken cage identification system, including the robot as described in the third aspect;

[0056] The robot is equipped with a sampling camera and an RFID reader, and magnetic strip tracks are laid in the chicken coop according to a set route; the robot can inspect the chicken coop according to the set route.

[0057] It also includes user terminals, as well as identification tags, QR codes, sound sensors, and temperature sensors installed on each chicken cage;

[0058] The sound sensor is used to collect the sound information of the chickens in each cage, and the temperature sensor is used to collect the temperature information of the chickens in each cage. Each identification tag stores the number information of the chicken cage it belongs to, as well as the sound information and temperature information collected at each sampling time by the sound sensor and the temperature sensor.

[0059] The robot is connected to the user terminal to transmit the determined category of the target chicken cage and the identity information obtained by the RFID reader to the user terminal in real time.

[0060] Fifthly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the above-described methods for identifying abnormal chicken cages.

[0061] In a sixth aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the abnormal chicken cage identification method as described above.

[0062] The abnormal chicken cage identification method, device, system and robot provided by the present invention are based on multi-sensor coordination and multi-source data fusion. After comprehensively extracting the internal and external features of individual chickens in the cage, the data is fed into the decision tree evaluation model for classification, which makes the data more reliable and the classification results more accurate. Attached Figure Description

[0063] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0064] Figure 1 This is one of the flowcharts illustrating the abnormal chicken cage identification method provided by the present invention;

[0065] Figure 2 This is the second flowchart of the abnormal chicken cage identification method provided by the present invention;

[0066] Figure 3 This is a schematic diagram of the training process for each network model provided by the present invention;

[0067] Figure 4 This is a schematic diagram of a virtual chicken coop provided by the present invention;

[0068] Figure 5 This is a schematic diagram of the structure of the abnormal chicken cage identification device provided by the present invention;

[0069] Figure 6 This is a schematic diagram of the structure of the robot provided by the present invention;

[0070] Figure 7This is a schematic diagram of the abnormal chicken cage identification system provided by the present invention;

[0071] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention;

[0072] The attached figures are labeled as follows:

[0073] 1: Virtual chicken coop diagram; 2: QR code; 3: RFID tag; 4: Sound sensor; 5: Robot; 6: Magnetic strip track; 7: Power wheel; 8: RFID reader; 9: Temperature sensor; 10: Sampling camera; 11: Rotating wheel. Detailed Implementation

[0074] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0075] It should be noted that in the description of the embodiments of the present invention, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. The terms "upper," "lower," etc., indicating orientation or positional relationships are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Unless otherwise expressly specified and limited, the terms "installed," "connected," and "linked" should be interpreted broadly, for example, as a fixed connection, a detachable connection, or an integral connection; a mechanical connection or an electrical connection; a direct connection or an indirect connection through an intermediate medium; or a connection within two elements. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0076] The following is combined with Figures 1-8 This invention describes the method, apparatus, system, and robot for identifying abnormal chicken cages provided in embodiments of the present invention.

[0077] Observations revealed that dead chickens found in cages, besides having lower body surface temperatures, typically had their heads buried at the bottom of the cage. Low-laying hens usually had upright combs with white spots or frost, or their combs were thinner than those of normal-laying hens. Based on these physiological characteristics, the abnormal chicken cage identification method provided by this invention utilizes multiple sensors and deep learning algorithms to intelligently identify and locate cages with dead chickens and / or low egg production by combining multiple surface features of the chickens with information such as the egg production of all chickens in the cage. Therefore, the abnormal chicken cage identification method provided by this invention can combine multiple surface features of the target object as the basis for location during the localization process, aiming to achieve the fusion of multiple information such as images, sounds, and temperatures of laying hens in large-scale farming, providing data support for the scientific management of chicken farms.

[0078] Figure 1 This is one of the flowcharts illustrating the abnormal chicken cage identification method provided by the present invention, such as... Figure 1 As shown, including but not limited to the following steps:

[0079] Step 101: Obtain a sampled image of the target chicken coop.

[0080] This invention allows workers to take images inside the chicken coop using a handheld sampling camera, or the sampling camera can be mounted on a mobile cart to take images inside the chicken coop. The mobile cart can be a robot that can patrol the chicken coop according to a preset inspection route and collect images of the chicken coop at a preset sampling frequency.

[0081] Furthermore, by cropping the images of the chicken coops collected, images of each chicken cage contained in each image can be cropped, and the image of any one of the chicken cages can be called the sampled image of the target chicken cage.

[0082] For example, if an image of a chicken coop contains images of three chicken cages, then the image of each chicken cage is cropped out separately from the image of the chicken coop. For any cropped image containing only one chicken cage, it is used as the sample image of the target chicken cage for analysis to determine whether there are dead chickens or low-laying chickens in that chicken cage.

[0083] Optionally, in the abnormal chicken cage identification method provided by this invention, the sampling camera model can be selected, its shooting parameters can be set appropriately, and the shooting distance between the sampling camera and the chicken cage can be adjusted to ensure that each image of the chicken coop captured by the sampling camera includes only one complete chicken cage. Simultaneously, the sampling frequency of the sampling camera is appropriately adjusted to match the robot's movement speed during inspection. This allows for the cropping of an image containing a complete chicken cage from the images of the chicken coop captured at each sampling moment during the robot's inspection process. This enables sequential analysis of the category of each chicken cage along the inspection route, avoiding repeated analysis of images of the same chicken cage, thus effectively improving detection efficiency.

[0084] Step 102: Identify the sampled image to obtain a first identification result and a second identification result respectively; the first identification result includes the number of chicken heads, the number of eggs, and the upright state of the combs of each chicken in the target chicken cage; the second identification result includes the comb color information of each chicken.

[0085] This invention mainly utilizes deep learning algorithms to achieve image recognition of sampled images of target chicken cages.

[0086] Specifically, the sampled image is input into a pre-built and trained deep neural network model, which uses the convolutional layers in the deep neural network model to extract the image features carried by the sampled image. Then, the fully connected layers and classifiers are used to analyze the extracted features and output the recognition result of the sampled image.

[0087] As an optional embodiment, the present invention may use only one deep neural network to recognize the sampled image and output the first recognition result and the second recognition result corresponding to the sampled image. Alternatively, multiple deep neural network models may be used to recognize the input sampled image respectively, so as to output the first recognition result and the second recognition result respectively. The present invention does not specifically limit the specific method.

