Laboratory animal behavior analysis method and system

Abstract

The invention discloses a laboratory animal behavior analysis method and system, and belongs to the technical field of animal behavior analys.The laboratory animal behavior analysis method includes the steps that A00, not less than three small animals are placed in a standard animal test box, and each small animal is dyed and distinguished; a01, the AI camera identifies and marks animal actions and postures and transmits marked video segments to the AI server; a02, the AI server carries out video reconstruction on the marked video segments to obtain 3D pictures, and repeated marks are removed; a03, the AI server counts the number of times of actions and postures of each animal. The AI camera carries out real-time edge processing on a shot video, animal postures and behaviors are recognized through an AI model library, and if it is judged that the postures and behaviors conform to actions and postures needing to be counted, relevant marks are synchronously marked in video stream signals; transmitting the marked video to an AI server; and the AI server performs screening statistics on the identification videos and gives a statistical result, so that the contrast experiment efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of animal behavior analysis, and in particular to a laboratory animal behavior analysis method and system. Background Art

[0002] In vivo experiments on drugs are an effective way to verify the effectiveness and side effects of drugs. Laboratory animal behavior analysis evaluates the psychological and physiological state of animals by observing and recording their behavioral responses at various dosages and in different environments; such small animals include rabbits, mice, guinea pigs and other limbed animals. This method has a wide range of applications in medicine, pharmacy, life sciences and other fields. It can help researchers understand the functional efficacy, side effects, anxiety and depression of drugs on animals.

[0003] The open field test (OFT) is a commonly used animal behavior experimental method. For example, it can detect the spontaneous activities and movements of small animals after medication. It is a method to evaluate the behavior, flexibility and tension of experimental animals under the influence of drugs. The application of OFT experiments is very wide, running through small animal behavioral research, and extended to neurological and pharmacological research. Taking antipsychotic drugs as an example, OFT is generally used to detect the efficacy and side effects of drugs. At present, OFT is often used as a control to reflect various spontaneous activities of experimental animals. Because of its simple experimental operation and rich data volume, one experiment can quantitatively evaluate the spontaneous activities, behavioral movements, anxiety and depression states of animals, and it is widely used in drug research and development.

[0004] The open-box test (OFT) usually uses a special animal test box, such as the common standard small animal test box, which contains one or more small animals. When conducting pharmacological research, the small animals can be fed or injected with corresponding drugs, and their specific behaviors and postures can be observed. By making statistical results over a certain period of time, we can understand the effects of drugs on animals, pharmacokinetic performance and side effects, and provide more accurate and objective drug efficacy evaluation results.

[0005] At present, animal behavior is usually observed and recorded by experimenters, which is time-consuming and labor-intensive, and prone to errors, resulting in large deviations in observation results and poor experimental repeatability. Moreover, when there are multiple animal experimental subjects, manual morphological observation and recording are not realistic. There are already systems with automatic analysis on the market, which use high-definition cameras to record in real time, identify and count various behaviors of animals, and can give corresponding reports, reducing the manual burden and improving the accuracy of observation. Currently, the commercial laboratory animal behavior video analysis systems on the market are all for the recognition of the actions and postures of individual animals, which have certain limitations.

[0006] Because many animal experiments on drugs require double-blind experiments or comparative experiments, it is necessary to feed or inject different doses of drugs to multiple animals at the same time. In the same environment, the differences in the behavior and posture of the animals are observed to understand the effects of drug dosage on the animals. To solve the above problems, the present invention provides a laboratory animal behavior analysis method and system. Summary of the invention

[0007] In view of the above-mentioned technical deficiencies, the purpose of the present invention is to provide a laboratory animal behavior analysis method and system. The AI ​​camera performs real-time edge processing on the captured video, and identifies the posture and behavior of the small animal through the AI ​​model library. If the posture and behavior of the small animal are determined to be consistent with the action and posture that need to be counted, the relevant mark can be synchronously added to the video stream signal. The marking information includes the type of action and the corresponding timestamp. The marked video is transmitted to the AI ​​server; the AI ​​server screens and counts the marked video and gives the statistical results, which solves the technical problems raised in the background technology.

