System and method for high-throughput phenotype acquisition and analysis of multi-target mice throughout their growth period

Abstract

The present invention relates to a high-throughput phenotype acquisition and analysis system and method for multi-target mice throughout their growth period. The present invention constructs a cage system suitable for the life of multi-target experimental mice, uses two depth cameras to observe the experimental mice from two rectangular holes on the top of the cage, obtains RGB image data and depth image data, and uses an RGB camera to observe downward from the rounded rectangular hole on the top of the cage to obtain video data of the experimental mice. Based on a deep learning network and a multi-target tracking algorithm, mice are identified in real time and each mouse is accurately tracked, and key points such as the nose, ears, neck, back center, and tail base of each mouse are extracted. The behavior of a single mouse and the social behavior between mice are estimated based on the key points, and the behavioral parameters such as the time, duration, and frequency of each behavior are determined, and the health status of the mice is analyzed based on these behavioral parameters. The present invention can accurately extract the phenotypes of multi-target mice throughout their life cycle, with low cost and convenient and quick operation.

Description

Technical Field

[0001] The invention belongs to the technical field of animal behavior and automated experimental equipment, and specifically relates to a high-throughput phenotype acquisition and analysis system and method for multi-target mice throughout their growth period. Background Art

[0002] Animal behavior is the body language of animals to express their psychology and physiology, and it is also a manifestation of the comprehensive functions of animals themselves. Animal behavior is a discipline that studies the interaction between animals and the environment and other organisms. It not only observes the natural behavior of animals, but more importantly, observes the behavior of animals and its changes under specific laboratory conditions, and then studies the neurological functions and psychological processes of experimental animals, analyzes the effects of drugs, etc.

[0003] Mice are typical model animals. As mammals, their genes are more than 98% similar to human genes. Many diseases that are difficult to cure in humans can be simulated in mice. The analysis of different behaviors of mice is widely used to measure experimental effects in biology, neuroscience, pharmacology, and genetics.

[0004] The traditional measurement method of mouse behavior analysis is to conduct statistical analysis based on the results of manual observation, which is time-consuming and labor-intensive. Not only can it not realize some experiments that require long-term observation, but it also cannot objectively evaluate and measure the behavior, and is greatly influenced by subjective factors.

[0005] There are still several obvious deficiencies in the existing machine vision-based mouse behavior detection. First, the detection flux is low, the target mice are few and mainly suitable for short-term experiments. There is no report on the equipment and methods for the whole life cycle mouse behavior detection, which leads to the inability of the prior art to simultaneously observe the phenotypes of different varieties of mice in the whole life cycle, thereby causing deviations in the later study of the correlation analysis between mouse gene-nutrition-phenotype. Secondly, due to the extremely similar appearance of mice, the multi-target mouse tracking accuracy is not high, and it is easy to lose and follow, resulting in the exchange of IDs between mice and data confusion. In addition, the prior art mostly uses RGB cameras to obtain the two-dimensional phenotype of mice, and there is no report on extracting the three-dimensional phenotype of mice and tracking multi-target mice. The three-dimensional point cloud adds depth information, which can use depth information to track multi-target mice more accurately, and the three-dimensional point cloud can obtain phenotypes that cannot be obtained by two-dimensional data, such as the weight of mice, the nesting behavior of mice, etc. Summary of the invention

[0006] 1. Technical issues to be resolved

[0007] In order to solve the technical problems of low throughput, inaccurate tracking, short test cycle and incomplete acquisition of phenotypic parameters, the present invention proposes a high-throughput phenotypic acquisition and analysis system and method for multi-target mice throughout their growth period.

[0008] (II) Technical solution

[0009] In order to solve the technical problem, the present invention provides a system and method for obtaining and analyzing multi-target mouse phenotypes during their entire growth period with high throughput.

