Chicken behavior statistical method, system and device and storage medium

Abstract

The invention provides a chicken behavior statistical method, system and device and a storage medium, and relates to the technical field of chicken breeding, and the method comprises the steps: obtaining to-be-analyzed image data; the to-be-analyzed image data are input into a preset target recognition model, chicken detection data are obtained, and the chicken detection data reflect chickens detected in the images; inputting the chicken detection data into a preset behavior recognition model to obtain chicken behavior data; according to the chicken behavior data, behavior statistical data is generated, and the behavior statistical data reflects the occurrence probability of each behavior type. Chicken recognition and behavior recognition are carried out on the image, manual observation is not needed, the behavior type of the chicken can be automatically obtained, the judgment standard unification of the behavior type is facilitated, the data error is reduced, the data stability is improved, statistics is carried out according to the behavior type of the chicken, the occurrence probability of each behavior type is obtained, and the efficiency is improved. And the comfort of chicken breeding can be evaluated more accurately and reliably.

Description

Technical Field

[0001] The present invention relates to the technical field of chicken breeding, and in particular to a chicken behavior statistics method, system, equipment and storage medium. Background Art

[0002] Modern chicken farming is increasingly moving towards large-scale production, using efficient farming techniques and management methods to increase production and reduce costs to meet people's demand for food. In the chicken farming process, it is necessary to monitor the behavior and health status of the chickens. Through manual observation, farmers can promptly find out whether the chickens are eating, drinking, active, and whether they have abnormal behavior. According to the chicken's behavior, the comfort of the chicken in the breeding cage can be known, and creating a better comfort level is conducive to the healthy growth and development of the chicken. These observation records are of great reference value for breeding management and chicken disease prevention.

[0003] However, as the scale of farming increases, the number of chickens also increases accordingly. Manual observation becomes cumbersome and time-consuming, and it is difficult to monitor the behavior of each chicken in a timely and comprehensive manner. Manual observation is affected by subjective factors of personnel, which may cause problems such as inconsistent data standards, large data errors, and poor data stability. As a result, the comfort assessment of chicken farming based on manual observation data of chicken behavior is unreliable. Summary of the invention

[0004] The present invention provides a chicken behavior statistics method, system, device and storage medium, which are used to solve the problems of subjective influence, inconsistent data standards, large data errors and poor data stability in manual observation of chickens in the prior art.

[0005] The present invention provides a chicken behavior statistics method, comprising:

[0006] Acquire image data to be analyzed;

[0007] Inputting the image data to be analyzed into a preset target recognition model to obtain chicken detection data, wherein the chicken detection data reflects the chickens detected in the image;

[0008] Inputting the chicken detection data into a preset behavior recognition model to obtain chicken behavior data, wherein the chicken behavior data reflects the behavior type of each chicken;

[0009] Behavior statistics are generated based on the chicken behavior data, and the behavior statistics reflect the occurrence probability of each behavior type.

[0010] According to a chicken behavior statistics method provided by the present invention, the step of obtaining image data to be analyzed includes:

[0011] Acquire captured image data, wherein the captured image data includes a plurality of image frames;

[0012] According to a preset frame period, image frames are selected from the captured image data at intervals of the frame period and saved to form image data to be analyzed.

[0013] According to a chicken behavior statistics method provided by the present invention, before obtaining the image data to be analyzed, the method further includes:

[0014] Acquire an original image dataset, wherein the original image dataset includes a plurality of chicken images;

[0015] Classifying the original image data set according to the growth cycle of chickens to obtain at least two groups of classified image data sets;

[0016] Performing an amplification process on each group of the classified image data sets to increase the number of images to form a corresponding training image data set;

[0017] According to each group of the training image data sets, a preset first neural network model is trained to obtain the target recognition model;

[0018] According to each group of the training image data sets, the preset second neural network model is trained to obtain the behavior recognition model.