[0088] It should be noted that the first recognition result obtained by identifying the sampled image of the target chicken cage mentioned above mainly includes, but is not limited to: the number of chicken heads, the number of eggs, and the upright state of the combs of each chicken in the target chicken cage.

[0089] Simultaneously, by identifying the sampled image of the target chicken coop, an image of the area containing the comb of each chicken inside the coop can be cropped from the sampled image; for ease of description, this is referred to as a comb sub-image. Then, RGB information can be extracted from each segmented comb sub-image to obtain the comb color information of the chicken corresponding to each comb sub-image.

[0090] It should be noted that, after using a rooster comb segmentation network model to distinguish the rooster comb sub-image from the background environment of the sampled image, the RGB values ​​of the rooster comb in the sub-image are extracted using the OpenCV library, and the RGB values ​​are clustered to find the RGB distribution characteristics of the rooster comb in low-laying hens. Then, a rooster comb detection network model is used to detect and identify whether the rooster comb is upright for classification. Finally, the feature data obtained above are normalized to provide a basis for the classification of chicken cage types.

[0091] Optionally, the above-mentioned normalization processing of feature data involves a calculation formula that can be expressed as follows:

[0092]

[0093] Where, x i For real-time collection of feature data; The normalized value; x min x is the theoretical minimum value of the feature data. max This represents the theoretical maximum value of the feature data.

[0094] Step 103: Input the first identification result and the second identification result into the decision tree model to determine the category of the target chicken cage.

[0095] The aforementioned chicken cages are mainly categorized into low-laying-producing chicken cages, normal chicken cages, and dead chicken cages.

[0096] Decision tree models are simple and easy-to-use nonparametric classifiers. They do not require any prior assumptions about the input data, are computationally fast, provide easily interpretable results, and are robust.

[0097] Optionally, the first and second identification results are used as influencing feature data for determining the category of the target chicken cage and input into a pre-trained decision tree model. The decision tree model outputs whether there are dead chickens and / or low-laying chickens in the target chicken cage, thereby determining the category of the target chicken cage.

[0098] It should be noted that if a dead chicken is found in the target chicken cage, it will be classified as a dead chicken cage; if a low-producing hen is found in the target chicken cage, it will be classified as a low-producing hen cage; if neither a low-producing hen nor a dead chicken is found in the target chicken cage, it will be classified as a normal chicken cage.

[0099] It should be noted that if both low-producing hens and dead hens are found in the target chicken cage, it will be classified as both a low-producing hen cage and a dead hen cage.

[0100] The abnormal chicken cage identification method, device, system and robot provided by the present invention are based on multi-sensor coordination and multi-source data fusion. After comprehensively extracting the internal and external features of individual chickens in the cage, the data is fed into the decision tree evaluation model for classification, which makes the data more reliable and the classification results more accurate.

[0101] Based on the above embodiments, as an optional embodiment, identifying the sampled image and obtaining the first identification result may include, but is not limited to, the following steps:

[0102] (1) Input the sampled image into a pre-trained chicken head detection network model to obtain multiple chicken head sub-images in the sampled image, and obtain the number of chicken head sub-images as the number of chicken heads in the target chicken cage.

[0103] (2) Input the sampled image into a pre-trained egg detection network model to obtain multiple egg sub-images in the sampled image, and obtain the number of egg sub-images as the number of eggs in the target chicken coop.

[0104] (3) Input the sampled image into a pre-trained rooster comb detection network model to obtain multiple rooster comb sub-images in the sampled image and determine the upright state of the rooster comb in each rooster comb sub-image.

[0105] It should be noted that in this embodiment of the invention, three different deep neural network models are used, including a chicken head detection network model, an egg detection network model, and a chicken comb detection network model, to identify and process the sampled images respectively, so as to obtain the number of chicken heads, the number of eggs, and the chicken comb image of each chicken in the target chicken cage.

[0106] As another alternative embodiment, the same deep neural network model can be pre-trained to identify the input sampled image and simultaneously obtain the number of chicken heads, the number of eggs, and the upright comb status of each chicken in the target chicken cage that can be observed.

[0107] It should be noted that since the heads of dead chickens found in chicken coops are usually buried at the bottom of the coop, the heads of dead chickens will not be counted when identifying the number of chicken heads in the sampled images.

[0108] The abnormal chicken cage identification method provided by this invention, with the help of a deep neural network model, can quickly and accurately perform image recognition on the sampled images of each chicken cage collected in real time, so as to obtain the number of chicken heads, the number of eggs, and the upright state of the chicken comb in each chicken cage, providing data support for subsequent automated judgment of whether there are dead chickens or low-producing chickens in the chicken cage.

[0109] Based on the above embodiments, as an optional embodiment, identifying the sampled image and obtaining the second identification result specifically includes, but is not limited to, the following steps:

[0110] The sampled image is input into a pre-trained rooster comb segmentation network model to obtain at least one rooster comb sub-image segmented from the sampled image by the rooster comb segmentation network model; the rooster comb color information of each chicken is determined according to the distribution of the RGB values ​​of each rooster comb sub-image.

[0111] Figure 2 This is the second flowchart illustrating the abnormal chicken cage identification method provided by the present invention, as follows: Figure 2 As shown, the present invention can use the Vision Transformer model as the basic model. After pre-training the model with a pre-built training set, the trained Vision Transformer model can be used to identify the output sampled image to obtain the number of chicken heads, the number of eggs, and the upright state of the combs of each chicken in the target chicken cage corresponding to the sampled image.

[0112] The process involves using the Vision Transformer model to identify the number of chicken heads and eggs in the target cage from the sampled images. The identified numbers are then compared to the standard numbers for chicken heads and eggs in the target cage, respectively. If the number of identified chicken heads is less than the standard number, the probability of dead chickens in the target cage is high; if the number is equal to the standard number, the probability is low. Similarly, if the number of identified eggs is less than the standard number, the probability of low-laying hens in the target cage is high; if the number is greater than or equal to the standard number, the probability is low.

[0113] Alternatively, the Mask R-CNN model can be used as the base model, and a rooster comb segmentation network model can be generated by pre-training it. By inputting the sampled image into the rooster comb segmentation network model, all rooster comb sub-images output after cropping the sampled image can be obtained. Then, RGB information is extracted from each cropped rooster comb sub-image to obtain the rooster comb color information corresponding to each rooster in each sub-image.

