Multi-modal animal abnormal data behavior classification method and device

Abstract

The invention discloses a multi-modal animal abnormal data behavior classification method and device which are applied to the technical field of data processing, and the method comprises the steps: obtaining abnormal data of a target animal in different development stages; the abnormal data comprises behavior video data and intestinal flora data of the same animal; performing feature extraction on the behavior video data to obtain behavior features representing animal action modes; analyzing the intestinal flora data to obtain intestinal flora characteristics; performing multi-modal data fusion on the behavior characteristics and the intestinal flora characteristics to generate fusion characteristics; inputting the fusion features into a pre-trained data model, classifying animal behaviors through the data model, and outputting a classification result; through the behavior-flora multi-modal fusion strategy, the classification accuracy is improved, and clues are provided for early intervention.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a multi-modal animal abnormal data behavior classification method and device. Background Art

[0002] Autism is a relatively common mental illness. It is a subtype of pervasive developmental disorder. The symptoms of the disease are mainly manifested in varying degrees of repetitive and stereotyped behaviors, social disorders, gastrointestinal problems, and sensory abnormalities. Most of the research subjects for autism are autism model animals. Among them, rodent mice are the traditional choice for studying the neural mechanisms of autism and testing treatment methods because of their high throughput, low cost, and multiple research paradigms. Autism model mice often show spontaneous repetitive activities, including spontaneous grooming, rotation, and jumping. Among them, repeated self-grooming is similar to the repetitive stereotyped behaviors of autistic patients and is considered to be one of the key factors in evaluating autism model mice. However, the current classification methods for autism model animals mainly rely on a single data type, and there is a problem of insufficient classification accuracy due to a single data source. At the same time, behavioral data are easily disturbed by the environment, microbiome data lack dynamic associations, and early detection capabilities are weak, and it is impossible to dynamically capture the key window period of disease development through multimodal features.

[0003] Therefore, in order to overcome these defects, the present application proposes a multimodal animal abnormal data behavior classification method and device. Summary of the invention

[0004] The purpose of this application is to provide a multimodal animal abnormal data behavior classification method and device, aiming to solve the above-mentioned problems.

[0005] To achieve the above objectives, this application provides the following technical solutions: In a first aspect, the present application provides a multimodal animal abnormal data behavior classification method, the steps comprising: Acquire abnormal data of target animals at different developmental stages; the abnormal data includes behavioral video data and intestinal flora data of the same animal; Extracting features from the behavior video data to obtain behavior features that characterize the animal's action patterns; analyzing the intestinal flora data to obtain intestinal flora features; Performing multimodal data fusion on the behavioral features and the intestinal flora features to generate fusion features; The fusion feature is input into a pre-trained data model, the animal's behavior is classified through the data model, and the classification result is output; wherein the data model is obtained by training based on a convolutional neural network.

[0006] In a second aspect, the present application provides a multimodal animal abnormal data behavior classification device, comprising: Data acquisition module: acquires abnormal data of target animals at different developmental stages; the abnormal data includes behavioral video data and intestinal flora data of the same animal; Feature extraction module: extract features from the behavior video data to obtain behavior features that characterize the animal's action patterns; analyze the intestinal flora data to obtain intestinal flora features; Feature fusion module: multi-modal data fusion of the behavior feature and the intestinal flora feature to generate fusion features; Classification module: input the fusion features into a pre-trained data model, classify the animal's behavior through the data model, and output the classification result; wherein the data model is trained based on a convolutional neural network.

[0007] In a third aspect, the present application provides a computer device, comprising a processor and a memory coupled to the processor, wherein the memory stores program instructions for implementing a multimodal animal abnormal data behavior classification method; and the processor is used to execute the program instructions stored in the memory to implement a multimodal animal abnormal data behavior classification method.

[0008] In a fourth aspect, the present application provides a storage medium storing program instructions executable by a processor, wherein the program instructions are used to execute a multimodal animal abnormal data behavior classification method.

[0009] The present application provides a multimodal animal abnormal data behavior classification method and device, which has the following beneficial effects: (1) By combining behavioral video data and intestinal flora data, a fusion analysis of multimodal data is achieved. Behavioral video data provides intuitive information about animal behavior, while intestinal flora data reflects changes in the animal's physiological state. The combination of the two can more comprehensively describe the animal's state, significantly improving the accuracy and reliability of classification compared to traditional classification methods that rely on a single data type.