[0019] According to a chicken behavior statistics method provided by the present invention, the preset first neural network model is trained according to each group of the training image data sets to obtain the target recognition model, including:

[0020] According to each group of the training image data sets, the chickens in each image are marked with identification frames to form a first marked image data set;

[0021] According to a preset first ratio, each group of the first annotated image data sets is divided into a first training image set and a first test image set;

[0022] Inputting the first training image set into a preset first neural network model for training;

[0023] Inputting the first test image set into the trained first neural network model, and judging whether the training is completed according to the accuracy rate of the target recognition result output by the first neural network model;

[0024] If the training is completed, the trained first neural network model is used as the target recognition model.

[0025] According to a chicken behavior statistics method provided by the present invention, the preset second neural network model is trained according to each group of the training image data sets to obtain the behavior recognition model, including:

[0026] According to each group of the first annotated image data sets, adding a behavior label to each of the identification boxes to form a second annotated image data set;

[0027] According to a preset second ratio, each group of the second annotated image data set is divided into a second training image set and a second test image set;

[0028] Inputting the second training image set into a preset second neural network model for training;

[0029] Inputting the second test image set into the trained second neural network model, and judging whether the training is completed according to the behavior type recognition accuracy output by the second neural network model;

[0030] If the training is completed, the trained second neural network model is used as the behavior recognition model.

[0031] According to a chicken behavior statistics method provided by the present invention, the amplification process is performed on each group of the classified image data sets to increase the number of images to form a corresponding training image data set, including:

[0032] Performing direction conversion processing on the classified image data set to form a direction conversion image data set;

[0033] and / or, performing color conversion processing on the classified image data set to form a color converted image data set;

[0034] and / or, performing size conversion processing on the classified image data set to form a size-converted image data set;

[0035] and / or, performing an interpolation algorithm on the classified image data set to form an interpolated image data set;

[0036] and / or, performing noise addition processing on the classified image dataset to form a noise-enhanced image dataset;

[0037] A training image dataset is formed according to the classification image dataset and the direction conversion image dataset and / or the color conversion image dataset and / or the size conversion image dataset and / or the interpolation image dataset and / or the noise enhancement image dataset.

[0038] According to a chicken behavior statistics method provided by the present invention, the behavior type recognition results output by the behavior recognition model include feeding behavior, drinking behavior, walking behavior, lying behavior, fighting behavior, gathering behavior and rioting behavior; generating behavior statistics data based on the chicken behavior data includes:

[0039] According to the chicken behavior data, respectively calculate a first probability corresponding to the feeding behavior, a second probability corresponding to the drinking behavior, a third probability corresponding to the walking behavior, a fourth probability corresponding to the lying behavior, a fifth probability corresponding to the fighting behavior, a sixth probability corresponding to the gathering behavior, and a seventh probability corresponding to the rioting behavior;

[0040] Classify the eating behavior, the drinking behavior, the walking behavior, and the lying behavior as normal behaviors, and classify the fighting behavior, the gathering behavior, and the rioting behavior as abnormal behaviors, and calculate the eighth probability corresponding to the normal behavior and the ninth probability corresponding to the abnormal behavior;

[0041] Behavior statistics are generated according to the first probability, the second probability, the third probability, the fourth probability, the fifth probability, the sixth probability, the seventh probability, the eighth probability, and the ninth probability.

[0042] The present invention also provides a chicken behavior statistics system, comprising: a processing device and at least two camera devices, at least one of the camera devices is arranged on the top of the breeding cage, and at least one of the camera devices is arranged on the side wall of the breeding cage. The processing device is communicatively connected to the camera device, and the processing device can execute the above-mentioned chicken behavior statistics method.

[0043] The present invention also provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the chicken behavior statistics method as described above when executing the program.

[0044] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the chicken behavior statistics method as described above is implemented.