[0010] A multi-target mouse full-growing period high-throughput phenotype acquisition and analysis system, characterized in that it includes a mouse cage system, a rack, a T-shaped bracket system, a depth camera, an RGB camera, a laptop computer, a switch, a video recorder, a display screen and a workstation, wherein:

[0011] The mouse cage system includes multiple sets of mouse cage systems, each set of mouse cage system includes a mouse cage, a drawer, a water bottle and a food box; the mouse cage adopts a rectangular box design, which is used to place experimental mice, so that the experimental mice can move freely inside the mouse cage; the top surface of the mouse cage is provided with multiple small round holes for ventilation, and rectangular holes are respectively opened on both sides of the top surface; the water bottle and the food box are detachably installed on the side wall of the mouse cage, respectively used to provide water and food for the experimental mice; the drawer is located directly below the mouse cage and is covered with corn cob bedding to prevent mouse excrement from contaminating the mouse cage;

[0012] The shelf adopts a multi-layer storage rack design, and each layer can place multiple cage systems side by side with equal spacing; each layer of the shelf is equipped with multiple buckles, and each cage system can be detachably installed in a fixed position with the buckles and the shelf;

[0013] The T-shaped bracket system includes a plurality of T-shaped brackets, which are fixedly mounted on the rack in an inverted T-shaped pattern so that they are located just above each mouse cage system;

[0014] Depth cameras are installed at the ends of both sides of the lower crossbar of each T-shaped bracket. The depth cameras face the inside of the cage through two rectangular holes on the top surface of the cage to observe the experimental mice.

[0015] An RGB camera is detachably installed at the center of the top surface of each mouse cage. The RGB camera has infrared night vision function and can continuously observe the experimental mice for 24 hours.

[0016] The laptop is connected to the depth camera and is used to store RGB image data and depth image data collected by the depth camera;

[0017] The video recorder is connected to the RGB camera through a switch and is used to store the video data collected by the RGB camera in real time;

[0018] The display screen is connected to the video recorder to display and monitor the activities of mice in all cages in real time;

[0019] The workstation is connected to a laptop and a video recorder to process and analyze the video and image files of all mice acquired by the RGB camera and the depth camera, obtain the two-dimensional and three-dimensional phenotypic parameters of multi-target mice throughout their life cycle, including the frequency and duration of various behaviors including eating, drinking, sleeping, standing, fighting and chasing, and analyze the health status of the mice based on the obtained result parameters.

[0020] Preferably, a rounded rectangular hole is opened at the center of the top surface of the mouse cage, and a small groove is provided on one side of the rounded rectangular hole for passing the cable; the RGB camera is detachably mounted on a rounded rectangular plate that matches the shape and size of the rounded rectangular hole, and the rounded rectangular plate is also provided with a groove corresponding to the rounded rectangular hole for passing the cable; the rounded rectangular plate is designed to be inserted into the rounded rectangular hole, so that the RGB camera is located in the internal space of the mouse cage and observes the mouse, and the cable of the RGB camera is led out through the small groove to connect to an external device; after the observation is completed, the rounded rectangular plate with the RGB camera installed can be removed from the rounded rectangular hole by lifting the cable.

[0021] Preferably, the rack is made of 304 stainless steel.

[0022] Preferably, the mouse cage is composed of 6 transparent polyphenylene sulfone resin plates, with a length, width and height of 265×265×310 mm.

[0023] Preferably, the two depth cameras above the cage are both tilted toward the interior of the cage, with the tilt angles of 65° to the horizontal direction, and the height difference between the bottom of the depth camera and the top of the cage is 1 cm.

[0024] A method for obtaining and analyzing multi-target mice throughout their growth period with high-throughput phenotypes, which uses the aforementioned multi-target mice throughout their growth period with high-throughput phenotypes, and specifically includes the following steps:

[0025] Step 1, disinfect the rack and mouse cage system;

[0026] Step 2, multiple cage systems are mounted on the rack by means of buckles, and the two depth cameras on the T-shaped bracket are ensured to be aligned with the interior space of the cage through the rectangular holes on both sides of the top surface of the cage;

[0027] Step 3: Insert the rounded rectangular plate with the RGB camera into the rounded rectangular hole on the top surface of the cage, making sure that the RGB camera is aligned with the interior space of the cage;

[0028] Step 4, cover the bottom of the drawer of the mouse cage with corn cob bedding and place the mouse in the mouse cage;

[0029] Step 5: The system is powered on, the depth camera acquires the RGB image and depth image of the mouse, the RGB camera acquires the video data of the mouse, and the data is stored in real time;

[0030] Step 6: The staff regularly removes the cage system, removes the mice from the cages, disinfects the cage system, replaces the corncob bedding, reinstalls the cage system and the rounded rectangular board, puts the mice in, and continues to observe;

[0031] Step 7: The workstation regularly performs digital image processing and analysis on the images and videos of mice, completes the observation of the entire life cycle of multi-target mice, obtains the phenotypic parameter information of the mice, and analyzes the health status of the mice based on the obtained result parameters.