[0010] (2) A deep learning model is used to track the key points of the animal's body and generate posture time series data. Action labels are then divided and behavioral parameters are calculated based on this data, reducing the impact of subjective judgment on the results. Specifically, the behavioral ratio of sniffing and grooming actions is calculated and its logarithm is taken as the key indicator for classification, which reflects the behavioral characteristics of autism model animals and helps to identify autism-like phenotypes at an early stage.

[0011] (3) This method can be applied to a variety of experimental animal models and can serve as an important clue for early disease prediction and intervention. It can also provide a strong basis for early drug screening and intervention in scientific research. By identifying autism-like phenotypes at an early stage, researchers can more accurately evaluate the effects of potential therapeutic drugs, thereby accelerating the drug development process. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A flowchart of a multimodal animal abnormal data behavior classification method according to Example 1 of the present application; Figure 2 This is a schematic diagram of the behavioral ratio of the duration of the sniffing action and the grooming action in Example 1 of the present application; Figure 3 This is a schematic diagram of using linear discriminant analysis (LDA) to verify the behavior ratio classification in Example 1 of the present application; Figure 4 This is a schematic diagram of the results of Bray-Curtis similarity index PCOA analysis and Adonis test of the intestinal flora data of 3 to 6-week-old male mice in Example 1 of the present application; Figure 5 This is a schematic diagram of the results of Bray-Curtis similarity index PCOA analysis and Adonis test of the intestinal flora data of 7-, 8-, and 12-week-old male mice in Example 1 of the present application; Figure 6 This is a schematic diagram of the results of Bray-Curtis similarity index PCOA analysis and Adonis test of the intestinal flora data of female mice aged 3 to 6 weeks in Example 1 of the present application; Figure 7 This is a schematic diagram of the results of Bray-Curtis similarity index PCOA analysis and Adonis test of the intestinal flora data of 7-, 8-, and 12-week-old female mice in Example 1 of the present application; Figure 8 This is a structural schematic diagram of a multimodal animal abnormal data behavior classification device according to Example 2 of the present application; Fig. 9 This is a schematic diagram of the computer device structure of Example 3 of the present application; Fig.10 This is a schematic diagram of the storage medium structure of Example 4 of the present application. DETAILED DESCRIPTION

[0013] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0014] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0015] Example 1 See also Figure 1 , is a flow chart of a multimodal animal abnormal data behavior classification method according to Example 1 of the present application; the steps include: S1: Acquire abnormal data of target animals at different developmental stages; the abnormal data include behavioral video data and intestinal flora data of the same animal.

[0016] In this embodiment, the target animal is an autism model animal, including but not limited to a rodent; the abnormal behavioral phenotype of the target animal is an autism-like phenotype, manifested as at least one of repetitive stereotyped behaviors, social disorders, or abnormal intestinal flora.

[0017] Among them, the behavioral video data can be obtained by shooting the spontaneous behavior of the same species from multiple angles through multiple high-definition resolution cameras or high-definition resolution cameras with infrared shooting mode. At this time, different animals have different emotions or behaviors. By changing the collection object, the collection of different states can be achieved. The duration of the collected video data can be set as needed. For example, the data collection time of a single animal is set between 15min-60min, and multiple angles can be collected. During the actual data collection, the viewing angles can also be increased or deleted as needed. For different animals, based on morphological analysis, when each species faces different emotions or different disease states, due to differences in physiological structure, there are subtle differences in the characteristics of the perceived external environment, and its behavioral tendencies will show obvious species limitations. Therefore, only the video data of the same species in different emotions or different disease states is obtained, so that the consistency of the data can be guaranteed. Specifically, by constructing animal experimental models in different states under the premise of complying with animal ethics, constructing experimental animals in different emotions, and collecting relevant video data for them.

[0018] In addition, the microorganisms in the intestines of animals are a complex micro-ecosystem composed of a variety of bacteria. During the collection process, animal feces samples or intestinal contents can be collected. For example, to collect intestinal flora data from mice, the tail-lifting method can be used, that is, the feces are picked up with sterilized tweezers after the mouse defecates; or the mouse can be placed in a feces collector or a sterilized empty cage, and the samples can be collected in time after it defecates naturally.

[0019] S2: extracting features from the behavior video data to obtain behavior features that characterize the animal's movement patterns; analyzing the intestinal flora data to obtain intestinal flora features.