[0045] The chicken behavior statistics method provided by the present invention has at least the following beneficial effects: according to the image data to be analyzed, the chickens in the image are identified by the target recognition model, and the chicken detection data is generated. According to the chicken detection data, the position and size of the chickens in the image can be known. According to the behavior recognition model, the chicken detection data is further processed to identify the behavior of the chickens and generate chicken behavior data. According to the chicken behavior data, the behavior type corresponding to each chicken in the image can be known, and then the number of chickens of each behavior type in the image is determined. All the images to be analyzed are counted, and the number of occurrences of each behavior type in all the images to be analyzed can be obtained, and then the probability of occurrence of each behavior type is calculated to generate behavior statistics. In this way, by performing chicken identification and behavior identification on the image, the behavior type of the chicken can be automatically known without relying on manual observation, which is conducive to the unification of the judgment standard of the behavior type, reducing data errors and improving the stability of the data, and according to the chicken behavior type, the occurrence probability of each behavior type is known, which is conducive to more accurate and reliable evaluation of the comfort of chicken breeding. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 This is one of the flow charts of a chicken behavior statistics method provided by the present invention;

[0048] Figure 2 The present invention provides a second flow chart of a chicken behavior statistics method;

[0049] Figure 3 The present invention provides a method for statistically analyzing chicken behaviors;

[0050] Figure 4 The present invention provides a fourth flowchart of a chicken behavior statistics method;

[0051] Figure 5 The fifth flow chart of the chicken behavior statistics method provided by the present invention;

[0052] Figure 6 This is a sixth flow chart of a chicken behavior statistics method provided by the present invention;

[0053] Figure 7 The present invention provides a schematic structural diagram of a chicken behavior statistics system in which a camera device is installed on the top of a breeding cage;

[0054] Figure 8 The present invention provides a schematic structural diagram of a chicken behavior statistics system in which a camera device is installed on the side wall of a breeding cage;

[0055] Fig. 9 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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.

[0057] Combine the following Figure 1-Figure 6 A chicken behavior statistical method of the present invention is described, comprising:

[0058] S100: Acquire image data to be analyzed;

[0059] S110: inputting the image data to be analyzed into a preset target recognition model to obtain chicken detection data, where the chicken detection data reflects the chickens detected in the image;

[0060] S120: Inputting the chicken detection data into a preset behavior recognition model to obtain chicken behavior data, wherein the chicken behavior data reflects the behavior type of each chicken;

[0061] S130: Generate behavior statistics data based on the chicken behavior data, wherein the behavior statistics data reflects the occurrence probability of each behavior type.

[0062] According to the image data to be analyzed, the target recognition model is used to identify the chickens in the image, and the chicken detection data is generated. According to the chicken detection data, the position and size of the chickens in the image can be known. The chicken detection data is further processed by the behavior recognition model to identify the chicken's behavior and generate chicken behavior data. According to the chicken behavior data, the behavior type corresponding to each chicken in the image can be known, and then the number of chickens of each behavior type in the image can be determined. All the images to be analyzed are counted, and the number of each behavior type in all the images to be analyzed can be obtained, and then the probability of each behavior type occurring can be calculated to generate behavior statistics. In this way, by performing chicken recognition and behavior recognition on the image, the behavior type of the chicken can be automatically known without relying on manual observation, which is conducive to the unification of the judgment standard of the behavior type, reducing data errors and improving the stability of the data. According to the chicken behavior type, statistics are performed to know the probability of occurrence of each behavior type, which is conducive to more accurate and reliable evaluation of the comfort of chicken farming.

[0063] The behavior recognition of chickens is divided into two steps. First, the target recognition model is used to identify the position and size of the chickens in the image to be analyzed, and then the behavior recognition model is used to identify the behavior type of the identified chickens. In this way, the image of the chicken flock can be directly recognized to obtain the behavior type of each chicken. The function division of the target recognition model and the behavior recognition model is clear, which is convenient for simplifying the complexity of a single model and improving the accuracy of recognition.