[0032] Preferably, in step 7, the deep learning neural network YOLOv7 and the improved StrongSORT multi-target tracking algorithm are used to process and analyze the video data collected by the RGB camera to obtain the two-dimensional phenotypic parameter information of the mouse, and the specific method is as follows:

[0033] (1) Label video images, create datasets, and train a mouse recognition model based on YOLOv7 offline;

[0034] (2) Mouse video data is input, and the YOLOv7 mouse recognition model is used to identify multiple target mice in each frame of the video and generate target detection frames. The multi-target tracking algorithm based on the improved StrongSORT is used to assign a specific ID number to each target detection frame and track it;

[0035] (3) Based on the YOLOv7 mouse detection results, identify multiple key points of each mouse, including the mouth, ears, neck, back center, and tail base; if there is occlusion, repair the key points based on the information of the key points of the previous and next frames and the repair technology;

[0036] (4) Estimate the posture of each mouse in each frame of video based on the key point position information;

[0037] (5) Calculate the behavior of each mouse and the social behavior between mice based on the posture and time series of the mice in each frame of the video image.

[0038] Preferably, in step 7, the specific improvement of the improved StrongSORT multi-target mouse tracking algorithm is as follows: in the current frame video, YOLOv7 is used to identify the mouse and generate a detection frame, and the generated detection frame is cascade matched and IOU matched with the detection frame, i.e., the trajectory, assigned a specific ID number in the previous frame video. If there are trajectories that do not match the detection frame and detection frames that do not match the trajectory, the Euclidean distance between them is calculated, and the ID number of the trajectory that does not match the detection frame with the smallest Euclidean distance is assigned to the detection frame that does not match the trajectory.

[0039] Preferably, in step 7, the RGB images and depth image data collected by two depth cameras are analyzed and processed to obtain the three-dimensional phenotypic parameter information of the mouse, and the specific method is as follows:

[0040] (1) Align the RGB image with the depth image, reconstruct the 3D point cloud through the depth image, assign RGB values, i.e., colors, to the 3D point cloud through the aligned RGB image, and reconstruct the 3D point cloud with two depth cameras respectively;

[0041] (2) The point clouds reconstructed by the two depth cameras are registered through the external parameter calibration between the two depth cameras and synthesized into a three-dimensional point cloud;

[0042] (3) The reconstructed 3D point cloud is divided into two categories: mice and background, the data set is annotated, and the pointnet++ network is trained to extract the mouse point cloud from the background;

[0043] (4) Predict the next position of the mouse based on Kalman filtering, and use the Hungarian algorithm to match the predicted position with the position of the mouse segmented in (3) and track it;

[0044] (5) Segment the mouse 3D point cloud into four parts: head, back, tail, and tail, annotate the data set, and train the pointnet++ network;

[0045] (6) estimating the posture of each mouse based on the position information of the mouse parts;

[0046] (7) Calculate the behavior of each mouse and the social behavior between mice based on the posture and time series of the mice in the point cloud.

[0047] Preferably, in step 7, the phenotypic parameter information of the mice is obtained, including the frequency and duration of the behaviors of eating, drinking, sleeping, standing, fighting and chasing, and the health of the mice is analyzed according to the obtained result parameters, specifically using the following detailed steps:

[0048] (1) Identify and detect various behaviors of mice:

[0049] Eating: if the mouse’s mouth was in the food box for more than 2 seconds, it was considered eating, and the time the mouse’s mouth was in the food box was recorded as the duration;

[0050] Drinking water: if the mouse's mouth is at the mouth of the water bottle for more than 2 seconds, it is judged as drinking water, and the time the mouse's mouth is at the mouth of the water bottle is recorded as the duration;

[0051] Sleeping: calculate the change in the position of key parts of the mouse, including the head, back and tail, in adjacent frames. If the change is less than 5 pixels, the mouse is considered to be stationary. If the stationary time is greater than 100 seconds, the mouse is considered to be sleeping, and the duration is recorded.