[0020] In this embodiment, the steps of extracting features from the behavior video data to obtain behavior features characterizing the animal's action patterns; and analyzing the intestinal flora data to obtain intestinal flora features further include: S21: Based on the behavioral video data, the key points of the animal's body are tracked through a deep learning model to generate posture time series data; wherein, the deep learning model is a model trained based on the key point markings of the animal's body, and according to the differences in body shape of animals of different ages, a special model for young animals and a general model for adult animals are constructed respectively.

[0021] In the above process, the video data is first preprocessed, including preprocessing operations such as frame rate unification and size adjustment to ensure the consistency of the input data. When there is background interference in the video data, background removal or segmentation processing is performed. The experimenter or user marks the acquired video data frame by frame, or marks the key points of the animal's body frame by frame according to the self-report of the collected individual and the doctor's auxiliary judgment, such as manually marking the joints, head, tail and other key parts of the animal on the video frame. Use the labeled data to train the deep learning model so that it can automatically identify the key points of the animal in the video and track the position changes of the key points of the animal. Finally, the animal's posture time series data is generated based on the tracked key point positions.

[0022] S22: dividing the action labels according to the posture timing data, and calculating the duration, occurrence frequency and transition mode between actions of each action.

[0023] Specifically, the action labels include: running, trotting, walking, right turning, left turning, stepping, jumping, climbing, rearing, hunching, rising, sniffing, grooming, and pause.

[0024] See also Figure 2, which is a schematic diagram of the behavioral ratio of the duration of the sniffing action and the grooming action in Example 1 of the present application. The behavioral characteristics include at least two parameters of preset behavior types, the two preset behavior types are exploratory behavior and stereotyped behavior, wherein the exploratory behavior includes the sniffing action, and the stereotyped behavior includes the grooming action. The behavioral ratio is calculated based on the duration of the sniffing action and the grooming action, and the behavioral ratio is logarithmically (log10) transformed.

[0025] See also Figure 3 , is a schematic diagram of using linear discriminant analysis (LDA) to verify the behavior ratio classification in Example 1 of the present application. Based on the behavioral characteristics of animals, the LDA method is used to verify the two parameters of the duration and frequency of the action.

[0026] In this embodiment, video data of mice of different ages in a square open field (50×50×30 cm) for 60 minutes is obtained through a spontaneous behavior experiment. On the second day after the behavioral experiment, the mice were collected independently using an empty cage box sterilized with high temperature and high pressure to avoid urine contamination and stored in an ultra-low temperature refrigerator at -80°C. 1000-3000 frames of pictures were manually marked for 16 body points of the animal (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, tail root, middle tail, tail tip). The marked images and animal posture time series data were trained through Deeplabcut deep learning. Among them, the models were trained for 3, 4, and 5-week-old videos respectively, and a model was trained for 6, 7, 8, and 12-week-old mixed marked images. Apply BehaviorAtlas Analyzer, upload the corresponding model and video data of different ages for separate analysis. Further determine its action label, and merge similar action labels. Based on this, the duration of each action label and the frequency of the action are calculated, and then the behavioral parameters of the animal's sniffing and grooming actions are extracted. The behavioral ratio is calculated based on the duration of the sniffing action and grooming action, and the behavioral ratio is logarithmically (log10) transformed to evaluate the animal's behavioral state. Finally, the inter-group differences in the ratio and its changing trends at different developmental stages can be statistically analyzed to identify potential disease characteristics.

[0027] Based on the above scheme, the emotion or disease state of animals can be identified by observing their behavioral characteristics without using external stimulation. It reflects the real spontaneous performance of animals in real state, such as fear, addiction withdrawal, pain, etc., and provides comparative indicators that can be quantified and analyzed for the evaluation of fear degree, pain degree, addiction withdrawal pain degree, etc.

[0028] S23: Analyze the intestinal flora data using high-throughput sequencing and bioinformatics.

[0029] See also Figure 4-7 , is a schematic diagram of the results of Bray-Curtis similarity index PCOA analysis and Adonis test of the intestinal flora data of male and female mice of different ages in Example 1 of the present application. Among them, p<0.05 means that there is a statistical difference, R 2 Size represents the degree of difference.