[0064] During the chicken breeding process, if the environment of the breeding cage is suitable, that is, the comfort level is high, the probability of abnormal behavior of the chicken is low. On the contrary, if the comfort level is low, the probability of abnormal behavior will increase significantly. Therefore, by statistically analyzing the probability of various types of chicken behaviors, it is possible to provide a basis for comfort evaluation, and take measures to improve the environment of the breeding cage when the probability of abnormal behavior is high, such as cooling down when the temperature is high, which is conducive to monitoring the environment of the breeding cage, improving the comfort level of the chickens, and allowing the chickens to grow healthily.

[0065] refer to Figure 2 In some embodiments of the chicken behavior statistics method of the present invention, the S100 includes:

[0066] S101: Acquire captured image data, where the captured image data includes a plurality of image frames;

[0067] S102: According to a preset frame period, image frames are selected from the captured image data at intervals of the frame period and saved to form image data to be analyzed.

[0068] Since the number of frames captured by the shooting device is relatively high, and the behavior of the chickens does not change too quickly, the behavior of the chickens does not change much between several consecutive image frames. Therefore, according to the preset frame period, one frame of the continuous image frames in the captured image data is selected and saved every frame period to remove a part of the image frames, and the saved image frames form the image data to be analyzed. The image data to be analyzed is processed and redundant image frames are removed, which can reduce the amount of data processing and help reduce the performance requirements of the processing equipment.

[0069] A larger frame period can reduce the processing burden, and a smaller frame period can improve the resolution of behavior recognition. By setting an appropriate frame period, the processing load and recognition resolution can be balanced. In some embodiments of the present invention, when the frame rate of the captured image data is 30 frames per second and the frame period is 10 frames, one image frame is selected for storage every 10 frames.

[0070] refer to Figure 3 In some embodiments of the chicken behavior statistics method of the present invention, before S100, it also includes:

[0071] S000: Acquire an original image dataset, where the original image dataset includes a plurality of chicken images;

[0072] S010: classifying the original image data set according to the growth cycle of the chickens to obtain at least two groups of classified image data sets;

[0073] S020: performing an amplification process on each group of the classified image data sets to increase the number of images to form a corresponding training image data set;

[0074] S030: training a preset first neural network model according to each group of the training image data sets to obtain the target recognition model;

[0075] S040: According to each group of the training image data sets, a preset second neural network model is trained to obtain the behavior recognition model.

[0076] Before the image data to be analyzed is processed for recognition, it is necessary to first train and form a target recognition model and a behavior recognition model. As chickens grow and develop, their appearance will change greatly. By classifying the original image data set according to the growth cycle of the chickens, it can be divided into three types: chicks, growing period, and degeneration period, and three groups of classified image data sets are formed accordingly. In this way, the samples of the training images are more compact and more representative according to the growth stage of the chickens in the image, which is conducive to improving the generalization ability and classification accuracy of the model. At the same time, the samples in each classified image data set are more similar, which can speed up the model training speed and save computing resources. After classification, each group of classified image data sets is amplified to increase the number of images in each group, improve the data scale, and form a training image data set, which is conducive to more comprehensive training of the model and improve the performance of the model.

[0077] According to the training image data set, the first neural network model and the second neural network model are trained respectively. It can be understood that the training objectives of the two are different. Although both input the training image data set, the first neural network model is aimed at identifying chickens in the image. Specifically, it can identify chickens based on the image contour and determine the position and size of the chickens. The second neural network model is aimed at the behavior of chickens. The training parameters of the two are different, and the output results are different. After the training is completed, models with different functions are formed, namely the target recognition model and the behavior recognition model.

[0078] The first neural network model may be a YOLO model, with Darknet53-CSP selected as the backbone network, and the output result is a chicken identification box that can reflect the position and size of the chicken. The second neural network model may be a convolutional neural network, with a global average pooling layer replacing the fully connected layer, reducing parameters to prevent overfitting, and using a spatial attention mechanism and a channel attention mechanism to improve feature extraction capabilities, wherein the spatial attention mechanism mainly focuses on different spatial positions in the input feature map, and assigns different importance weights to features at different positions, so that more attention can be paid to important areas when processing images, while ignoring unimportant areas; the channel attention mechanism mainly focuses on different channels in the input feature map, and assigns different importance weights to features of different channels, thereby focusing more attention on important channels and reducing attention to irrelevant channels.