[0114] Finally, the number of chicken heads, the number of eggs, and the upright state of each chicken's comb, along with the comb color information of each chicken obtained later, can be used as multi-sensor data features and input into a pre-trained decision tree model. This allows the model to obtain the identification results of whether the target chicken cage contains dead chickens or low-laying chickens. Based on the identification results, it can then be determined whether the target chicken cage belongs to the category of low-laying chicken cage, dead chicken cage, or normal chicken cage.

[0115] Figure 3 This is a schematic diagram of the training process for each network model provided by the present invention, such as... Figure 3 As shown, this invention provides another method for pre-training the Vision Transformer model and the Mask R-CNN model, the steps of which include:

[0116] First, multiple sample images (each containing a certain number of chickens and eggs) are collected to construct a sample image set for pre-training. All sample images in the set are then preprocessed, such as removing poorly captured images. Next, data augmentation is performed on each preprocessed sample image, including but not limited to image cropping, image flipping, image scaling, and color gamut transformation, to expand the sample image set and improve the robustness of pre-training.

[0117] For each sampled image in the enhanced image set, data labels are created, annotating the coordinates of the chicken's head, egg, and comb, as well as the comb's color and upright position. Training labels are then constructed for each sampled image. Data standardization is performed by mapping these labels to each sampled image, and then sequentially inputting them into the Vision Transformer and Mask R-CNN models. Based on the discrepancy between the outputs of the Vision Transformer and Mask R-CNN models and the labels after each input, the model parameters of both models are adjusted until the output converges, yielding the final detection and segmentation network models.

[0118] The abnormal chicken cage identification method provided by this invention acquires images of chicken cages and performs image recognition based on a deep learning network model to obtain data features related to the presence of dead chickens and low-laying hens in the chicken cages. Then, a multi-data fusion method is used to determine whether there are dead chickens or low-laying hens in the chicken cages, thereby determining the category of each chicken cage and providing data support for the scientific management of chicken farms.

[0119] Based on the above embodiments, as an optional embodiment, the abnormal chicken cage identification method provided by the present invention further includes: acquiring the sound information of each chicken in the target chicken cage;

[0120] Accordingly, the sound information, the first recognition result, and the second recognition result are input into the decision tree model to determine the category of the target chicken cage.

[0121] The sound information of each chicken in the target chicken cage can be obtained by using a sound sensor installed above the target chicken cage.

[0122] Specifically, acquiring the sound information of each chicken in the target chicken coop includes:

[0123] Collect the sound signals of each chicken in the target chicken cage within a preset time period;

[0124] The characteristic parameters related to the sound signal are obtained. These characteristic parameters are linear predictive cepstral coefficients extracted after the start and end points of the sound signal are determined by analyzing the zero-crossing rate and energy of the sound signal.

[0125] The feature parameters are input into a pre-trained sound classification model to obtain the sound information output by the sound classification model;

[0126] The sound classification model is obtained by training a Hidden Markov Model as the initial model.

[0127] Specifically, the abnormal chicken cage identification method provided by the present invention uses sound recognition technology and identifies and classifies the sound signals of each chicken in the target chicken cage through the sound recognition method of Hidden Markov Model (HMM).

[0128] First, sound signals are collected within a preset time period using a sound sensor. Since the sound sensor is installed on the top of the target chicken coop, the zero-crossing rate and energy of the collected sound signals can be analyzed, and these two parameters can be used as the basis for determining the start and end points of the sound signals under different states. Then, the linear prediction cepstral coefficients of the sound information are extracted as feature parameters and input into the sound classification model to obtain the classification result of the sound information output by the sound classification model. This classification result can be loud, soft, or the sound can be a laying egg call, coughing sound, etc., and this invention does not specifically limit it.

[0129] It should be noted that the above sound classification model was obtained after training with a Hidden Markov Model as the initial model.

[0130] Combination Figure 2As shown, after obtaining the sound information of each chicken in the target chicken cage, the present invention classifies the sound information through a sound classification model. The obtained sound classification result can be used as a type of data feature. This feature is combined with the aforementioned information on the number of chicken heads, the number of eggs, the upright state of each chicken's comb, and the color of the comb. By using a multi-data fusion approach, the present invention can determine whether there are dead chickens or low-laying chickens in the chicken cage, and thus determine the category to which each chicken cage belongs, providing data support for the scientific management of chicken farms.

[0131] Based on the above embodiments, as an optional embodiment, the abnormal chicken cage identification method provided by the present invention further includes obtaining the body temperature information of each chicken in the target chicken cage;

[0132] Accordingly, the sound information, body temperature information, first recognition result, and second recognition result related to the target chicken cage are input into the decision tree model to determine the category of the target chicken cage.

[0133] The temperature sensor can be placed above the target chicken cage or on the robot used for chicken house inspection, enabling rapid measurement of the body temperature of chickens in each cage.

[0134] As an optional embodiment, the above-mentioned acquisition of the body temperature information of each chicken in the target chicken cage may specifically include:

[0135] The body surface temperature of each chicken was collected at multiple sampling times;

[0136] Based on the least squares method, a binary linear regression was performed on the body surface temperature of each chicken at the multiple sampling times to obtain the body temperature information of each chicken.

[0137] This invention utilizes a temperature sensor to detect the body temperature of chickens. To minimize the stress on chickens caused by traditional manual temperature measurement, the temperature sensor used in this invention can be a non-contact infrared temperature transmitter. This infrared temperature transmitter uses RS485 for data transmission to upload the body surface temperature collected at each sampling moment to the main control unit in real time and associate it with an RFID tag.

[0138] In actual temperature measurement, the measurement sites will vary for different livestock and poultry. When measuring the temperature of chickens, the main measurement sites are the abdomen and head.

[0139] After obtaining the body surface temperature of each chicken in the target cage and integrating it into temperature information, it is bound to the RFID tag of each corresponding chicken cage. This allows for the recording of the body temperature changes of the chickens before and after laying eggs. By combining the body temperature patterns of each chicken, the possibility of dead chickens and low-laying chickens in each chicken cage can be determined.

