A method for automatically identifying the movement state of a rat in a rat horizontal bridge experiment and application thereof

Abstract

The application discloses a rat motion state automatic identification method in a rat horizontal bridge experiment and application, and belongs to the technical field of experimental automation analysis; the application collects a rat motion video, extracts a target to obtain a target motion image sequence containing only a rat target, then adopts a pre-trained ResNet-LSTM model to capture spatial features and time sequence features of the target motion image sequence, so as to identify the motion state of the rat in each frame; compared with an existing professional motion capture system (such as VICON), the application does not need to perform additional marking on a rat body in a video shooting process to assist shooting and subsequent motion state identification, simplifies experimental operation, reduces the influence of unnecessary human factors on experimental results, and does not need other professional motion capture equipment except a camera, so that the rat motion state in the rat horizontal bridge experiment can be automatically and accurately identified at a low cost.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of experimental automation analysis, and more particularly relates to a rat motion state automatic recognition method in a rat horizontal bridge experiment and application. BACKGROUND

[0002] The horizontal bridge experiment is a common method for evaluating the motion state of a rat proposed by Gerlinde A. Metz and Ian Q. Whishaw. The horizontal bridge is composed of multiple levels of adjustable horizontal ladders and transparent baffles on both sides, and the rat is restricted to pass through the horizontal bridge in one direction. During the experiment, the rat walks through the horizontal ladder with its limbs, and a camera is used to record the motion process of the rat. Then, the experimenter analyzes the video and counts the gait, angle of the paw, and the situation of the limb lifting and falling to evaluate the motion state of the rat.

[0003] At present, in the rat horizontal bridge experiment, researchers need to shoot a large number of videos of rats moving on the horizontal bridge, and then manually analyze and count the motion state of the rats in the video. This process is time-consuming and laborious. The existing motion capture system such as VICON is expensive, and it also needs to paste markers (such as color points, infrared balls, etc.) on the captured object to assist in shooting and subsequent recognition. It can only capture the motion trajectory of the human body and large animal limbs, and is not suitable for small animals such as rats. Moreover, the rat horizontal bridge experiment mainly focuses on the motion state of the rat lifting the hind legs and dropping the hind legs on the horizontal bridge, and the existing motion capture system cannot judge the motion state of the rat on the horizontal bridge based on the captured motion trajectory of the target object, so it cannot be applied to the rat horizontal bridge experiment. SUMMARY

[0004] In view of the above defects or improvement needs of the prior art, the present application provides a rat motion state automatic recognition method in a rat horizontal bridge experiment and application, to solve the technical problem that the prior art cannot automatically and accurately recognize the motion state of the rat in the rat horizontal bridge experiment at a low cost.

[0005] To achieve the above purpose, in a first aspect, the present application provides a rat motion state automatic recognition method in a rat horizontal bridge experiment, comprising the following steps:

[0006] S1, sequentially performing target detection on each frame of image in the rat motion video, extracting the rat target region in the image to obtain a target image, and performing preprocessing; sorting the target images after preprocessing according to their time sequence order in the rat motion video to obtain a target motion image sequence;

[0007] S2, inputting the target motion image sequence into the pre-trained motion state recognition model to obtain a motion state recognition result of the rat in each frame image of the target motion image sequence;

[0008] The rat motion video is a motion video of a single rat collected during a horizontal bridge experiment on the rat, the motion state recognition model is a ResNet-LSTM model, and the motion state of the rat includes lifting the hind legs, walking, and falling of the hind legs.

[0009] Further preferably, the frame images in the target motion image sequence are input into the pre-trained motion state recognition model in a sliding window manner to obtain the motion state recognition result of the rat in each frame image of the target motion image sequence, and for the image frames identified multiple times by the motion state recognition model, the final recognition result is the recognition result with the highest quantity proportion in the multiple recognition results.

[0010] Further preferably, the motion state recognition model includes a cascaded ResNet network and an LSTM network.

[0011] The ResNet network is used to sequentially perform spatial feature extraction on each input frame image and input into the LSTM network.

[0012] The LSTM network is used to further extract the time sequence features of the frame image on the basis of the spatial features of the frame image, to obtain the spatio-temporal features of each frame image, and then obtain the motion state recognition result of the rat in each frame image.

[0013] Further preferably, the method for pre-processing the target image includes adjusting the size of the target image to have a consistent aspect ratio, and normalizing the adjusted target image to make the pixel values conform to a normal distribution.