[0079] The original image data set used to train the neural network model can be image data of breeding cages taken by a photographing device.

[0080] refer to Figure 4 In some embodiments of the chicken behavior statistics method of the present invention, the S030 includes:

[0081] S031: According to each group of the training image data sets, the chickens in each image are marked with identification frames to form a first marked image data set;

[0082] S032: Dividing each group of the first annotated image data sets into a first training image set and a first test image set according to a preset first ratio;

[0083] S033: inputting the first training image set into a preset first neural network model for training;

[0084] S034: inputting the first test image set into the trained first neural network model, and judging whether the training is completed according to the accuracy rate of the target recognition result output by the first neural network model;

[0085] S035: If the training is completed, the trained first neural network model is used as the target recognition model.

[0086] The images in the training image data set are annotated to mark the location and size of each chicken in the image with an identification box to form a first annotated image data set. Then, the first annotated image data set is divided into a first training image set and a first test image set according to a preset first ratio. The first training image set is used to train the first neural network model, and the first test image set is used to test the trained first neural network model to verify whether the training of the first neural network model is complete. When the output target recognition result of the trained first neural network model is compared with the first test image set, and the accuracy exceeds the preset threshold, the training can be considered complete, and the first neural network model is used as the target recognition model. In this way, after training with each group of training image data sets, the location and size of chickens at different growth stages in the image can be identified, thereby achieving the purpose of chicken recognition.

[0087] refer to Figure 5 In some embodiments of the chicken behavior statistics method of the present invention, the S040 comprises:

[0088] S041: adding a behavior label to each identification box according to each group of the first annotated image data sets to form a second annotated image data set;

[0089] S042: Dividing each group of the second annotated image data sets into a second training image set and a second test image set according to a preset second ratio;

[0090] S043: inputting the second training image set into a preset second neural network model for training;

[0091] S044: inputting the second test image set into the trained second neural network model, and judging whether the training is completed according to the behavior type recognition accuracy output by the second neural network model;

[0092] S045: If the training is completed, the trained second neural network model is used as the behavior recognition model.

[0093] After using an identification frame to identify the location and size of the chicken in the training image data set, a behavior label that reflects the current behavior type of the chicken is added to each identification frame to form a second annotated image data set. Then, the second annotated image data set is divided into a second training image set and a second test image set according to a preset second ratio. The second training image set is used to train the second neural network model, and the second test image set is used to verify whether the second neural network model has been trained. After the training, when the output behavior type recognition result of the second neural network model is compared with the second test image set, and the accuracy exceeds the threshold, the training can be considered complete, and the second neural network model is used as a behavior recognition model. In this way, after training with each group of training image data sets, the current behavior type of chickens at different growth stages in the image can be identified, thereby achieving the purpose of chicken behavior type recognition.

[0094] The preset first ratio and second ratio may be 8:2, that is, the ratio of the number of images included in the training image set to the number of images included in the test image set is 8:2. During the training process, if it is determined that the training is not completed, the training is repeated.

[0095] In some embodiments of the chicken behavior statistics method of the present invention, the S020 includes:

[0096] Performing direction conversion processing on the classified image data set to form a direction conversion image data set;

[0097] and / or, performing color conversion processing on the classified image data set to form a color converted image data set;

[0098] and / or, performing size conversion processing on the classified image data set to form a size-converted image data set;

[0099] and / or, performing an interpolation algorithm on the classified image data set to form an interpolated image data set;

[0100] and / or, performing noise adding processing on the classified image data set to form a noise-enhanced image data set;

[0101] A training image dataset is formed according to the classification image dataset and the direction conversion image dataset and / or the color conversion image dataset and / or the size conversion image dataset and / or the interpolation image dataset and / or the noise enhancement image dataset.