[0030] The 16S rRNA sequencing technology combined with bioinformatics analysis methods was used to conduct an in-depth analysis of the intestinal flora of mice. Genomic DNA was extracted from fecal samples using a DNA extraction kit, and DNA concentration and purity were detected by agarose gel electrophoresis and NanoDrop2000. Specific primers were used to target the 16S rDNA V3-V4 region, and the target fragment was obtained by PCR amplification. The PCR product was purified by magnetic beads and quantified by Qubit. After equal amounts of mixed samples were sequenced on the machine (Illumina NovaSeq platform). The original sequencing data (FASTQ format) was removed from the primer sequence by Cutadapt, and quality control was performed using DADA2 to generate an ASV (Amplicon Sequence Variant) abundance table. ASV was annotated for species based on the Silva database (version138), and annotation results with a confidence level of >97% were retained. PCoA analysis based on the Bray-Curtis similarity index and Adonis test were used to screen the intestinal flora characteristics according to statistical significance to achieve classification of mice in different groups.

[0031] S3: Perform multimodal data fusion on the behavioral features and the intestinal flora features to generate fusion features.

[0032] In this embodiment, the behavior feature is represented as , is the number of behavioral characteristics; the intestinal flora characteristics are expressed as , is the number of intestinal flora features. Then the fusion feature is expressed as , is the fused feature vector, whose dimension is .

[0033] S4: Input the fusion features into a pre-trained data model, classify the animal's behavior through the data model, and output the classification result; wherein the data model is trained based on a convolutional neural network.

[0034] In this embodiment, the data model consists of an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer, and is used for data analysis of autism. The data model is represented as , It is the parameter optimized by the back propagation algorithm during the training of the data model.

[0035] The fused feature vector Input into the data model to obtain the classification result : , Select the animal behavior type with the highest probability as the prediction result: , in, To predict the type of animal behavior, To output the classification results Corresponding type The probability value of Used to select the animal behavior type with the highest probability from the probability distribution vector as the prediction result.

[0036] According to the above steps, this embodiment designed a systematic verification experiment of multiple genotypes (wild-type WT, heterozygous HE, homozygous KO), cross-sex, and full developmental cycle. The experiment was divided into 6 groups, each containing 15 mice, and spontaneous behavior experiments were conducted at 3, 4, 5, 6, 7, 8, and 12 weeks of age. Spontaneous behavior was monitored for 60 minutes per age period, using a standardized square open field apparatus (50×50×30 cm), and video was collected by cameras at the four corners. Through analysis and calculation of behavioral data, it was verified that the logarithm (log10) of the ratio of sniffing and grooming actions can accurately distinguish autism model mice of different genotypes. There were significant statistical differences between homozygous (KO) and heterozygous (HE) autism model mice of each age and the control wild-type (WT) mice ( Figure 2 ). At 3 and 5 weeks of age, this value changed dramatically compared with other weeks and dropped to negative values, which is considered to be the critical window period for disease development. Although this ratio fluctuated within the same group of mice, it was generally stable and maintained within a certain range. This is highly consistent with the results obtained by the LDA method ( Figure 3 In terms of gender differences, the experimental results showed that there were significant differences between different genotypes of female mice at different ages (p<0.01), while the differences between different genotypes of male mice were not significant, which shows that the method of the present invention can effectively distinguish gender differences ( Figure 4-Figure 7 ). The above experimental results show that this method has high stability and good generalization.

[0037] In summary, Example 1 of the present application proposes a multimodal animal abnormal data behavior classification method, which is mainly used for the classification and early detection of autism model animals. This method realizes high-precision identification of autism-like phenotypes based on animal behavioral characteristics and intestinal flora diversity data analysis. Specifically, it includes calculating the behavioral ratio of sniffing and grooming based on behavioral characteristics. This ratio can reflect the balance between the animal's exploration of the outside world and its own attention behavior, and is an important indicator for judging autism-like phenotypes. Furthermore, the LDA method is used to verify the accuracy of the ratio classification, and the changing trend of the ratio at different developmental stages is analyzed, which provides clues for identifying the key window period of disease development. At the same time, the intestinal flora data is deeply analyzed by high-throughput sequencing technology, and the behavioral characteristics and intestinal flora characteristics are integrated for classification, which improves the accuracy of classification and research efficiency. It does not rely on a single behavioral threshold judgment, but combines behavioral and physiological multimodal data to provide accuracy in animal behavior classification.