[0140] As an optional embodiment, after obtaining the body surface temperature of each chicken at continuous sampling time and using DB wavelet preprocessing, a BP neural network model is established to classify and study the temperature changes before and after laying eggs. This allows the determination of the temperature status of each chicken in the observed cage before and after laying eggs, and the judgment of the current health status and production performance of the chickens.

[0141] Furthermore, since chickens exhibit regular changes in body temperature before and after laying eggs, these temperature changes can be compared with the chickens' historical body temperatures to identify any abnormalities in the chicken's health or egg production. However, due to the significant error between surface temperature and actual body temperature, it is insufficient to reflect normal body temperature. Therefore, this invention uses surface temperature measured by a temperature sensor at multiple sampling times as a variable. Least squares regression analysis is employed to obtain the most accurate value, which is then compared with the historical body temperature data of each chicken and the body temperature data of other chickens in the cage to determine whether the laying hen has died or is laying eggs abnormally.

[0142] Optionally, the calculation formula for binary linear regression analysis is specifically designed as follows:

[0143]

[0144] Where a and b are regression coefficients; The variables x are respectively i y i The average value; n is the total number of samples.

[0145] It should be noted that the present invention associates the body surface temperature data of each chicken in the target chicken cage collected at each sampling time with the RFID tag. As an optional embodiment, the sound information of all chickens in each chicken cage, the upright status of the chickens' combs, and the number information of each chicken cage can also be saved and associated with the RFID tag.

[0146] In this way, when the robot inspects the target chicken cage, it can read the aforementioned data information in real time by recognizing the RFID tag through its onboard RFID reader. At the same time, it can input the chicken comb color information obtained by segmenting and recognizing the sampled image, along with all the data information read above, into the decision tree model in its control unit, thereby determining the category of the target chicken cage in real time.

[0147] Based on the above embodiments, as an optional embodiment, the step of inputting the sound information, the body temperature information, the first recognition result, and the second recognition result into a decision tree model to determine the category of the target chicken coop includes:

[0148] Based on a pre-constructed feature data table for determining the category of chicken cages, the feature data table records the mapping relationship between sound information, body temperature information, matching degree of chicken head number, matching degree of egg number, chicken comb upright status and chicken comb color information and the category of chicken cages.

[0149] Using the feature data table, a confusion matrix table is generated based on the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state, and chicken comb color information related to the target chicken cage, so as to calculate the probability that the target chicken cage belongs to any category based on the confusion matrix table;

[0150] Based on the probability that the target chicken cage belongs to any category, the information entropy corresponding to the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state and chicken comb color information are determined respectively.

[0151] Based on the information entropy corresponding to the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state, and chicken comb color information, the information gain corresponding to the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state, and chicken comb color information is determined respectively.

[0152] The information gains corresponding to the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state, and chicken comb color information are normalized respectively to determine the weights corresponding to the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state, and chicken comb color information.

[0153] Based on the matching degree of the number of chicken heads and its corresponding weight, and the body temperature information and its corresponding weight, calculate the confidence level that the target chicken cage is a dead chicken cage;

[0154] Based on the matching degree of the number of eggs and its corresponding weight, the upright state of the rooster comb and its corresponding weight, the rooster comb color information and its corresponding weight, the body temperature information and its corresponding weight, and the sound information and its corresponding weight, the confidence level of classifying the target chicken cage as a low-laying chicken cage is calculated.

[0155] The category of the target chicken cage is determined based on the confidence level that the target chicken cage is a dead chicken cage and the confidence level that the target chicken cage is a low-laying chicken cage.

[0156] The chicken head count matching degree is determined by comparing the number of chicken heads identified in the target chicken cage with the standard number of chicken heads in the target chicken cage. The egg count matching degree is determined by comparing the number of eggs identified in the target chicken cage with the standard number of eggs in the target chicken cage.

[0157] Specifically, this invention is the first to use a feature selection method to weight the acquired sound information, body temperature information, first recognition results (mainly including the number of chicken heads, the number of eggs, and the upright state of the chicken combs) and second recognition results (mainly including the color information of the chicken combs) and other multi-sensor feature data. It has the characteristics of being simple and practical, robust, and having good real-time performance.

[0158] For example, by calculating the information gain, different weights are assigned to the feature data corresponding to the body temperature sensor, sound sensor, and RFID tag to make judgments on the behavior of laying hens; by comparing the number of chicken heads in each cage with the standard number of chicken heads to obtain the matching degree of the number of chicken heads, cages with abnormal life are screened out, and then combined with the temperature information and sound information of each chicken in each cage, it is determined whether there are dead chickens.

[0159] At the same time, by integrating the matching degree of the number of eggs in each chicken cage, the comb color information of each chicken in the chicken cage, and audio and temperature information obtained from other sensors, abnormal egg production is judged, and finally, classification indicators of dead chickens and low-producing chickens are generated for users to select and call.

[0160] The abnormal chicken cage identification method provided by this invention can effectively improve the identification rate of chicken cage classification by combining the fusion and discrimination of multiple sensors. It overcomes the defects of existing positioning devices, such as inaccurate data from a single sensor and insufficient classification basis, and can greatly reduce the problem of inaccurate classification.

[0161] The following describes in detail the method provided by this invention for determining the type of chicken coop using information gain as multi-type feature data. Its specific implementation steps mainly include:

[0162] (1) Based on the contents recorded in the feature data table, calculate the confusion matrix table related to each feature data to determine the probability P of the feature data affecting a certain category. i .

[0163] (2) P i Substitute into formula (1) to obtain the information entropy (s) of each feature data.

[0164] (3) Substitute Entropy(s) into formula (2) to obtain the information gain (i) of each feature data.

[0165] (4) Substitute Information Gain(i) into formula (3) to obtain the weight W of each feature data. i .

[0166] (5) Based on the determined weights W of each feature data iBy combining the characteristic data of each chicken cage collected, the confidence level of each chicken cage type can be calculated as low-laying chicken cage, normal chicken cage, and dead chicken cage.

[0167] (6) Finally, the type with a confidence level greater than a certain preset threshold is determined as the actual type of chicken cage.

[0168] For example, the feature data table mentioned above can be displayed as shown in Table 1:

[0169] Table 1 Feature Data Table

[0170]

[0171] Taking the rooster comb color information as an example, to calculate the information gain of this feature data, based on the feature data table shown in Table 1, the corresponding confusion matrix table is shown in Table 2:

[0172] Table 2. Confusion matrix corresponding to rooster comb color information.