[0102] The images in the classification image data set are processed by direction conversion, color conversion, size conversion, interpolation algorithm, and noise addition, which can form new images accordingly to increase the amount of training image data and form a training image data set, which is conducive to improving the performance of the model after training. Among them, the direction conversion can be left-right flipping, angle rotation and other processing, so that the orientation of the chickens in the training image is more comprehensive; the size conversion can be proportional scaling to a preset size, such as 640 pixels, 1280 pixels and other sizes, so that the size of the chickens in the training image is more comprehensive; the image is processed by the interpolation algorithm, which can change the clarity of the image, so that the clarity of the chickens in the training image is more comprehensive; the noise processing can be Gaussian noise, salt and pepper noise, etc., which can increase the noise in the image and simulate the image under low brightness conditions, so that the situation of the chickens in the training image under different brightness is more comprehensive. In this way, increasing the amount of training image data is conducive to improving the versatility and applicability of the target recognition model and behavior recognition model obtained by the final training.

[0103] refer to Figure 6 In some embodiments of the chicken behavior statistics method of the present invention, the behavior type recognition results output by the behavior recognition model include feeding behavior, drinking behavior, walking behavior, lying behavior, fighting behavior, gathering behavior and rioting behavior; S130 includes:

[0104] S131: Calculate, according to the chicken behavior data, a first probability corresponding to the feeding behavior, a second probability corresponding to the drinking behavior, a third probability corresponding to the walking behavior, a fourth probability corresponding to the lying behavior, a fifth probability corresponding to the fighting behavior, a sixth probability corresponding to the gathering behavior, and a seventh probability corresponding to the rioting behavior;

[0105] S132: classifying the eating behavior, the drinking behavior, the walking behavior, and the lying behavior as normal behaviors, and classifying the fighting behavior, the gathering behavior, and the rioting behavior as abnormal behaviors, and calculating an eighth probability corresponding to the normal behavior and a ninth probability corresponding to the abnormal behavior;

[0106] S133: Generate behavior statistical data according to the first probability, the second probability, the third probability, the fourth probability, the fifth probability, the sixth probability, the seventh probability, the eighth probability and the ninth probability.

[0107] In the process of chicken farming, the behaviors of chickens can be summarized into seven types, namely, feeding behavior, drinking behavior, walking behavior, lying behavior, fighting behavior, gathering behavior and rioting behavior. Among them, feeding behavior, drinking behavior, walking behavior and lying behavior are normal activities of chickens, while fighting behavior, gathering behavior and rioting behavior are abnormal activities of chickens. When the comfort level is low, the probability of fighting behavior, gathering behavior and rioting behavior will increase significantly. Therefore, by calculating the probabilities of the seven types of behaviors, the probability of normal behavior and the probability of abnormal behavior, and forming behavioral statistical data, we can know the current situation of chicken farming, which is conducive to more accurate and reliable evaluation of the comfort level of chicken farming.

[0108] In some embodiments of the present invention, the probabilities corresponding to various behavior types may be calculated using the following formula:

[0109]

[0110]

[0111]

[0112] Where n is the total number of image frames; T is the time for the recognition process; S is the frame period; P i is the density of the corresponding behavior type in the i-th image frame; A i is the number of corresponding behavior types in the i-th image frame; N is the total number of chickens; P is the probability of the corresponding behavior type.