[0038] Example 2 See also Figure 8 , is a structural diagram of a multimodal animal abnormal data behavior classification device according to Example 2 of the present application; the specific contents include: Data acquisition module: acquires abnormal data of target animals at different developmental stages; the abnormal data includes behavioral video data and intestinal flora data of the same animal; Feature extraction module: extract features from the behavior video data to obtain behavior features that characterize the animal's action patterns; analyze the intestinal flora data to obtain intestinal flora features; Feature fusion module: fusing the behavior feature and the intestinal flora feature into multimodal data to generate fusion features; Classification module: input the fusion features into a pre-trained data model, classify the animal's behavior through the data model, and output the classification result; wherein the data model is trained based on a convolutional neural network.

[0039] In this embodiment, the multimodal animal abnormal data behavior classification device includes a data acquisition module, a feature extraction module, a feature fusion module, and a classification module. The specific implementation method of each module is as follows: Data acquisition module: integrated with high-definition camera and automatic feces sampling device, supporting real-time data transmission. Specifically, data is acquired at 3, 4, 5, 6, 7, 8, and 12 weeks of age. Shank3bAutism model mice with gene mutation (KO type), heterozygous (HE type) and wild type (WT type), 15 mice in each group, half male and half female. A standardized open field apparatus (50×50×30 cm) was used to continuously record spontaneous behavior videos for 60 minutes using a four-corner camera. The collection of intestinal flora data was performed by obtaining mouse fecal samples at the same time point and storing them in an ultra-low temperature refrigerator at -80℃. Finally, the obtained behavioral video data and intestinal flora data were input into the feature extraction module for feature extraction.

[0040] Feature extraction module: 16 key points of the body were manually marked in videos of 3, 4, and 5-week-old mice to train a special model for young mice; mixed markings were performed on videos of 6, 7, 8, and 12-week-old mice to train a general model for adults. The trained model was used to track key points of the body, generate posture time series data, divide action labels, and finally calculate the duration and frequency of each action. In addition, the extraction of microbial features: the fecal samples were sequenced using the Illumina NovaSeq platform, and the ASV abundance table was generated using the DADA2 process. PCoA analysis was performed based on the Bray-Curtis similarity index to screen for significantly different microbial features.

[0041] Feature fusion module: The behavioral features obtained by the feature extraction module (such as the sniffing / grooming duration ratio log 10 ) and intestinal flora characteristics (such as differential bacterial genus abundance) into a fusion feature vector.

[0042] Classification module: Input the fusion features output by the feature fusion module into the pre-trained data model, output three types of probabilities (normal, heterozygous, and homozygous), and select the highest probability as the classification result. The classification module also includes a visualization unit, which is used to generate an interpretable heat map of the classification results, annotate the behavioral characteristics and intestinal flora characteristics that contribute most to the classification decision, such as grooming frequency and Lactobacillus abundance, which can help researchers focus on core phenotypic characteristics. In addition, through this device, users can upload mouse videos and feces data to the cloud, and the device automatically generates fusion features and classifies them, and feedbacks results and visualization reports.

[0043] Example 3 See also Fig. 9 , is a schematic diagram of the computer device structure of Example 3 of the present application. The computer device 50 includes a processor 51 and a memory 52 coupled to the processor 51 .

[0044] The memory 52 stores program instructions for implementing the above-mentioned multimodal animal abnormal data behavior classification method.

[0045] The processor 51 is used to execute program instructions stored in the memory 52 to implement a multi-modal animal abnormal data behavior classification.

[0046] The processor 51 may also be referred to as a CPU (Central Processing Unit).

[0047] The processor 51 may be an integrated circuit chip with signal processing capability. The processor 51 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0048] Example 4 See also Fig.10 , which is a schematic diagram of the structure of the storage medium of Example 4 of the present application. The storage medium of the embodiment of the present application stores a program file 61 that can implement all the above methods, wherein the program file 61 can be stored in the above storage medium in the form of a software product, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk, or a computer, server, mobile phone, tablet and other devices.

[0049] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0050] The above description is only a preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

[0051] Although the embodiments of the present application have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present application, and that the scope of the present application is defined by the appended claims and their equivalents.

[0052] Of course, the present invention may have many other implementations. Based on this implementation, other implementations obtained by ordinary technicians in this field without any creative work are all within the scope of protection of the present invention.