[0173]

[0174] Furthermore, before calculating the information gain corresponding to the rooster comb color information, it is necessary to first calculate the information entropy corresponding to the rooster comb color information. The expression of the corresponding calculation formula (1) is as follows:

[0175]

[0176] Where c is the number of categories; p i Let this feature data affect the probability of the i-th category.

[0177] After calculating the information entropy related to the rooster comb color information, the information gain corresponding to the rooster comb color information can be obtained through the information gain formula. The expression of the corresponding calculation formula (2) is as follows:

[0178]

[0179] After obtaining the information gain corresponding to the rooster comb color information, it is first normalized to obtain its corresponding weight. The expression of its corresponding calculation formula (3) is as follows:

[0180]

[0181] The abnormal chicken cage identification method provided by this invention is based on multi-sensor coordination and multi-source data fusion. It extracts internal and external features of individual chickens within the cage and feeds them into a decision tree evaluation model for classification. This eliminates the limitations imposed by inaccurate single-feature data, making the data more reliable and the judgment more accurate. Furthermore, it allows for flexible customization of classification standards according to user needs, providing more flexible service to users.

[0182] Based on the above embodiments, as an optional embodiment, a QR code of the same shape is set in the upper left corner of each chicken cage in the chicken house, and each QR code is used to store the location information of the chicken cage; the step of obtaining the sampling image of the target chicken cage includes:

[0183] Acquire any image of a chicken coop captured by the sampling camera;

[0184] The coordinate information of two adjacent QR codes in the chicken coop image is obtained to determine the coordinate information of the target chicken cage in the chicken coop image; the target chicken cage is the chicken cage included in the area defined by the same side of the two adjacent QR codes in the chicken coop image.

[0185] The sampled image of the target chicken cage is cropped from the chicken coop image based on the coordinate information of the chicken cage.

[0186] It should be emphasized that the QR code placed in the upper left corner of each chicken cage serves two main purposes:

[0187] First, each QR code stores the location information of its corresponding chicken cage. When the robot inspects a cage and determines its type, it simultaneously reads the QR code in the upper left corner to obtain the cage's location information. This location information, combined with its type, is then stored in a database for remote user terminals to access and use, or displayed directly on a virtual chicken coop map. This allows a single virtual chicken coop map to simultaneously display the types of all chicken cages within the entire coop, providing services for user management of the chicken coop.

[0188] Secondly, since the images of chicken coops captured by the sampling camera generally include not just one chicken cage, but multiple chicken cages at the same time, if it is necessary to analyze the type of each chicken cage, it is necessary to crop the images of chicken coops captured by the sampling camera in order to separate the images corresponding to each chicken cage.

[0189] In view of this, the abnormal chicken coop identification method provided by the present invention involves pre-attaching a QR code (each QR code is the same size) to the upper left corner (or the upper right, lower left, or lower right corner) of each chicken coop. When cropping the chicken coop image acquired by the sampling camera, the method counts when two adjacent QR codes are detected and simultaneously identifies the coordinate information of these two QR codes in the chicken coop image. Then, a coordinate transformation is used to calculate the coordinate information of the left and right boundaries of the chicken coop in the chicken coop image. It should be noted that in the present invention, the upper and lower boundaries of each chicken coop are assumed to be fixed. Only the left and right boundaries of each chicken coop need to be determined to define the area where the chicken coop is located in the chicken coop image. Cropping this area yields the sampling image of the chicken coop.

[0190] This invention uses pre-placed QR codes to divide the sampled images corresponding to each chicken cage using the coordinate information of the QR code boundaries, which can effectively reduce distortion caused by camera movement.

[0191] Based on the above embodiments, as an optional embodiment, obtaining the coordinate information of two adjacent QR codes in the chicken coop image to determine the coordinate information of the target chicken cage in the chicken coop image includes:

[0192] Based on the coordinate information of each QR code in the chicken coop image, combined with the optical center coordinates of the target chicken cage from the sampling camera, the intrinsic parameters of the sampling camera, and the focusing distance of the target chicken cage, the world coordinate information of each QR code is determined;

[0193] Based on the world coordinate information of the two QR codes, the coordinate information of the target chicken coop is determined.

[0194] As an alternative embodiment, on the one hand, when identifying chicken heads and eggs in the sampled images, the robot's movement can cause errors in the delineation of the boundaries of each chicken cage, making it easy to make mistakes in the division.

[0195] The abnormal chicken coop identification method provided by this invention uses a monocular camera calibration model to identify the image coordinates (img) of the QR code. x img y Perform coordinate transformation to obtain its corresponding world coordinates (w) x w y w z Then, by calculating the object distance from each chicken coop to the sampling camera lens, the image coordinates of each chicken coop can be obtained, which can then correct the boundary division of the chicken coops. The specific coordinate transformation formula is as follows:

[0196]

[0197] Formula (4) above can be further converted to:

[0198]

[0199] Among them, img x ,img y c represents the position of the left vertex of the QR code in the image coordinate system. x c y f represents the optical center coordinates of the sampling camera; x f y z1 is the intrinsic parameter of the sampling camera; z1 is the focusing distance of the QR code; w x w y z1 is the coordinates of the left vertex of the QR code in the world coordinate system; z2 is the focusing distance of the chicken coop.

[0200] The abnormal chicken cage identification method provided by the present invention uses a monocular camera calibration model to correct the distortion of the chicken cage during the sampling camera's motion and shooting process, thereby preventing errors in counting laying hens and eggs in the cage.

[0201] Based on the above embodiments, as an optional embodiment, after determining the category of the target chicken cage, the method further includes:

[0202] Locate the virtual area of ​​the target chicken coop in the virtual chicken coop map, and mark the category of the target chicken coop within the virtual area.

[0203] Figure 4 This is a schematic diagram of a virtual chicken coop provided by the present invention, such as... Figure 4 As shown, to further reduce stress in chickens caused by excessive human intervention, the robot can automatically inspect along a set route without human intervention. Wireless communication can be established between the user terminal and the robot's main control unit, such as through wireless sensing technology using Zigbee sensors. This virtual chicken coop... Figure 1 It can be displayed on user terminals deployed outside the chicken house, allowing users to remotely monitor and manage the status of each chicken cage inside the chicken house.