[0052] Standing, assuming that the position of the mouse's mouth in the image is A, the position of the center of the back is B, the position of the base of the tail is C, the distance between the mouth and the center of the back is a, and the distance between the center of the back and the base of the tail is b. If 0.9<a / b<1.1 and the angle between AB and BC is less than 30°, it is judged to be standing, and the duration is recorded;

[0053] Fighting: if the relative movement speed of the two mice was greater than 10 cm / s and they were in contact with each other in the next frame, it was considered a fight and the duration was recorded;

[0054] Chase: two mice move in phase, the mouth of one mouse is behind the tail of the other mouse, and the speed of both mice is greater than 5 cm / s, and the relative speed of the two mice is less than 2 cm / s. This is considered a chase, and the duration is recorded;

[0055] (2) Count the number of times each behavior occurs in 24 hours a day, which is the frequency. Add up the duration of each behavior and record the frequency and duration of each behavior.

[0056] (3) A health index is established based on various behaviors, their frequency of occurrence, and duration, and the health status of mice is analyzed based on the health index.

[0057] (III) Beneficial effects

[0058] Compared with the prior art, the beneficial effects of the present invention are embodied in at least three aspects.

[0059] Firstly, the mouse cage system designed and manufactured by the present invention meets the requirements for SPF-level animal breeding. It adopts an inverted T-shaped bracket to fix the depth camera, a rounded rectangular plate to fix the RGB camera, and a drawer to place the bedding, which is convenient for the installation and disassembly of the hardware structure and sensor components, making it more convenient and quick to regularly disinfect the mouse cage and replace the bedding, and provides a solid hardware guarantee for obtaining and analyzing the two-dimensional and three-dimensional phenotypes of multi-species and multi-target mice throughout the reproductive period.

[0060] Secondly, the StrongSORT multi-target tracking algorithm was improved. During the tracking process, after cascade matching and IOU matching, if there are tracks that are not matched with the detection frame and detection frames that are not matched with the tracks, a secondary matching is performed. This solves the problems of blank frames and mismatches in the multi-target mouse tracking process, effectively improves the multi-target tracking accuracy, and ensures the quality of scientific research data related to mouse gene-nutrition-phenotype in the later stage.

[0061] In addition, the present invention can simultaneously and accurately extract the two-dimensional and three-dimensional phenotypes of multiple mice in multiple cages, and is suitable for both short-term and full-life cycle experiments on mice. The RGB camera and two depth cameras are used to accurately obtain the two-dimensional and three-dimensional phenotypes, and the phenotypic parameters are more comprehensive. The health status of mice can be evaluated based on the two-dimensional and three-dimensional phenotypes of mice, which is of great significance for the later establishment of the association analysis between gene-nutrition-phenotype.

[0062] Finally, the squirrel cage system provided by the present invention has a modular structure design, relatively low cost, convenient and quick operation, and has the potential for large-scale promotion and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a schematic structural diagram of the rack and cage system of the present invention.

[0064] Figure 2 Schematic diagram of the structure of the mouse cage system and sensor components (head view).

[0065] Figure 3 Schematic diagram of the structure of the mouse cage system and sensor components (oblique top view).

[0066] Figure 4 This is a workflow diagram embodiment of the present invention.

[0067] Figure 5 The multi-target mouse tracking results based on the original StrongSORT.

[0068] Figure 6 The results of multi-target mouse tracking based on the improved StrongSORT. DETAILED DESCRIPTION

[0069] In order to solve the technical problem, the present invention provides a multi-target mouse full-growing period high-throughput phenotype acquisition and analysis system and method. The technical solution of the present invention is now further explained in conjunction with the accompanying drawings of the specification.

[0070] The present invention provides a multi-target mouse full-growing period high-throughput phenotype acquisition and analysis system and method, the system hardware design is as follows Figure 1 , Figure 2 and Figure 3 As shown, it includes a cage system, a rack, a T-bracket system, a depth camera, an RGB camera, a laptop, a switch, a video recorder, a display screen, and a workstation.