[0014] Further preferably, in the step S1, a YOLO4 model is used to sequentially perform target detection on each frame image in the motion video.

[0015] In a second aspect, the present application provides a method for evaluating the motion ability of a rat in a rat horizontal bridge experiment, including:

[0016] The recognition results of the frame images in the rat motion video are counted to obtain the total number of steps and the number of times of falling of the hind legs of the rat in the rat motion video, and the ratio of the total number of steps to the number of times of falling of the hind legs of the rat in the rat motion video is calculated to evaluate the motion ability of the rat; the larger the ratio, the stronger the motion ability of the rat;

[0017] The identification result of each frame image in the rat movement video is obtained by using the rat movement state automatic identification method provided in the first aspect of the present application; and the rat movement video is a movement video of a rat collected during a horizontal bridge experiment on a single rat.

[0018] Further preferably, the method for counting the identification results of each frame image in the rat movement video comprises:

[0019] If the rat movement video has continuous M frames of images, which are composed of m1 frames of images identified as walking, m2 frames of images identified as hind leg lifting, and m3 frames of images identified as walking, then it is determined that the rat has moved one step, and the total number of steps of the rat is increased by one;

[0020] If the identification results of continuous N frames of images in the rat movement video are all hind leg falling, then it is determined that the rat has fallen once, and the number of times of hind leg falling of the rat is increased by one;

[0021] Wherein, m1+m2+m3=M; m1, m2, m3, N are all greater than or equal to 3.

[0022] In a third aspect, the present application provides a rat movement state automatic identification system in a rat horizontal bridge experiment, comprising: a memory and a processor, the memory stores a computer program, and the processor executes the computer program to execute the rat movement state automatic identification method provided in the first aspect of the present application.

[0023] In a fourth aspect, the present application provides a rat movement state automatic identification system in a rat horizontal bridge experiment, comprising: a camera, a horizontal bridge experiment device and an identification device;

[0024] The camera is located on one side of the horizontal bridge experiment device, and is used to collect a movement video in a rat horizontal bridge experiment in real time and output to the identification device;

[0025] The identification device is used to execute the rat movement state automatic identification method provided in the first aspect of the present application.

[0026] Further preferably, the horizontal bridge experiment device comprises a plurality of horizontal ladders suspended in the air and with adjustable spacing, and transparent baffles located on both sides of the horizontal ladders;

[0027] The center of the camera picture is consistent with the center of the horizontal ladder to ensure that the horizontal ladder is fully displayed in the camera picture and remains horizontal.

[0028] In a fifth aspect, the present application further provides a computer readable storage medium comprising a stored computer program, wherein the computer program, when executed by a processor, controls a device in which the storage medium is located to perform the rat motion state automatic identification method provided in the first aspect of the present application and / or the rat motion ability evaluation method in a rat horizontal bridge experiment provided in the second aspect of the present application.

[0029] Overall, the above technical solutions conceived by the present application can achieve the following beneficial effects:

[0030] 1. The present application provides a rat motion state automatic identification method in a rat horizontal bridge experiment, which comprises the following steps: collecting a rat motion video, performing target extraction on the rat motion video to obtain a target motion image sequence containing only a rat target, and then using a pre-trained ResNet-LSTM model to capture the spatial features and time sequence features of the target motion image sequence, so as to identify the motion state of the rat in each frame. Compared with existing professional motion capture systems (such as VICON), the present application does not need to perform additional marking on the rat's body during video shooting to assist shooting and subsequent motion state identification, which simplifies the experimental operation and also reduces the influence of unnecessary human factors on the experimental results. In addition to the camera, no other professional motion capture equipment is needed, and the rat motion state in the rat horizontal bridge experiment can be automatically and accurately identified at a lower cost.

[0031] 2. Compared with other regression methods for identifying rat limb key points, the rat motion state automatic identification method in the rat horizontal bridge experiment provided by the present application first proposes a classification model based on deep learning, and first applies it to the rat horizontal bridge experiment. The model has better robustness, and the prediction result of the model can be directly used to calculate motion parameters such as the number of motion steps, the number of falls and the comprehensive motion ability score of the rat on the horizontal bridge.