[0204] Users can view the category information of each chicken cage and the distribution location of each chicken cage in the virtual chicken house map on the user terminal.

[0205] In the virtual chicken coop Figure 1 Each chicken cage is pre-assigned a number, such as 1001-1024, which are used as the numbers for the 24 chicken cages. The number of each chicken cage is stored in an RFID tag 3 installed on each chicken cage. When the robot 5 inspects each chicken cage, it can obtain the number of each chicken cage by reading the RFID tag 3 using the onboard RFID reader.

[0206] Furthermore, each chicken cage is also equipped with a sound sensor 4, which can collect the sound information of the chickens inside the cage in real time.

[0207] It should be noted that the sound sensor 4 is connected to the RFID tag 3 to transmit the collected sound information to the RFID tag 3 in real time. In this way, after the robot 5 reads the RFID tag 3, it can simultaneously obtain the chicken coop's identification number and the sound information.

[0208] A QR code 2 is set in the upper left corner of each chicken cage. The function of QR code 2 has been described in the above embodiment and will not be repeated here.

[0209] Furthermore, in order to reduce manual operation, magnetic strip tracks 6 are pre-laid on the passageways inside the chicken coop for the robot 5. The robot 5's travel route can be set in advance according to the layout of the magnetic strip tracks 6, and instructions can be issued to the robot 5 according to specific markers to control its forward, backward, left, and right turns.

[0210] This invention places the user terminal outside the chicken farm, effectively reducing the number of times people enter and exit the farm, thus reducing stress on the chickens. The placement of the magnetic strip track can be manually adjusted, making it suitable for intelligent positioning services in various chicken farms. Furthermore, a virtual chicken coop location map is created proportionally based on the actual chicken coop layout, displaying the distribution of all chicken cages within the entire coop to improve positioning accuracy. Once the type of each chicken cage is determined, its location can be identified by reading its RFID tag, significantly improving positioning accuracy.

[0211] Figure 5 This is a schematic diagram of the abnormal chicken cage identification device provided by the present invention, as shown below. Figure 5 As shown, it mainly includes an image acquisition unit 51, an image detection unit 52, and a chicken cage classification unit 53, wherein:

[0212] The image acquisition unit 51 is mainly used to acquire sampled images of the target chicken cage;

[0213] The image detection unit 52 is mainly used to identify the sampled image to obtain a first identification result and a second identification result respectively; the first identification result includes the number of chicken heads, the number of eggs, and the upright state of the combs of each chicken in the target chicken cage; the second identification result includes the comb color information of each chicken.

[0214] The chicken cage classification unit 53 is mainly used to input the first identification result and the second identification result into the decision tree model to determine the category of the target chicken cage. The categories include low-laying hen cages, normal hen cages, and dead hen cages.

[0215] It should be noted that the abnormal chicken cage identification device provided in this embodiment of the invention can execute the abnormal chicken cage identification method described in any of the above embodiments during specific operation, which will not be elaborated in this embodiment.

[0216] The abnormal chicken cage identification device and system provided by this invention are based on the coordinated operation of multiple sensors and the fusion of multiple source data. After comprehensively extracting the internal and external features of individual chickens in the cage, the data is fed into a decision tree evaluation model for classification, resulting in more reliable data and more accurate classification results.

[0217] Figure 6 This is a structural schematic diagram of the robot provided by the present invention, as shown below. Figure 6 As shown, the robot provided by the present invention includes a main control unit, which implements the abnormal chicken cage identification method as described in any of the above embodiments during operation.

[0218] The robot is equipped with a sampling camera 10, a supplementary lighting device, an RFID reader / writer 8, drive wheels 7, a temperature sensor 9, and rotating wheels 11. The robot can move along a pre-set inspection route, intelligently locate itself based on information captured by the sampling camera, and ultimately feed the location information back to the user terminal.

[0219] The power wheel 7 serves as the robot's power driver and gravity support point. The supplementary lighting device is used when the environment inside the chicken coop is too dim, affecting the image acquisition of the sampling camera. It provides supplementary lighting without disturbing the chickens. The rotating wheel 11 can automatically adjust the distance between the sampling camera lens and each chicken cage according to instructions.

[0220] The robot provided by this invention can be used in conjunction with deep learning and target detection technology to determine whether there are dead chickens in the cage based on the number of chickens in the cage and the surface temperature of the chickens in the cage; and to determine whether there are low-producing chickens in each cage by using feature data such as the number of eggs in the cage, the upright state of the chicken comb, the color information of the chicken comb, the temperature information of the chickens and the sound information.

[0221] Figure 7 This is a schematic diagram of the abnormal chicken cage identification system provided by the present invention, as shown below. Figure 7 As shown, it mainly includes a robot 5; a sampling camera 10 and an RFID reader 8 are installed on the robot 5, and a magnetic strip track 6 is laid in the chicken house according to a set route; the robot 5 can inspect the chicken house according to the set route.

[0222] It also includes a user terminal, as well as RFID tags 3, QR codes 2, sound sensors 4 and temperature sensors 9 installed on each chicken cage;

[0223] The sound sensor 4 is used to collect the sound information of each chicken in each cage, and the temperature sensor 9 is used to collect the temperature information of each chicken in each cage. Each RFID tag 3 stores the number information of the chicken cage it belongs to, as well as the sound information and temperature information collected at each sampling time uploaded by the sound sensor and the temperature sensor.

[0224] The robot is connected to the user terminal to transmit the determined category of the target chicken cage and the identity information obtained by the RFID reader to the user terminal in real time.

[0225] The present invention also provides an abnormal chicken cage identification system, which mainly comprises a user terminal, a main control unit, a detection unit, a segmentation unit, a classification unit, and a positioning unit.

[0226] The user terminal exists in the form of an APP on the host computer. Users can check the location information of different types of chicken cages at any time through the human-computer interaction interface of the user terminal.

[0227] The main control unit is a data processing center that integrates data reception and transmission;

[0228] The detection unit mainly includes three modules. The first detection module detects the number of chicken heads and eggs and collects egg information. The second module detects whether the chicken comb is upright and sends the feature data obtained from the above two modules to the main control unit. The third module acquires relevant information such as RFID tags, temperature sensors, and sound sensors, processes it, and transmits it to the main control unit.