[0071] The mouse cage system includes multiple sets of mouse cage systems, each set of mouse cage system includes a mouse cage, a drawer, a water bottle and a food box; the mouse cage is customized, adopts a rectangular box design, and is fixed by six transparent PPSU plates through 304 stainless steel M4 screws. PPSU material and 304 stainless steel are corrosion-resistant, high temperature-resistant and durable. The mouse cage is used to place experimental mice, which is convenient for experimental mice to move freely inside the mouse cage; the top surface of the mouse cage is provided with 26 circular holes with a diameter of 15mm, a length × width of 96 × 56mm, and a hole with a diameter of 15mm. There are rounded rectangular holes and two rectangular holes with a length × width of 110 × 40 mm, of which 26 holes with a diameter of 15 mm are used to diffuse the odor in the mouse cage; the water bottle and food box are detachably installed on the side wall of the mouse cage, with a height of 150 mm and 50 mm from the bottom of the mouse cage, respectively, to provide water and food for the experimental mice; the drawer is located directly below the mouse cage and is covered with corn cob bedding to prevent mouse excrement from contaminating the mouse cage. The drawer structure makes it more convenient and quick to replace the bedding and manually weigh the mice.

[0072] The shelf adopts a multi-layer storage rack design, and multiple mouse cage systems can be placed side by side at equal intervals on each layer; each layer of the shelf is equipped with multiple buckles, and each mouse cage system can be detachably installed in a fixed position with the shelf through the buckles, which is convenient for regular disinfection of the mouse cages. The position after each disassembly and installation is fixed, so that the field of view of the RGB camera and depth camera remains unchanged when collecting data.

[0073] The T-shaped bracket system includes a plurality of T-shaped brackets, which are fixedly mounted on the rack in an inverted T-shaped pattern so as to be located just above each mouse cage system.

[0074] Two depth cameras are installed at the two ends of the lower crossbar of each T-shaped bracket. Both depth cameras observe the experimental mice in the cage through two rectangular holes on the top surface of the cage in an inclined manner. The inclination angle is 65° to the horizontal direction. The height difference between the bottom of the depth camera and the top of the cage is 1 cm. The two depth cameras are connected by an audio cable to ensure that the timestamps of the two depth cameras when collecting data are consistent. The laptop is connected to the depth camera via a USB cable to store the RGB image data and depth image data collected by the depth camera.

[0075] The rounded rectangular hole on the top surface of the mouse cage is provided with a small groove on one side for passing the cable; the RGB camera is detachably mounted on a rounded rectangular plate that matches the shape and size of the rounded rectangular hole, and the rounded rectangular plate is also provided with a small groove corresponding to the rounded rectangular hole for passing the cable; the rounded rectangular plate is designed to be inserted into the rounded rectangular hole, so that the RGB camera is located in the internal space of the mouse cage and faces the mouse for observation, and the cable of the RGB camera is led out through the small groove to connect to external equipment; after the observation is completed, the rounded rectangular plate with the RGB camera installed can be removed from the rounded rectangular hole by lifting the cable.

[0076] The RGB camera has infrared night vision function and can observe the experimental mice continuously for 24 hours. The video recorder is connected to the RGB camera through a switch to store the video data collected by the RGB camera in real time.

[0077] The display screen is connected to a video recorder to display and monitor the activities of mice in all cages in real time.

[0078] The workstation is connected to a laptop and a video recorder to process and analyze the video and image files of all mice acquired by the RGB camera and the depth camera, obtain the two-dimensional and three-dimensional phenotypic parameters of multi-target mice throughout their life cycle, including the frequency and duration of various behaviors including eating, drinking, sleeping, standing, fighting and chasing, and analyze the health status of the mice based on the obtained result parameters.

[0079] A method for a high-throughput phenotype acquisition and analysis system for multi-target mice throughout their growth period, the workflow diagram is as follows Figure 4 As shown, the following steps are taken: (1) Connect the mouse cage, RGB camera, two depth cameras, laptop computer, switch, video recorder, and monitor. (2) Cover the bottom of the mouse cage drawer with corn cob bedding. (3) Place three mice on the bedding and put the drawer containing the mice into the mouse cage. (4) Turn on the power, and the laptop computer and video recorder collect image data and video data respectively. (5) The workstation regularly performs digital image processing and analysis on the image data and video data to obtain the two-dimensional and three-dimensional phenotypes of the mice, including activity paths, behavioral states, and the frequency and duration of various behaviors.