[0008] In order to solve the above technical problems, the present invention adopts the following technical solution: The present invention provides a laboratory animal behavior analysis method, comprising the following steps:

[0009] A00: Place no less than three small animals in a standard animal test box, and dye each small animal with a different color to distinguish them;

[0010] A01: Several AI cameras are symmetrically placed on the top of the standard animal test box to identify and mark the movements and postures of each animal, and transmit the marked video segments to the AI ​​server;

[0011] A02: The AI ​​server reconstructs 3D images from multiple camera perspectives for the marked video segments to remove repeated markings of the same animal and the same action by multiple cameras;

[0012] A03: The AI ​​server counts the number of movements and gestures of each animal.

[0013] The method for analyzing laboratory animal behavior according to claim 1, wherein the motion skeleton including the estimation of the animal's behavior posture is determined before A01:

[0014] Define the motion skeleton of the small animal, and use 16 structural points to construct the motion skeleton of the small animal, namely: nose, left ear, right ear, neck, left forelimb, right forelimb, left hind limb, right hind limb, left front paw, right front paw, left hind paw, right hind paw, back center, tail root, tail middle and tail tip; among them, the nose, left ear and right ear locate the movement of the small animal's head; the neck is the connection point between the head and the trunk; the neck, left forelimb, right forelimb, left hind limb, right hind limb, back center and tail root constitute the body outline of the small animal, which can describe the changes in the body outline; the left forelimb and left forepaw, right forelimb and right forepaw, left hind limb and left hind paw, and right forelimb and right hind paw simplify the small animal's limbs into line segments respectively, describing the movement state of the small animal's limbs.

[0015] Preferably, A01 also includes animal behavior posture estimation and AI deep learning training:

[0016] In the marking process, the unobstructed parts of the animal's body are directly marked, and the obstructed parts of the animal's body are estimated by observing the synchronized video frames from other camera positions; the behavior postures of the animals that meet the motion models in the AI ​​model library will be marked and manually cross-validated;

[0017] Use AI servers to implement DLC model training and model prediction. Through massive data model training, the labeled small animal behavior video and the model prediction labeling reach the required accuracy; an AI model library will be generated for AI chips to call.

[0018] Preferably, A02 includes recognition and classification of animal behavior postures:

[0019] The AI ​​camera's built-in AI main control chip calls the AI ​​model library to identify, classify and mark the effective behavioral postures of small animals based on the captured video, and uploads the classified and marked video to the AI ​​server for processing.

[0020] Preferably, A03 includes the statistical and analytical results of animal behavior posture:

[0021] The server reconstructs and filters the specific action videos of each marked animal, removes duplicate tags, and provides data statistics and analysis results of each animal's behavior and posture.

[0022] A laboratory animal behavior analysis system includes a plurality of AI cameras, an AI server and an AI model library; the plurality of AI cameras are centrally symmetrically installed in a standard test box for shooting small animals without blind spots; the AI ​​cameras call the AI ​​model library through a built-in AI main control chip, compare the shot videos, identify, classify and mark the effective behavior postures of the small animals, and upload the classified and marked videos to the AI ​​server for processing; the AI ​​server reconstructs and screens the specific action videos of each marked animal, removes duplicate marks, and provides data statistics and analysis results of the behavior and posture of each small animal.

[0023] Preferably, the AI ​​model library stores the behavior and posture video models of small animals; the posture video model is obtained by customized training of the selected open source large model by the AI ​​server, and learning and training the large model with a large number of known small animal behaviors and postures.

[0024] Preferably, the AI ​​main control chip performs preliminary recognition and screening of full-time videos, distinguishing between invalid time period videos and valid time period videos; performs behavior and status detection for valid time period video processing, and recognizes, classifies and locates the behavior and posture of each small animal.