Claims

1. A multimodal animal abnormal data behavior classification method, characterized in that: include: Obtain abnormal data of target animals at different developmental stages; The abnormal data includes behavioral video data and intestinal flora data of the same animal; Extracting features from the behavior video data to obtain behavior features that characterize the animal's action patterns; Analyzing the intestinal flora data to obtain intestinal flora characteristics; Performing multimodal data fusion on the behavioral features and the intestinal flora features to generate fusion features; The fusion feature is input into a pre-trained data model, the animal's behavior is classified through the data model, and the classification result is output; wherein the data model is obtained by training based on a convolutional neural network.

2. A multimodal animal abnormal data behavior classification method according to claim 1, characterized in that: The step of extracting features from the behavior video data to obtain behavior features that characterize the animal's action pattern specifically includes: Based on the behavior video data, the key points of the animal's body are tracked through a deep learning model to generate posture time series data; wherein the deep learning model is a model trained based on the key point markings of the animal's body, and a special model for young animals and a general model for adult animals are constructed according to the body shape differences of animals of different ages; Action labels are divided according to the posture timing data, and the duration, occurrence frequency and transition mode between actions of each action are calculated.

3. A multimodal animal abnormal data behavior classification method according to claim 2, characterized in that: The behavior characteristics include at least two parameters of preset behavior types, the two preset behavior types are exploratory behavior and stereotyped behavior, wherein the exploratory behavior includes sniffing action, and the stereotyped behavior includes grooming action; Behavioral ratios were calculated based on the duration of sniffing and grooming actions and log-transformed.

4. A multimodal animal abnormal data behavior classification method according to claim 1, characterized in that: The step of analyzing the intestinal flora data to obtain intestinal flora characteristics specifically includes: The intestinal flora data are analyzed using high-throughput sequencing and bioinformatics; The fecal samples of the target animals were sequenced by 16S rRNA gene to obtain the characteristics of the intestinal flora.

5. A multimodal animal abnormal data behavior classification method according to claim 1, characterized in that: The step of inputting the fusion feature into a pre-trained data model, classifying the behavior of the animal through the data model, and outputting the classification result; wherein the data model is obtained by training based on a convolutional neural network, specifically includes: Assume that the behavior characteristics are expressed as , is the number of behavioral characteristics; the intestinal flora characteristics are expressed as , is the number of intestinal flora features; the fusion feature is expressed as , is the fused feature vector, whose dimension is ; The data model consists of an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer; the data model is represented as , The parameters optimized by the back propagation algorithm during the training of the data model; The fused feature vector Input into the data model to obtain the classification result : ， Select the animal behavior type with the highest probability as the prediction result: ， in, To predict the type of animal behavior, To output the classification results Corresponding type The probability value of Used to select the animal behavior type with the highest probability from the probability distribution vector as the prediction result.

6. A multimodal animal abnormal data behavior classification method according to claim 1, characterized in that: The target animal is an autism model animal, including but not limited to a rodent; The abnormal behavioral phenotype of the target animal is an autism-like phenotype, manifested as at least one of repetitive stereotyped behaviors, social disorders, or abnormal intestinal flora.

7. A multimodal animal abnormal data behavior classification device, characterized in that: include: Data acquisition module: acquires abnormal data of target animals at different developmental stages; the abnormal data includes behavioral video data and intestinal flora data of the same animal; Feature extraction module: extracting features from the behavior video data to obtain behavior features that characterize the animal's action patterns; Analyzing the intestinal flora data to obtain intestinal flora characteristics; Feature fusion module: fusing the behavior feature and the intestinal flora feature into multimodal data to generate fusion features; Classification module: input the fusion features into a pre-trained data model, classify the animal's behavior through the data model, and output the classification result; wherein the data model is trained based on a convolutional neural network.

8. A multimodal animal abnormal data behavior classification device according to claim 7, characterized in that: The classification module also includes a visualization unit, which is used to generate an interpretable heat map of the classification results, marking the behavioral features and intestinal flora features that contribute most to the classification decision.

9. A computer device, characterized in that: The computer device includes a processor and a memory coupled to the processor, wherein the memory stores program instructions for implementing a multimodal animal abnormal data behavior classification method as described in any one of claims 1-6; and the processor is used to execute the program instructions stored in the memory to implement a multimodal animal abnormal data behavior classification.

10. A storage medium, characterized in that: The invention stores program instructions executable by a processor, wherein the program instructions are used to execute a multimodal animal abnormal data behavior classification method as described in any one of claims 1 to 6.

Images