[0032] 3. The present application provides a rat motion state automatic identification system in a rat horizontal bridge experiment, which can automatically identify the motion state of the rat during the horizontal bridge experiment by only shooting a video with a camera and combining the rat motion state automatic identification method provided in the first aspect of the present application, greatly reducing the work burden of researchers. In addition to the camera, the present application does not need to rely on other motion capture equipment, which reduces the experimental cost and also reduces the influence of human factors on the experimental results without the need for special marking and other processing of the rat. The present application also has the ability to analyze whether the rat's motion function is normal and the degree of impairment from the results of video detection of the rat's motion state (for example, judging the motion function recovery after spinal cord injury). BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1A flow chart of a rat motion state automatic identification method in a rat horizontal bridge experiment provided for the embodiment 1 of the present application;

[0034] Figure 2 A process schematic diagram of recognizing the motion state of a rat in each frame of a target motion image sequence by using a ResNet-LSTM model provided for the embodiment 1 of the present application;

[0035] Figure 3 A whole process schematic diagram of recognizing and evaluating the motion state of a rat in a rat horizontal bridge experiment provided for the embodiment 2 of the present application. DETAILED DESCRIPTION

[0036] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0037] Embodiment 1,

[0038] A rat motion state automatic identification method in a rat horizontal bridge experiment, as shown in FIG. 1, comprises the following steps: Figure 1

[0039] S1, sequentially detecting each frame of image in the rat motion video, extracting the rat target area in the image to obtain a target image, and pre-processing; sorting the pre-processed target image according to its time sequence order in the rat motion video to obtain a target motion image sequence;

[0040] The rat motion video is a motion video of a single rat collected when the rat is subjected to a horizontal bridge experiment; the horizontal bridge is composed of multiple levels of adjustable horizontal ladders and transparent baffles on both sides; the length of the horizontal bridge is about 100 cm, the width of the horizontal ladder is about 10 cm, and the height of the transparent baffle is about 20 cm. When the rat moves on the horizontal bridge, the camera is placed on one side of the rat to take pictures. The average duration of all collected videos is about 20 s, the average frame number is about 1200, the resolution is 3840*2160, and a total of 356 videos are collected (it should be noted that the parameters are not limited to these parameters, and other parameters can also be used, which are only examples here). The rat is limited to pass through the experimental device in one direction, and rats with different degrees of spinal cord injury will exhibit different motion states on the horizontal bridge.

[0041] ​Specifically, since the original image contains a large amount of background information, the proportion of the rat in the original image is small, and the background area with too large proportion will interfere with the subsequent ResNet-LSTM model, therefore, the image is first subjected to target detection. There are various methods for target detection of the image, such as YOLOv2, YOLOv3, YOLOv4 model and the like. In the embodiment, the trained YOLOv4 model is used to sequentially perform target detection on each frame of image in the rat movement video. The input of the YOLOv4 model is a single frame of image in the video, the image content is the entire horizontal bridge movement scene and a single rat moving on the horizontal bridge, and the output of the model is an image containing only the rat. YOLOv4 is a single-stage target detection algorithm, which has the characteristics of high precision and high real-time. Specifically: first, the training set is created by using yolo_mark software, and the YOLOv4 model is trained, and then the image to be detected is input into the trained YOLOv4 model. The output of the model is the image in the detection box, the coordinates of the detection box and the confidence of each detected target. The confidence threshold is set to 0.75, and the detection results below the threshold will not be output. The model may output other misrecognized detection results, such as human hands, other rats in the rat incubator, and the naming of the target rat detection image is generally dest0, and other misrecognized targets are named as dest1-dest9. Only the image with the name dest0 in the folder is retained, that is, the detection image containing only the target rat is obtained.

[0042] Since the rat only occupies a small part of the space in the input image, and occupies a large part of the space in the output image, the target image also needs to be preprocessed. In the embodiment, the method for pre-processing the target image comprises: adjusting the size of the target image to have a consistent aspect ratio, and standardizing the adjusted target image to make the pixel value conform to the normal distribution. Specifically, the image is pre-processed as follows: 1) the picture size is adjusted. The image size output by the YOLOv4 model is not uniform, which does not meet the input requirements of the subsequent ResNet-LSTM model, therefore, the transforms.Resize function in pytorch is used to change it to 448x448; 2) the picture is converted to a vector form by using the transforms.ToTensor function in pytorch, and the image is standardized. The standardization operation is as follows: wherein, I(x, y) out represents the pixel value of the processed image at the coordinate point (x, y), I(x, y) in represents the pixel value of the image before processing at the coordinate point (x, y), I mean represents the pixel value mean of the image before processing, I std represents the pixel value standard deviation of the image before processing.

[0043] S2, inputting the target motion image sequence into the pre-trained motion state recognition model to obtain a motion state recognition result of the rat in each frame image of the target motion image sequence; wherein the motion state of the rat includes: hind leg lifting, walking and hind leg falling.