[0025] The beneficial effects of the present invention are as follows: the AI ​​camera of the present invention performs real-time edge processing on the captured video, and identifies the posture and behavior of the small animal through the AI ​​model library. If the posture and behavior of the small animal are determined to be consistent with the action and posture that need to be counted, relevant marks can be synchronously added to the video stream signal, and the marking information includes the type of action and the corresponding timestamp. The marked video is transmitted to the AI ​​server; the AI ​​server screens and counts the marked video, gives the statistical results, and realizes the simultaneous monitoring and statistics of the action postures of multiple experimental mice, reducing manpower and material resources, and improving the efficiency of comparative experiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only 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.

[0027] Figure 1 A flowchart of a laboratory animal behavior analysis method provided in an embodiment of the present invention.

[0028] Figure 2 A schematic diagram of the structure of a laboratory animal behavior analysis system provided in an embodiment of the present invention.

[0029] Explanation of the accompanying drawings: 1-AI camera, 2-AI server, 3-standard test box. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] Example: Figure 1 As shown, the present invention provides a method for analyzing laboratory animal behavior, comprising the following steps, which are described using mice as an example in this embodiment:

[0032] A00: Place no less than three mice in a standard animal test box, and dye each mouse with a different color to distinguish them so that the AI ​​camera can distinguish different mice. In this embodiment, six medical mice were used to observe their activities in a test box for 2-48 hours around the clock; the behavior after drug intervention was recorded and analyzed; the mouse behavior was divided into two categories: pathological behavior (scratching the nose, turning in circles, etc.) and non-pathological behavior (eating, sleeping, excretion, etc.).

[0033] A01: Several AI cameras are set up symmetrically on the top of the standard animal test box, which can automatically identify a single or multiple mice, realize the movement tracking of a single or multiple mice, identify and mark the movements and postures of each mouse, and transmit the marked video segments to the AI ​​server. This step also includes the following processes:

[0034] The estimated motion skeleton of the mouse behavior posture is determined: the motion skeleton of the mouse is defined, and 16 structural points are used to construct the motion skeleton of the mouse, namely: nose, left ear, right ear, neck, left forelimb, right forelimb, left hindlimb, right hindlimb, left front paw, right front paw, left hind paw, right hind paw, back center, tail root, tail middle and tail tip; among them, the nose, left ear and right ear locate the movement of the mouse head; the neck serves as the connection point between the head and the trunk; the neck, left forelimb, right forelimb, left hind limb, right hind limb, back center point and tail root constitute the body contour of the mouse, which can describe the changes in the body contour; the left forelimb and left forepaw, right forelimb and right forepaw, left hind limb and left hind paw, and right forelimb and right hind paw simplify the mouse's limbs into line segments, respectively, to describe the movement state of the mouse's limbs.

[0035] Estimation of animal behavior posture and AI deep learning training: Three AI cameras are used in this embodiment, each camera is distributed on the upper edge of the experimental box, and the lenses are 120 degrees to each other and face the central bottom surface of the experimental box; when shooting, the AI ​​camera captures the mouse observed from an oblique upper perspective. In the marking, the unobstructed part of the mouse's body is directly marked, and the obstructed part of the mouse's body is estimated by observing the synchronized video frames under other camera positions; the behavioral posture of the filmed mouse that conforms to the motion model in the AI ​​model library will be marked and manually cross-validated. Three AI cameras, each AI camera is distributed on the upper edge of the experimental box to mark about hundreds of frames of the mouse behavior video, which can track the 16 structural points of the mouse. Use the AI ​​server to implement DLC model training and model prediction, and train the model through massive data until the marked mouse behavior video and the model prediction reach the required accuracy; an AI model library will be generated for the AI ​​chip to call.

[0036] A02: The AI ​​server reconstructs 3D images from multiple camera perspectives for the marked video segments, which is used to remove repeated markings of the same animal and the same action by multiple cameras. Specifically, it includes the recognition and classification of mouse behavior postures: the AI ​​camera has a built-in AI main control chip that calls the AI ​​model library to recognize, classify and mark the effective behavior postures of mice based on the captured video, and upload the classified and marked videos to the AI ​​server for processing. In fact, the AI ​​high-definition camera can automatically track the movement of one or more mice, and recognize, mark and classify the behavior and postures of mice. The classification information of the recognized actions is marked at the time when the relevant actions in the video stream occur. The server can quickly retrieve these marks, make analysis and statistics. When there is doubt about the recognition of mouse actions, you can quickly jump to the relevant time period to play back the video, confirm or correct it. For video segments that are confirmed to have misjudgment, they should be collected as samples for deep algorithm learning to increase the accuracy of judgment.