[0080] The deep learning neural network YOLOv7 and the improved StrongSORT multi-target tracking algorithm were used to process and analyze the video data collected by the RGB camera to obtain the two-dimensional phenotypic parameter information of mice, which specifically included the following aspects: (1) annotating video images, making a data set, and offline training a mouse recognition model based on YOLOv7; (2) modifying the StrongSORT-based multi-target mouse tracking algorithm, and performing cascade matching and IOU matching on the detection frame generated by YOLOv7 in the current frame and the detection frame, i.e., the trajectory, assigned a specific ID number in the previous frame of video. If there are trajectories that do not match the detection frame and detection frames that do not match the trajectory, the Euclidean distance between them is calculated, and the ID number of the trajectory that does not match the detection frame with the smallest Euclidean distance is assigned to the detection frame that does not match the trajectory. Figure 5 The multi-target tracking results based on the original StrongSORT. The left picture shows the tracking result when the mouse climbs up the wire mesh, and the right picture shows the tracking result when the mouse jumps from the wire mesh to the ground. It can be seen from the figure that the ID number of the mouse is changed from 3 to 4 when it climbs up the wire mesh and jumps to the ground. Figure 6The figure below shows the multi-target tracking result based on the improved StrongSORT. The left picture shows a mouse climbing up the wire mesh, and the right picture shows a mouse jumping onto the ground. As can be seen from the figure, the ID number of the mouse does not change when it climbs up the wire mesh and jumps onto the ground, that is, the tracking ID number of the same mouse is accurate. (3) Based on the YOLOv7 mouse detection results, the key points of each mouse, including the mouth, ears, neck, back center, and tail base, are identified. If there is occlusion, the key points are repaired based on the information and repair technology of the key points of the previous and next frames. (4) The posture of each mouse in each frame of the video image is estimated based on the key point position information. (5) The behavior of each mouse and the social behavior between mice are calculated based on the posture and time series of the mouse in each frame of the video image.

[0081] The RGB images and depth image data collected by two depth cameras are analyzed and processed to obtain the three-dimensional phenotypic parameter information of mice, which specifically includes the following aspects: (1) Align the RGB image with the depth image, reconstruct the three-dimensional point cloud through the depth image, assign the RGB value, i.e., color, to the three-dimensional point cloud through the aligned RGB image, and reconstruct the three-dimensional point cloud respectively with the two depth cameras; (2) The point clouds reconstructed by the two depth cameras are registered through the external parameter calibration between the two depth cameras to form a three-dimensional point cloud; (3) The reconstructed three-dimensional point cloud is divided into two categories, mice and background, and the data set is annotated, and the pointnet++ network is trained to extract the mouse point cloud from the background; (4) The next position of the mouse is predicted based on the Kalman filter, and the Hungarian algorithm is used to match and track the predicted position with the position of the mouse segmented in (3); (5) The three-dimensional point cloud of the mouse is segmented into four parts: head, back, tail, and tail, and the data set is annotated, and the pointnet++ network is trained; (6) The posture of each mouse is estimated based on the position information of the mouse parts; (7) The behavior of each mouse and the social behavior between mice are calculated based on the posture and time series of the mouse in the point cloud.

[0082] The specific embodiments described in this application are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