[0044] Preferably, the frame images in the target motion image sequence are input into the pre-trained motion state recognition model in a sliding window manner to obtain the motion state recognition result of the rat in each frame image of the target motion image sequence; wherein for the image frames identified multiple times by the motion state recognition model, the final recognition result is the recognition result with the highest quantity proportion in the multiple recognition results.

[0045] In this embodiment, the motion state recognition model is trained using a labeled rat image dataset; specifically, a plurality of consecutive rat images in the rat motion video processed by step S1 are taken as input, and the motion state corresponding to each rat image (the motion state of the rat includes: hind leg lifting, walking and hind leg falling) is taken as the expected output, and the motion state recognition model is trained until convergence. The motion state recognition model used in this embodiment is a ResNet-LSTM model, and cross-entropy is used as the loss function during training. Specifically, the above motion state recognition model includes a cascaded ResNet network and an LSTM network; wherein the ResNet network is used to sequentially perform spatial feature extraction on each input frame image and input into the LSTM network; the LSTM network is used to further extract the time sequence features of the frame image on the basis of the spatial features of the frame image, to obtain the spatio-temporal features of each frame image, and then to obtain the motion state recognition result of the rat in each frame image.

[0046] The present application integrates the traditional ResNet network and the LSTM network, develops a new motion state recognition model based on deep learning in the rat horizontal bridge experiment, called ResNet-LSTM, as shown in Figure 2The input of the ResNet-LSTM model is 10 consecutive images (not limited to 10, other parameters can also be used, which is only an example here) in a sliding window form, and the output of the model is the classification value of the rat motion state in each image, including three types: hind leg lifting, walking, and hind leg falling. The network structure mainly includes a ResNet50 module and a bidirectional LSTM module. The ResNet50 module includes 49 convolutional layers and one fully connected layer. The spatial features of each single image are extracted in sequence through the ResNet50 module, and then the information extracted from the single image is input into the bidirectional LSTM module in sequence. The information extracted from the previous single image is also input into the bidirectional LSTM module, so that the information extracted by the bidirectional LSTM module at each time is transmitted in sequence, thereby using the LSTM to extract the temporal features between different image frames to better identify the different motion states of the rat in the horizontal bridge experiment. Further, the cross-entropy loss is used as the loss function in the training process. Since each image in the video (except the first 9 images) is detected 10 times, the motion state of the rat in each image is determined by the voting results of the 10 detections, and the motion state with the highest number of votes is determined as the final motion state.

[0047] To further verify the recognition effect of the rat motion state automatic recognition method provided by the present application in the rat horizontal bridge experiment, in an experimental example, the data set used includes 356 rat horizontal bridge motion videos with a duration of about 20s. Among them, 200 videos are randomly selected as training data, and the remaining 156 videos are used as test data. A sliding window with a length of 10 frames and a step of 1 frame is used to extract image samples from the starting time of each video. After each sample extraction is completed, the sliding window is moved back by one frame for the next extraction. When the remaining number of image sequences is less than 10, the sampling is completed. Each sample extracted includes 10 images, and each image corresponds to a motion state label (hind leg lifting or walking or hind leg falling).

[0048] Then, the image samples extracted in the training data are taken as inputs, and the labels of the corresponding images are taken as outputs to train the ResNet-LSTM model, so as to obtain a trained rat motion state recognition model. Specifically, the model is trained for 50 rounds, the training batch size is 8, the Adam optimizer is used, and the initial learning rate is set to 0.001 and is adjusted according to the validation set accuracy.

[0049] The trained ResNet-LSTM model is tested on 100 test video data. Each frame of image in the video (except for the first 9 frames of image) is detected for 10 times, and the label of each frame of image is determined by comprehensive voting of the 10 detection results, and the label with the highest number of votes is output. On the single frame image level, the recall rate of the "leg lifting" image recognition is 0.974, and the specificity is 0.984; the recall rate of the "walking" image recognition is 0.983, and the specificity is 0.977; the recall rate of the "leg falling" image recognition is 0.982, and the specificity is 0.998. The results show that the rat motion state recognition model can achieve very high recognition accuracy.