[0037] A03: The AI ​​server counts the number of actions and postures of each mouse. Specifically, it includes the statistical and analytical results of the mouse's behavior and posture: the server reconstructs and filters the specific action videos of each marked mouse, removes duplicate marks, and gives the statistical and analytical results of each mouse's behavior and posture.

[0038] In this embodiment, several AI cameras are symmetrically arranged on the test box to perform video monitoring of the behavior postures of mice with different identifications. The AI ​​camera compares the captured video with the AI ​​model library to identify the posture behavior of the mice, and transmits the identified video to the AI ​​server; the AI ​​server screens and counts the identified video, gives statistical results, and realizes simultaneous monitoring and statistics of the movement postures of multiple experimental mice, reducing manpower and material resources and improving the efficiency of comparative experiments.

[0039] See also Figure 2 As shown, the laboratory animal behavior analysis system includes several AI cameras 1, AI server 2 and AI model library; several AI cameras 1 are centrally symmetrically installed in a standard test box 3 for shooting mice without blind spots; the AI ​​camera 1 calls the AI ​​model library through the built-in AI main control chip, compares the shooting video, identifies, classifies and marks the effective behavior posture of the mouse, and uploads the classified and marked video to the AI ​​server for processing; the AI ​​server 2 reconstructs and screens the specific action video of each marked mouse, removes duplicate marks, and gives the data statistics and analysis results of each mouse's behavior and posture. In addition, the system can set an alarm function to remind staff to intervene when mice die, twitch, bite each other or other emergencies; the software can be customized to meet the needs of specific drugs or different experimental mice.

[0040] Specifically, the AI ​​model library stores the behavior and posture video model of mice; the posture video model is obtained by learning and training the large model with a large number of known mouse behaviors and postures based on the customized training of the selected open source large model by the AI ​​server 2. The action video of the mouse needs to be manually judged and marked before training the model library through AI big data. With the continuous accumulation of raw data of the experiment, the recognition rate of the model library will gradually increase. There will also be databases for different drugs. The effective video is trained by deep learning customized by the large model AI to generate database information for identifying various actions. After meeting the required behavior and posture recognition accuracy, the corresponding library file is generated for the AI ​​chip to call as a tool library for recognition. The database is the artificial neurons generated after deep learning of big data, the information of neuron connections and excitation functions, and the best matching results will be given after inputting the pictures and video signals to be recognized. This is the basic principle of recognition. Usually a lot of training is required to achieve relatively accurate recognition results.

[0041] More specifically, the AI ​​main control chip performs preliminary identification and screening of full-time videos, distinguishing between invalid time period videos and valid time period videos; for valid time period video processing, behavior and state detection is performed, and the behavior and posture of each mouse are identified, classified and located. For different drug experiments, such as animal experiments on psychiatric drugs and dermatological drugs, the statistical actions are different. For example, the frequency of mice running, falling, wiping eyes, pinching noses, scratching, etc., will be used as the basis for statistics. The video server can manage one or more standard test chambers, can realize video storage, mark fast playback, retrieve and count the markers of the above-mentioned various animal behaviors and postures, and provide relevant analysis reports.

[0042] In this embodiment, the AI ​​camera performs real-time edge processing on the captured video, and identifies the posture and behavior of the small animals through the AI ​​model library. If the posture and behavior of the small animals are determined to be consistent with the actions and postures that need to be counted, relevant marks can be synchronously added to the video stream signal. The marking information includes the type of action and the corresponding timestamp. The marked video is then transmitted to the AI ​​server; the AI ​​server screens and counts the marked video, and gives the statistical results, so as to realize the simultaneous monitoring and statistics of the action postures of multiple experimental mice, reduce manpower and material resources, and improve the efficiency of comparative experiments.