Claims

1. A high-throughput phenotype acquisition and analysis system for multi-target mice throughout their growth period, characterized in that: It includes a cage system, a rack, a T-bracket system, a depth camera, an RGB camera, a laptop, a switch, a video recorder, a display screen and a workstation, among which: The mouse cage system includes multiple sets of mouse cage systems, each set of mouse cage system includes a mouse cage, a drawer, a water bottle and a food box; the mouse cage adopts a rectangular box design, which is used to place experimental mice, so that the experimental mice can move freely inside the mouse cage; the top surface of the mouse cage is provided with multiple small round holes for ventilation, and rectangular holes are respectively opened on both sides of the top surface; the water bottle and the food box are detachably installed on the side wall of the mouse cage, respectively used to provide water and food for the experimental mice; the drawer is located directly below the mouse cage and is covered with corn cob bedding to prevent mouse excrement from contaminating the mouse cage; The shelf adopts a multi-layer storage rack design, and each layer can place multiple cage systems side by side with equal spacing; each layer of the shelf is equipped with multiple buckles, and each cage system can be detachably installed in a fixed position with the shelf through the buckles; The T-shaped bracket system includes a plurality of T-shaped brackets, which are fixedly mounted on the rack in an inverted T-shaped pattern so that they are located just above each mouse cage system; Depth cameras are installed at the ends of both sides of the lower crossbar of each T-shaped bracket. The depth cameras face the inside of the cage through two rectangular holes on the top surface of the cage to observe the experimental mice. An RGB camera is detachably installed at the center of the top surface of each mouse cage. The RGB camera has infrared night vision function and can continuously observe the experimental mice for 24 hours. The laptop is connected to the depth camera and is used to store the RGB image data and depth image data collected by the depth camera; the video recorder is connected to the RGB camera through a switch and is used to store the video data collected by the RGB camera in real time; The display screen is connected to the video recorder to display and monitor the activities of mice in all cages in real time; The workstation is connected to a laptop and a video recorder to process and analyze all the video and image files of mice acquired by the RGB camera and the depth camera, obtain the two-dimensional and three-dimensional phenotypic parameters of the multi-target mice throughout their life cycle, including the frequency and duration of various behaviors such as eating, drinking, sleeping, standing, fighting and chasing, and analyze the health status of the mice based on the obtained result parameters; A rounded rectangular hole is provided at the center of the top surface of the mouse cage, and a small groove is provided on one side of the rounded rectangular hole for passing the cable; the RGB camera is detachably mounted on a rounded rectangular plate having a shape and size matching the rounded rectangular hole, and the rounded rectangular plate is also provided with a groove corresponding to the rounded rectangular hole for passing the cable; the rounded rectangular plate is designed to be inserted into the rounded rectangular hole, so that the RGB camera is located in the inner space of the mouse cage and faces the mouse for observation, and the cable of the RGB camera is led out through the small groove to connect to an external device; After the observation is completed, the rounded rectangular plate with the RGB camera installed can be removed from the rounded rectangular hole by lifting the cable.

2. The multi-target mouse full-growing period high-throughput phenotype acquisition and analysis system according to claim 1, characterized in that: The rack is made of 304 stainless steel.

3. The multi-target mouse full-growing period high-throughput phenotype acquisition and analysis system according to claim 1, characterized in that: The mouse cage is composed of 6 transparent polyphenylene sulfone resin plates, and has a length, width and height of 265×265×310 mm.

4. The multi-target mouse full-growing period high-throughput phenotype acquisition and analysis system according to claim 1, characterized in that: The two depth cameras above the cage are both tilted toward the interior of the cage, with the tilt angle being 65° from the horizontal direction, and the height difference between the bottom of the depth camera and the top of the cage is 1 cm.

5. A method for acquiring and analyzing multi-target mice throughout their growth period with high-throughput phenotypes, which uses the multi-target mice throughout their growth period with high-throughput phenotypes acquisition and analysis system as described in any one of claims 1 to 4, and specifically comprises the following steps: Step 1, disinfect the rack and mouse cage system; Step 2, multiple cage systems are mounted on the rack by means of buckles, and the two depth cameras on the T-shaped bracket are ensured to be aligned with the interior space of the cage through the rectangular holes on both sides of the top surface of the cage; Step 3: Insert the rounded rectangular plate with the RGB camera into the rounded rectangular hole on the top surface of the cage, making sure that the RGB camera is aligned with the interior space of the cage; Step 4, cover the bottom of the drawer of the mouse cage with corn cob bedding and place the mouse in the mouse cage; Step 5: The system is powered on, the depth camera acquires the RGB image and depth image of the mouse, the RGB camera acquires the video data of the mouse, and the data is stored in real time; Step 6: The staff regularly removes the cage system, removes the mice from the cages, disinfects the cage system, replaces the corncob bedding, reinstalls the cage system and the rounded rectangular board, puts the mice in, and continues to observe; Step 7: The workstation regularly performs digital image processing and analysis on the images and videos of mice, completes the observation of the entire life cycle of multi-target mice, obtains the phenotypic parameter information of the mice, and analyzes the health status of the mice based on the obtained result parameters.

6. The method for obtaining and analyzing multi-target mouse phenotypes during the entire growth period according to claim 5, characterized in that: In step 7, the deep learning neural network YOLOv7 and the improved StrongSORT multi-target tracking algorithm are used to process and analyze the video data collected by the RGB camera to obtain the two-dimensional phenotypic parameter information of the mice. The specific method is as follows: (1) Label video images, create datasets, and train a mouse recognition model based on YOLOv7 offline; (2) Mouse video data is input, and the YOLOv7 mouse recognition model is used to identify multiple target mice in each frame of the video and generate target detection frames. The multi-target tracking algorithm based on the improved StrongSORT is used to assign a specific ID number to each target detection frame and track it; (3) Based on the YOLOv7 mouse detection results, identify multiple key points of each mouse, including the mouth, ears, neck, back center, and tail base; If there is occlusion, the key points are repaired based on the key point information and repair technology of the previous and next frames; (4) Estimate the posture of each mouse in each frame of video based on the key point position information; (5) Calculate the behavior of each mouse and the social behavior between mice based on the posture and time series of the mice in each frame of the video image.