[0050] Embodiment 2,

[0051] A method for evaluating the motion ability of a rat in a rat horizontal bridge experiment, comprising:

[0052] The recognition results of each frame of image in the rat motion video are counted to obtain the total number of steps and the number of times of leg falling of the rat in the rat motion video, and the ratio of the total number of steps to the number of times of leg falling of the rat in the rat motion video is calculated to evaluate the motion ability of the rat; the larger the ratio, the stronger the motion ability of the rat;

[0053] The recognition results of each frame of image in the rat motion video are counted to obtain the total number of steps and the number of times of leg falling of the rat in the rat motion video, and the ratio of the total number of steps to the number of times of leg falling of the rat in the rat motion video is calculated to evaluate the motion ability of the rat; the larger the ratio, the stronger the motion ability of the rat;

[0054] Specifically, the method for counting the recognition results of each frame of image in the rat motion video comprises:

[0055] If the continuous M frames of images in the rat motion video are composed of m1 frames of images recognized as walking images, m2 frames of images recognized as leg lifting images, and m3 frames of images recognized as walking images, it is determined that the rat has moved one step, and the total number of steps of the rat is increased by one;

[0056] If the recognition results of the continuous N frames of images in the rat motion video are all leg falling, it is determined that the rat has fallen once, and the number of times of leg falling of the rat is increased by one;

[0057] Wherein, m1+m2+m3=M; m1, m2, m3, N are all greater than or equal to 3.

[0058] Specifically, the whole process from recognition to evaluation of the motion state of the rat in the rat horizontal bridge experiment is as shown in the figure. Figure 3 To further illustrate the performance of the method for evaluating the motion ability of the rat in the rat horizontal bridge experiment provided by the present application, the correlation coefficient between the calculated motion ability scores of the rat in the 156 videos and the pre-labeled motion ability scores of the rat is calculated, and the correlation coefficient is 0.942, indicating that the recognition result of the automatic rat motion state recognition method provided by the present application has approached the recognition level of experts, and can automatically and accurately evaluate the motion ability of the rat in the horizontal bridge motion.

[0059] The related technical solutions are the same as those in Embodiment 1, which will not be repeated here.

[0060] Embodiment 3,

[0061] A system for automatically recognizing the motion state of a rat in a rat horizontal bridge experiment, comprising: a memory and a processor, the memory storing a computer program, and the processor executing the computer program to execute the automatic rat motion state recognition method provided in Embodiment 1 of the present application.

[0062] The related technical solutions are the same as those in Embodiment 1, which will not be repeated here.

[0063] Embodiment 4,

[0064] A system for automatically recognizing the motion state of a rat in a rat horizontal bridge experiment, comprising: a camera, a horizontal bridge experiment device and a recognition device; wherein the horizontal bridge experiment device comprises a plurality of horizontal ladders suspended in the air with adjustable spacing, and transparent baffles located on both sides of the horizontal ladders; the center of the camera picture is consistent with the center of the horizontal ladder to ensure that the horizontal ladder appears entirely in the camera picture and remains horizontal; the rat is restricted to pass through the experiment device in one direction, and rats with different degrees of spinal cord injury will exhibit different motion states on the horizontal bridge.

[0065] Specifically, when the rat moves on the horizontal bridge, the camera is located on one side of the horizontal bridge experiment device, which is used to collect the motion video of the rat in the horizontal bridge experiment in real time and output to the recognition device; in one embodiment, the horizontal bridge is about 100 cm long, the width of the horizontal ladder is about 10 cm, the height of the transparent baffle is about 20 cm, the average duration of all collected videos is about 20 s, the average frame number is about 1200, the resolution is 3840*2160, and a total of 356 videos are collected.

[0066] The recognition device is used to execute the automatic rat motion state recognition method provided in Embodiment 1 of the present application.

[0067] The application can automatically identify the motion state of a rat during a horizontal bridge experiment only by shooting a video through a camera and combining a developed deep learning algorithm, greatly reducing the work burden of researchers; the application does not need to rely on other motion capture devices except the camera, reduces the experimental cost, and does not need to perform special marking and other processing on the rat, reducing the influence of human factors on the experimental results; the application also has the ability to analyze whether the motion function of the rat is normal and the degree of damage from the video detection result of the motion state of the rat (for example, judging the recovery of the motion function after spinal cord injury).

[0068] The related technical solutions are the same as those in Embodiment 1, and will not be repeated here.

[0069] Embodiment 5,

[0070] A computer readable storage medium, comprising a stored computer program, wherein the computer program, when executed by a processor, controls a device where the storage medium is located to perform the rat motion state automatic identification method provided in Embodiment 1 of the application and / or the rat motion ability evaluation method in a rat horizontal bridge experiment provided in Embodiment 2 of the application.