[0113] Regarding the seven types of behaviors, when a chicken stands in front of a trough, from the time the chicken's head reaches into the trough to the time the chicken's head extends from the trough and no longer reaches into the trough, it is recorded as a feeding behavior. During this period, the identification box of the chicken is given the label of eating behavior (eat). When a chicken stands in front of a waterer, from the time the chicken's head extends toward the waterer to the time the chicken's head extends from the waterer and no longer extends toward the waterer, it is recorded as a drinking behavior. During this period, the identification box of the drinking chicken is given the label of drinking behavior (drink). When a chicken stands in a breeding area, from the time the chicken moves from a certain position to the time the chicken's position in the breeding area does not change, it is recorded as a moving behavior. During this period, the identification box of the moving chicken is given the label of moving behavior (move). When a chicken is in the breeding area, from the time the chicken's belly touches the ground and lies on the ground to the time the chicken stands or performs other behaviors, it is recorded as a lying behavior. During this period, the identification box of the lying chicken is given the label of lying behavior (lie). When the chickens are in the breeding area, a fight is recorded from the time the chickens chase, peck or scratch each other with their beaks, until the chickens have other behaviors. During this period, the identification boxes of the fighting chickens are given the label of fighting behavior (fight). When the chickens are in the breeding area, a gathering behavior is recorded from the time when multiple chickens gather, stand or lie down, until the chickens have other behaviors or disperse. During this period, the identification boxes of the gathered chickens are given the label of gathering behavior (gather). When the chickens are in the breeding area, a disorder behavior is recorded from the time when multiple chickens move disorderly, until all chickens stop moving. During this period, the identification boxes of the disorderly moving chickens are given the label of disorder behavior (disorder).

[0114] The present invention also provides a chicken behavior statistics system, comprising: a processing device and at least two camera devices 900, at least one of the camera devices 900 is arranged on the top of the breeding cage, and at least one of the camera devices 900 is arranged on the side wall of the breeding cage. The processing device is communicatively connected to the camera device 900, and the processing device can execute the above-mentioned chicken behavior statistics method.

[0115] By arranging the camera devices 900 on the top and sides of the breeding cage, the chicken images can be taken from the top and from the sides at the same time to obtain a more comprehensive chicken image, and the camera device 900 transmits the taken images to the processing device. The processing device can automatically know the behavior type of the chicken by identifying the chicken and the behavior of the captured images without relying on manual observation, which is conducive to unifying the judgment standards of the behavior type, reducing data errors and improving the stability of the data, and statistics are performed according to the chicken behavior type to know the occurrence probability of each behavior type, which is conducive to more accurate and reliable evaluation of the comfort of chicken breeding.

[0116] refer to Figure 7The camera device 900 located on the top of the breeding cage can completely obtain the image of the entire breeding cage without blind spots, but the camera device 900 is far away from the chickens, and the image resolution of the chickens may be low. Figure 8 The camera device 900 located on the side wall of the breeding cage can shoot the chickens at a close distance with a high resolution, but there is a certain blind spot directly below the camera device 900 and there may be obstructions between the chickens on the other side. By arranging the camera device 900 on the top and the side wall, it is possible to obtain more comprehensive images of the chickens in the breeding cage, which is conducive to statistical analysis.

[0117] Fig. 9 An example of a physical structure diagram of an electronic device is shown in FIG. Fig. 9 As 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 may call the logic instructions in the memory 830 to execute a chicken behavior statistics method.

[0118] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0119] In yet another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute a chicken behavior statistics method provided by the above methods.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A chicken behavior statistics method, characterized in that: include: Obtaining image data to be analyzed; Inputting the image data to be analyzed into a preset target recognition model to obtain chicken detection data, wherein the chicken detection data reflects the chickens detected in the image; Inputting the chicken detection data into a preset behavior recognition model to obtain chicken behavior data, wherein the chicken behavior data reflects the behavior type of each chicken; Behavioral statistical data is generated based on the chicken behavior data, and the behavioral statistical data reflects the occurrence probability of each of the behavior types.

2. A chicken behavior statistics method according to claim 1, characterized in that, The obtaining of the image data to be analyzed includes: Acquiring captured image data, wherein the captured image data includes a plurality of image frames; According to a preset frame period, image frames are selected from the captured image data at intervals of the frame period and saved to form image data to be analyzed.