[0229] The segmentation unit mainly performs segmentation and cropping of the rooster comb image, extracts the RGB information of the rooster comb from the segmented rooster comb image, and sends the rooster comb color information to the main control unit;

[0230] The classification unit mainly integrates the feature data obtained from the detection unit and the segmentation unit, and uses these feature data to classify each chicken cage into different categories, and sends the obtained category results information to the main control unit.

[0231] The positioning unit receives the classification results from the main control unit, intelligently locates the positions of dead chickens and low-laying hens on the chicken house location distribution map, and sends the information to the main control unit.

[0232] Furthermore, users can view the category information of chicken cages and the location of each chicken cage in the virtual chicken house location distribution map on the user terminal, and store the category results, chicken cage QR codes, chicken cage RFID identification information, date and other information in the relevant database of the user terminal for later query and comparison.

[0233] To better demonstrate the technical content, structural features and effects of the present invention, the working process of the abnormal chicken cage identification system is described clearly and completely below.

[0234] The entire system's detection and classification process is as follows: Operating hours are 11:00 AM daily. In the initial system state, the robot activates its sampling camera to prepare for scanning. As the robot patrols along its set route, the sampling camera detects scanned targets such as chicken cage QR codes, chickens, eggs, and chicken combs. Simultaneously, a supplementary lighting device is activated to collect chicken comb color information. An RFID reader identifies and reads the chicken cage number, a temperature sensor records the temperature of the chickens inside the cage, and a sound sensor identifies the chickens' sounds. During data acquisition, the detection unit detects the chicken cage QR code, cage number, number of chickens, number of eggs, chicken comb upright position, temperature information, and sound information, and transmits the processed feature data to the main control unit in real time.

[0235] The segmentation unit performs real-time segmentation on the rooster comb image and sends the obtained rooster comb color information to the main control unit.

[0236] The main control unit performs multi-feature fusion on the features received by the detection unit and the segmentation unit, and sends the fused features to the classification unit. The classification unit divides the chicken cages into three categories based on the multiple features: normal chicken cages, low-producing chicken cages, and dead chicken cages.

[0237] After classifying each chicken cage, the type results are returned to the main control unit, which then sends the type results to the positioning unit.

[0238] The positioning unit intelligently locates each chicken cage in the virtual chicken coop map based on the type result, and transmits the chicken cage type and location information to the main control unit. The main control unit then uploads the final result to the user terminal system via Zigbee sensors for permanent storage.

[0239] Thus, after completing one round of inspection, the robot returns to its initial state and can move to the next chicken coop to continue inspection and begin the next round of work.

[0240] The abnormal chicken cage identification system provided by this invention uses Zigbee for wireless communication between the user terminal and the robot. After reading the RFID tag information, the system can retrieve the category information and location information of dead chickens and low-laying hens in each chicken cage from the terminal database based on the cage's number. Simultaneously, the category information is saved to the terminal system in real time for permanent storage, for future retrieval and use.

[0241] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 8As shown, the electronic device may include a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute an abnormal chicken cage identification method, which includes: acquiring a sampled image of the target chicken cage; identifying the sampled image to obtain a first identification result and a second identification result; the first identification result includes the number of chicken heads, the number of eggs, and the upright state of the combs of each chicken in the target chicken cage; the second identification result includes the comb color information of each chicken; inputting the first identification result and the second identification result into a decision tree model to determine the category of the target chicken cage; the category includes low-laying chicken cages, normal chicken cages, and dead chicken cages.

[0242] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0243] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein when the program instructions are executed by a computer, the computer is able to execute the abnormal chicken cage identification method provided in the above embodiments, the method comprising: acquiring a sampled image of a target chicken cage; identifying the sampled image to acquire a first identification result and a second identification result respectively; the first identification result comprising the number of chicken heads, the number of eggs, and the upright state of the combs of each chicken in the target chicken cage; the second identification result comprising the comb color information of each chicken; inputting the first identification result and the second identification result into a decision tree model to determine the category of the target chicken cage; the category comprising low-laying chicken cages, normal chicken cages, and dead chicken cages.

[0244] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method for identifying abnormal chicken cages provided in the above embodiments. The method includes: acquiring a sampled image of a target chicken cage; identifying the sampled image to obtain a first identification result and a second identification result respectively; the first identification result includes the number of chicken heads, the number of eggs, and the upright state of the combs of each chicken in the target chicken cage; the second identification result includes the comb color information of each chicken; inputting the first identification result and the second identification result into a decision tree model to determine the category of the target chicken cage; the category includes low-laying chicken cages, normal chicken cages, and dead chicken cages.

[0245] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0246] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0247] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying abnormal chicken cages, characterized in that, include: Acquire a sampled image of the target chicken coop; The sampled image is identified to obtain a first identification result and a second identification result respectively; the first identification result includes the number of chicken heads, the number of eggs, and the upright state of the combs of each chicken in the target chicken cage; the second identification result includes the comb color information of each chicken. The first and second identification results are input into the decision tree model to determine the category of the target chicken cage; The categories include low-laying hen cages, normal hen cages, and dead hen cages; Acquire the sound information of each chicken in the target chicken cage; Accordingly, the sound information, the first recognition result, and the second recognition result are input into the decision tree model to determine the category of the target chicken cage; The step of acquiring the sound information of each chicken in the target chicken cage includes: Collect the sound signals of each chicken in the target chicken cage within a preset time period; The characteristic parameters related to the sound signal are obtained. These characteristic parameters are linear predictive cepstral coefficients extracted after the start and end points of the sound signal are determined by analyzing the zero-crossing rate and energy of the sound signal. The feature parameters are input into a pre-trained sound classification model to obtain the sound information output by the sound classification model; The sound classification model is obtained by training a Hidden Markov Model as the initial model.

2. The method for identifying abnormal chicken cages according to claim 1, characterized in that, Identifying the sampled image and obtaining the first identification result includes: The sampled image is input into a pre-trained chicken head detection network model to obtain multiple chicken head sub-images in the sampled image, and the number of chicken head sub-images is taken as the number of chicken heads in the target chicken cage; The sampled image is input into a pre-trained egg detection network model to obtain multiple egg sub-images in the sampled image, and the number of the egg sub-images is taken as the number of eggs in the target chicken coop; The sampled image is input into a pre-trained rooster comb detection network model to obtain multiple rooster comb sub-images in the sampled image, and the upright state of the rooster comb in each rooster comb sub-image is determined.