[0043] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for analyzing laboratory animal behavior, characterized in that: The steps include: A00: Place no less than three small animals in a standard animal test box, and dye each small animal with a different color to distinguish them; A01: Several AI cameras are symmetrically placed on the top of the standard animal test box to identify and mark the movements and postures of each animal, and transmit the marked video segments to the AI ​​server; A02: The AI ​​server reconstructs 3D images from multiple camera perspectives for the marked video segments to remove repeated markings of the same animal and the same action by multiple cameras; A03: The AI ​​server counts the number of movements and gestures of each animal.

2. The method for analyzing laboratory animal behavior according to claim 1, characterized in that: A01 includes the following steps to determine the motion skeleton of the animal: Define the motion skeleton of the small animal, and use 16 structural points to construct the motion skeleton of the small animal, namely: nose, left ear, right ear, neck, left forelimb, right forelimb, left hind limb, right hind limb, left front paw, right front paw, left hind paw, right hind paw, back center, tail root, tail middle and tail tip; among them, the nose, left ear and right ear locate the movement of the small animal's head; the neck is the connection point between the head and the trunk; the neck, left forelimb, right forelimb, left hind limb, right hind limb, back center and tail root constitute the body outline of the small animal, which can describe the changes in the body outline; the left forelimb and left forepaw, right forelimb and right forepaw, left hind limb and left hind paw, and right forelimb and right hind paw simplify the small animal's limbs into line segments respectively, describing the movement state of the small animal's limbs.

3. The method for analyzing laboratory animal behavior according to claim 2, characterized in that: Before A01, it also included animal behavior posture estimation and AI deep learning training: In the marking process, the unobstructed parts of the animal's body are directly marked, and the obstructed parts of the animal's body are estimated by observing the synchronized video frames from other camera positions; the behavior postures of the animals that meet the motion models in the AI ​​model library will be marked and manually cross-validated; Use AI servers to implement DLC model training and model prediction. Through massive data model training, the labeled small animal behavior video and the model prediction labeling reach the required accuracy; an AI model library will be generated for AI chips to call.

4. The method for analyzing laboratory animal behavior according to claim 3, characterized in that: A02 includes the recognition and classification of animal behavior postures: The AI ​​camera's built-in AI main control chip calls the AI ​​model library to identify, classify and mark the effective behavioral postures of small animals based on the captured video, and uploads the classified and marked video to the AI ​​server for processing.

5. The method for analyzing laboratory animal behavior according to claim 4, characterized in that: A03 includes the statistical and analytical results of animal behavior: The server reconstructs and filters the specific action videos of each marked animal, removes duplicate tags, and provides data statistics and analysis results of each animal's behavior and posture.

6. The laboratory animal behavior analysis system according to any one of claims 1 to 6, characterized in that: The system comprises a plurality of AI cameras (1), an AI server (2) and an AI model library; the plurality of AI cameras (1) are centrally symmetrically installed in a standard test box (3) for shooting small animals without blind spots; the AI ​​cameras (1) call the AI ​​model library through a built-in AI main control chip, compare the shot videos, identify, classify and mark the effective behavior postures of the small animals, and upload the classified and marked videos to the AI ​​server for processing; the AI ​​server (2) reconstructs and screens the specific action videos of each marked animal and removes duplicate marks; and provides data statistics and analysis results of the behavior and posture of each small animal.

7. The laboratory animal behavior analysis system according to claim 6, characterized in that: The AI ​​model library stores the behavior and posture video models of small animals; the posture video models are obtained by customizing and training the selected open source large model based on the AI ​​server (2), using a large number of known small animal behaviors and postures to learn and train the large model.

8. The laboratory animal behavior analysis system according to claim 7, characterized in that: The AI ​​main control chip performs preliminary recognition and screening of all-time videos, distinguishing between invalid time period videos and valid time period videos; performs behavior and status detection on valid time period videos, and recognizes, classifies and locates the behavior and posture of each small animal.

Images