3. A chicken behavior statistics method according to claim 1, characterized in that, Before acquiring the image data to be analyzed, the method further includes: Acquire an original image dataset, wherein the original image dataset includes multiple chicken images; Classifying the original image dataset according to the chicken growth cycle to obtain at least two sets of classified image datasets; Performing an amplification process on each set of the classified image datasets to increase the number of images to form a corresponding training image dataset; Training a preset first neural network model according to each group of the training image data sets to obtain the target recognition model; According to each group of the training image data sets, the preset second neural network model is trained to obtain the behavior recognition model.

4. A chicken behavior statistics method according to claim 3, characterized in that, The step of training a preset first neural network model according to each set of training image data sets to obtain the target recognition model includes: According to each group of the training image datasets, the chickens in each image are labeled with identification boxes to form a first labeled image dataset; Dividing each set of the first labeled image data sets into a first training image set and a first test image set according to a preset first ratio; Inputting the first training image set into a preset first neural network model for training; Inputting the first test image set into the trained first neural network model, and determining whether the training is complete based on the accuracy of the target recognition result output by the first neural network model; If the training is completed, the trained first neural network model is used as the target recognition model.

5. A chicken behavior statistics method according to claim 4, characterized in that, The step of training a preset second neural network model according to each set of training image data sets to obtain the behavior recognition model includes: According to each group of the first annotated image data sets, adding a behavior label to each of the identification boxes to form a second annotated image data set; Dividing each set of the second labeled image data sets into a second training image set and a second test image set according to a preset second ratio; Inputting the second training image set into a preset second neural network model for training; inputting the second test image set into the trained second neural network model, and determining whether the training is complete based on the behavior type recognition accuracy output by the second neural network model; If the training is completed, the trained second neural network model is used as the behavior recognition model.

6. A chicken behavior statistics method according to claim 3, characterized in that, The amplification process is performed on each group of the classification image dataset to increase the number of images to form a corresponding training image dataset, including: performing direction conversion processing on the classified image dataset to form a direction conversion image dataset; and / or, performing color conversion processing on the classified image dataset to form a color converted image dataset; and / or, performing size conversion processing on the classified image dataset to form a size-converted image dataset; and / or, performing an interpolation algorithm on the classified image dataset to form an interpolated image dataset; and / or, performing noise processing on the classified image dataset to form a noise-enhanced image dataset; A training image dataset is formed according to the classification image dataset and the direction conversion image dataset and / or the color conversion image dataset and / or the size conversion image dataset and / or the interpolation image dataset and / or the noise enhancement image dataset.

7. A chicken behavior statistics method according to claim 1, characterized in that, The behavior type recognition results output by the behavior recognition model include feeding behavior, drinking behavior, walking behavior, lying behavior, fighting behavior, gathering behavior and rioting behavior; Generating behavioral statistical data based on the chicken behavior data includes: According to the chicken behavior data, respectively calculating a first probability corresponding to the feeding behavior, a second probability corresponding to the drinking behavior, a third probability corresponding to the walking behavior, a fourth probability corresponding to the lying behavior, a fifth probability corresponding to the fighting behavior, a sixth probability corresponding to the gathering behavior, and a seventh probability corresponding to the rioting behavior; Classifying the eating behavior, the drinking behavior, the walking behavior, and the lying behavior as normal behaviors, and classifying the fighting behavior, the gathering behavior, and the rioting behavior as abnormal behaviors, and calculating an eighth probability corresponding to the normal behavior and a ninth probability corresponding to the abnormal behavior; Behavior statistics are generated according to the first probability, the second probability, the third probability, the fourth probability, the fifth probability, the sixth probability, the seventh probability, the eighth probability, and the ninth probability.

8. A chicken behavior statistics system, characterized in that: include: A processing device and at least two camera devices, at least one of which is arranged on the top of the breeding cage, and at least one of which is arranged on the side wall of the breeding cage, the processing device is communicatively connected to the camera devices, and the processing device is capable of executing a chicken behavior statistics method as described in any one of claims 1 to 7.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the chicken behavior statistics method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the chicken behavior statistics method according to any one of claims 1 to 7 is implemented.

Images