3. The method for identifying abnormal chicken cages according to claim 1, characterized in that, Identifying the sampled image and obtaining the second identification result specifically includes: The sampled image is input into a pre-trained rooster comb segmentation network model to obtain at least one rooster comb sub-image segmented from the sampled image by the output of the rooster comb segmentation network model; Based on the distribution of RGB values ​​of each chicken comb image, the comb color information of each chicken is determined.

4. The method for identifying abnormal chicken cages according to claim 1, characterized in that, Also includes: Obtain the body temperature information of each chicken in the target chicken cage; Accordingly, the sound information, the body temperature information, the first recognition result, and the second recognition result are input into the decision tree model to determine the category of the target chicken cage.

5. The method for identifying abnormal chicken cages according to claim 4, characterized in that, The step of obtaining the body temperature information of each chicken in the target chicken cage includes: The body surface temperature of each chicken was collected at multiple sampling times; Based on the least squares method, a binary linear regression was performed on the body surface temperature of each chicken at the multiple sampling times to obtain the body temperature information of each chicken.

6. The method for identifying abnormal chicken cages according to claim 4, characterized in that, The step of inputting the sound information, the body temperature information, the first recognition result, and the second recognition result into a decision tree model to determine the category of the target chicken coop includes: Based on a pre-constructed feature data table for determining the category of chicken cages, the feature data table records the mapping relationship between sound information, body temperature information, matching degree of chicken head number, matching degree of egg number, chicken comb upright status and chicken comb color information and the category of chicken cages. Using the feature data table, a confusion matrix table is generated based on the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state, and chicken comb color information related to the target chicken cage, so as to calculate the probability that the target chicken cage belongs to any category based on the confusion matrix table; Based on the probability that the target chicken cage belongs to any category, the information entropy corresponding to the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state and chicken comb color information are determined respectively. Based on the information entropy corresponding to the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state, and chicken comb color information, the information gain corresponding to the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state, and chicken comb color information is determined respectively. The information gains corresponding to the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state, and chicken comb color information are normalized respectively to determine the weights corresponding to the sound information, body temperature information, chicken head number matching degree, egg number matching degree, chicken comb upright state, and chicken comb color information. Based on the matching degree of the number of chicken heads and its corresponding weight, and the body temperature information and its corresponding weight, calculate the confidence level that the target chicken cage is a dead chicken cage; Based on the matching degree of the number of eggs and its corresponding weight, the upright state of the rooster comb and its corresponding weight, the rooster comb color information and its corresponding weight, the body temperature information and its corresponding weight, and the sound information and its corresponding weight, the confidence level of classifying the target chicken cage as a low-laying chicken cage is calculated. The category of the target chicken cage is determined based on the confidence level that the target chicken cage is a dead chicken cage and the confidence level that the target chicken cage is a low-laying chicken cage. The chicken head count matching degree is determined by comparing the number of chicken heads identified in the target chicken cage with the standard number of chicken heads in the target chicken cage. The egg count matching degree is determined by comparing the number of eggs identified in the target chicken cage with the standard number of eggs in the target chicken cage.

7. The method for identifying abnormal chicken cages according to claim 1, characterized in that, Each chicken cage in the chicken house has a QR code of the same shape in the upper left corner. Each QR code is used to store the location information of the chicken cage. The acquisition of the sampled image of the target chicken cage includes: Acquire any image of a chicken coop captured by the sampling camera; The coordinate information of two adjacent QR codes in the chicken coop image is obtained to determine the coordinate information of the target chicken cage in the chicken coop image; the target chicken cage is the chicken cage included in the area defined by the same side of the two adjacent QR codes in the chicken coop image. The sampled image of the target chicken cage is cropped from the chicken coop image based on the coordinate information of the chicken cage.

8. The method for identifying abnormal chicken cages according to claim 7, characterized in that, Obtaining the coordinate information of two adjacent QR codes in the chicken coop image to determine the coordinate information of the target chicken cage in the chicken coop image includes: Based on the coordinate information of each QR code in the chicken coop image, combined with the optical center coordinates of the target chicken cage from the sampling camera, the intrinsic parameters of the sampling camera, and the focusing distance of the target chicken cage, the world coordinate information of each QR code is determined; Based on the world coordinate information of the two QR codes, the coordinate information of the target chicken coop is determined.

9. The method for identifying abnormal chicken cages according to claim 7, characterized in that, After determining the category of the target chicken cage, the process also includes: Locate the virtual area of ​​the target chicken coop in the virtual chicken coop map, and mark the category of the target chicken coop within the virtual area.

10. A device for identifying abnormal chicken cages, characterized in that, include: Image acquisition unit, used to acquire sampled images of the target chicken cage; An image detection unit is used to identify the sampled image to obtain a first identification result and a second identification result respectively; the first identification result includes the number of chicken heads, the number of eggs, and the upright state of the combs of each chicken in the target chicken cage; the second identification result includes the comb color information of each chicken. The chicken cage classification unit is used to input the first identification result and the second identification result into the decision tree model to determine the category of the target chicken cage; The categories include low-laying hen cages, normal hen cages, and dead hen cages; Acquire the sound information of each chicken in the target chicken cage; Accordingly, the sound information, the first recognition result, and the second recognition result are input into the decision tree model to determine the category of the target chicken cage; The step of acquiring the sound information of each chicken in the target chicken cage includes: Collect the sound signals of each chicken in the target chicken cage within a preset time period; The characteristic parameters related to the sound signal are obtained. These characteristic parameters are linear predictive cepstral coefficients extracted after the start and end points of the sound signal are determined by analyzing the zero-crossing rate and energy of the sound signal. The feature parameters are input into a pre-trained sound classification model to obtain the sound information output by the sound classification model; The sound classification model is obtained by training a Hidden Markov Model as the initial model.

11. A system for identifying abnormal chicken cages, characterized in that, The method for identifying abnormal chicken cages as described in any one of claims 1-9 is executed. It also includes user terminals, as well as identification tags, QR codes, sound sensors, and temperature sensors installed on each chicken cage; The sound sensor is used to collect the sound information of the chickens in each cage, and the temperature sensor is used to collect the temperature information of the chickens in each cage. Each identification tag stores the number information of the chicken cage it belongs to, as well as the sound information and temperature information collected at each sampling time by the sound sensor and the temperature sensor.

Images