7. The method for obtaining and analyzing multi-target mouse phenotypes during the entire growth period according to claim 6, characterized in that: In step 7, the specific improvements of the improved StrongSORT multi-target mouse tracking algorithm are as follows: in the current frame video, YOLOv7 is used to identify the mouse and generate a detection frame, and the generated detection frame is cascade matched and IOU matched with the detection frame, i.e., the trajectory, assigned a specific ID number in the previous frame video. If there are trajectories that do not match the detection frame and detection frames that do not match the trajectory, the Euclidean distance between them is calculated, and the ID number of the trajectory that does not match the detection frame with the smallest Euclidean distance is assigned to the detection frame that does not match the trajectory.

8. The method for obtaining and analyzing multi-target mouse phenotypes during the entire growth period according to claim 7, characterized in that: In step 7, the RGB images and depth image data collected by the two depth cameras are analyzed and processed to obtain the three-dimensional phenotypic parameter information of the mouse, and the specific method is as follows: (1) Align the RGB image with the depth image, reconstruct the 3D point cloud through the depth image, assign RGB values, i.e., colors, to the 3D point cloud through the aligned RGB image, and reconstruct the 3D point cloud with two depth cameras respectively; (2) The point clouds reconstructed by the two depth cameras are registered through the external parameter calibration between the two depth cameras and synthesized into a three-dimensional point cloud; (3) The reconstructed 3D point cloud is divided into two categories: mice and background, the data set is annotated, and the pointnet++ network is trained to extract the mouse point cloud from the background; (4) Predict the next position of the mouse based on Kalman filtering, and use the Hungarian algorithm to match the predicted position with the position of the mouse segmented in (3) and track it; (5) Segment the mouse 3D point cloud into four parts: head, back, tail, and tail, annotate the data set, and train the pointnet++ network; (6) estimating the posture of each mouse based on the position information of the mouse parts; (7) Calculate the behavior of each mouse and the social behavior between mice based on the posture and time series of the mice in the point cloud.

9. The method for obtaining and analyzing multi-target mouse phenotypes during the entire growth period according to claim 5, characterized in that: In step 7, the phenotypic parameter information of the mice is obtained, including the frequency and duration of the behaviors of eating, drinking, sleeping, standing, fighting and chasing, and the health status of the mice is analyzed based on the obtained result parameters, specifically using the following detailed steps: (1) Identify and detect various behaviors of mice: Eating: if the mouse’s mouth was in the food box for more than 2 seconds, it was considered eating, and the time the mouse’s mouth was in the food box was recorded as the duration; Drinking water: if the mouse's mouth is at the mouth of the water bottle for more than 2 seconds, it is judged as drinking water, and the time the mouse's mouth is at the mouth of the water bottle is recorded as the duration; Sleeping: calculate the change in the position of key parts of the mouse, including the head, back and tail, in adjacent frames. If the change is less than 5 pixels, the mouse is considered to be stationary. If the stationary time is greater than 100 seconds, the mouse is considered to be sleeping, and the duration is recorded. Standing, let the position of the mouse's mouth in the image be A, the position of the center of the back be B, the position of the base of the tail be C, the distance between the mouth and the center of the back be a, and the distance between the center of the back and the base of the tail be b. If 0.9 fights, the relative movement speed of the two mice is greater than 10 cm / s, and they touch each other in the next frame, it is judged as a fight, and the duration is recorded; Chase: two mice move in phase, the mouth of one mouse is behind the tail of the other mouse, and the speed of both mice is greater than 5 cm / s, and the relative speed of the two mice is less than 2 cm / s. This is considered a chase, and the duration is recorded; (2) Count the number of times each behavior occurs in 24 hours a day, which is the frequency. Add up the duration of each behavior and record the frequency and duration of each behavior. (3) A health index is established based on various behaviors, their frequency of occurrence, and duration, and the health status of mice is analyzed based on the health index. ​

Images