[0071] The related technical solutions are the same as those in Embodiments 1 and 2, and will not be repeated here.

[0072] Those skilled in the art can easily understand that the above description is only the preferred embodiment of the application, and is not intended to limit the application. Any modification, equivalent replacement and improvement made within the spirit and principle of the application shall be included in the protection scope of the application.

Claims

1. A method for automatically identifying the movement state of a rat in a rat horizontal bridge experiment, characterized in that, The method comprises the following steps: S1. Target detection is performed on each frame of image in the rat movement video in sequence, the rat target region in the image is extracted, a target image is obtained, and pre-processing is performed; the target image after pre-processing is sorted according to the time sequence order thereof in the rat movement video, and a target movement image sequence is obtained; S2. The target movement image sequence is input into a pre-trained movement state recognition model, and a movement state recognition result of the rat in each frame of image in the target movement image sequence is obtained; S3. The recognition results of each frame of image in the rat movement video are counted, the total number of steps and the number of times of falling of the rat in the rat movement video are obtained, and the ratio of the total number of steps to the number of times of falling of the rat in the rat movement video is calculated, so as to evaluate the movement ability of the rat; the larger the ratio is, the stronger the movement ability of the rat is; The rat movement video is a movement video of a single rat collected when the rat is subjected to a horizontal bridge experiment; the movement state recognition model is a ResNet-LSTM model; and the movement state of the rat comprises the following: lifting of the hind legs, walking, and falling of the hind legs. The method for counting the recognition results of each frame of image in the rat movement video comprises the following steps: If there are M consecutive frames of image in the rat movement video, which are composed of m1 frames of image with the recognition result of walking, m2 frames of image with the recognition result of lifting of the hind legs, and m3 frames of image with the recognition result of walking, then it is determined that the rat has moved one step, and the total number of steps of the rat is increased by one; If the recognition results of N consecutive frames of image in the rat movement video are all falling of the hind legs, then it is determined that the rat has fallen once, and the number of times of falling of the rat is increased by one; Wherein, m1+m2+m3=M; m1, m2, m3, and N are all greater than or equal to 3.

2. The method of claim 1, wherein the method is characterized by, The frame of image in the target movement image sequence is input into the pre-trained movement state recognition model in a sliding window manner, and the movement state recognition result of the rat in each frame of image in the target movement image sequence is obtained; wherein, for the image frame which is recognized multiple times by the movement state recognition model, the final recognition result thereof is the recognition result with the highest quantity ratio in the multiple recognition results thereof.

3. The method of claim 1, wherein the method is characterized by, The movement state recognition model comprises a cascaded ResNet network and an LSTM network; The ResNet network is used for spatial feature extraction of each input frame of image in sequence, and input into the LSTM network; The LSTM network is used for further extraction of the time sequence feature of the frame of image on the basis of the spatial feature of the frame of image, so as to obtain the spatio-temporal feature of each frame of image, and further obtain the movement state recognition result of the rat in each frame of image.

4. The method according to any one of claims 1 to 3, wherein The method for pre-processing the target image comprises the following steps: adjusting the size of the target image so that it has a consistent aspect ratio, and normalizing the target image after adjustment so that the pixel value thereof conforms to a normal distribution.

5. A system for automatically identifying the state of motion of a rat in a rat level crossing experiment, characterized in that, The method comprises the following steps: The memory stores a computer program, and the processor executes the computer program to execute the rat movement state automatic recognition method in any one of claims 1-4.

6. A system for automatically identifying the state of motion of a rat in a rat level crossing experiment, characterized in that, ​ A camera, a horizontal bridge experimental device and a recognition device; The camera is located on one side of the horizontal bridge experimental device, and is used for collecting motion video in real time during a rat horizontal bridge experiment, and outputting to the recognition device; The recognition device is used for executing the rat motion state automatic recognition method in any one of claims 1-4.

7. The automatic rat locomotor state recognition system according to claim 6, wherein The horizontal bridge experimental device comprises a plurality of horizontal ladders suspended in the air and having adjustable spacing, and transparent baffles located on both sides of the horizontal ladders. The center of the camera picture is consistent with the center of the horizontal ladder, so as to ensure that the horizontal ladder is entirely present in the camera picture and remains horizontal.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium comprises a stored computer program, wherein the computer program controls a device where the storage medium is located to execute the rat motion state automatic recognition method in any one of claims 1-4 when the computer program is run by a processor.

Images