Method and device for determining invalid archives, electronic equipment and storage medium

Abstract

The invention discloses an invalid file determination method and device, electronic equipment and a storage medium, and the method comprises the steps: determining a candidate file in each target object file, the candidate file comprises at least two target object tracks, and each target object track corresponds to a target object feature of a preset dimension; determining at least two target dimensions meeting a characteristic difference condition in preset dimensions of each target object track of the candidate file; according to the target object features of the preset dimensions of the target object tracks of the candidate archives and the target object features of the target dimensions, target object feature fusion is carried out; and judging whether the candidate file is an invalid file or not according to the candidate file after the target object feature fusion. According to the invention, the effectiveness of the target object file can be judged, invalid files can be accurately identified, and computing resources are saved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, electronic device, and storage medium for determining invalid files. Background Technology

[0002] With the development of video processing technology, archives are increasingly used in the field of video processing. Based on video captured by cameras, target object recognition is performed, and the identified target objects are clustered to obtain trajectory information of the same target object, forming target object archives. However, target object recognition often cannot achieve good results in all complex scenes. In low-light environments or when target objects are occluded, they are easily grouped into the same archive by clustering algorithms. Such archives, due to occlusion or the presence of multiple different target objects, can be called invalid archives. Invalid archives have low actual utilization rates and can interfere with subsequent in-depth information mining. Therefore, it is necessary to identify and remove invalid archives from each target object archive.

[0003] Current technologies for identifying invalid files involve verifying the spatiotemporal information of target object trajectories within the target object file to determine the rationality of this information. However, extracting only a portion of the target object trajectories from the file and verifying only those extracted trajectories can easily lead to false positives and false negatives. Conversely, verifying the spatiotemporal information of all target object trajectories would consume significant computational resources. Summary of the Invention

[0004] This invention provides a method, apparatus, electronic device, and storage medium for determining invalid files, so as to judge the validity of target files, accurately identify invalid files, and save computing resources.

[0005] In a first aspect, embodiments of the present invention provide a method for determining invalid files, the method comprising:

[0006] Candidate files are determined from each target object file, wherein each candidate file includes at least two target object trajectories, and each target object trajectory corresponds to a target object feature of a preset dimension.

[0007] Among the preset dimensions of the trajectory of each target object in the candidate archive, at least two target dimensions that satisfy the feature difference condition are determined.

[0008] Based on the target object features of each target object trajectory in the candidate files in the preset dimension, and the target object features in the target dimension, target object feature fusion is performed.

[0009] Based on the candidate files after fusing the features of the target object, determine whether the candidate files are invalid files.

[0010] Secondly, embodiments of the present invention also provide an apparatus for determining invalid files, the apparatus comprising:

[0011] The candidate file determination module is used to determine candidate files from each target object file. Each candidate file includes at least two target object trajectories, and each target object trajectory corresponds to a target object feature of a preset dimension.

[0012] The target dimension determination module is used to determine at least two target dimensions that satisfy the feature difference condition from the preset dimensions of the trajectory of each target object in the candidate file.

[0013] The target object feature fusion module is used to perform target object feature fusion based on the target object features of each target object trajectory in the candidate file in a preset dimension, as well as the target object features in the target dimension.

[0014] The invalid file determination module is used to determine whether a candidate file is invalid based on the candidate files after the feature fusion of the target object.

[0015] Thirdly, embodiments of the present invention also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for determining invalid files as described in any of the embodiments of the present invention.

[0016] Fourthly, embodiments of the present invention also provide a storage medium for storing computer-executable instructions, which, when executed by a computer processor, are used to perform the invalid file determination method as described in any of the embodiments of the present invention.

[0017] The technical solution of this invention involves filtering candidate files from various target object files, selecting target dimensions that meet feature difference conditions for each preset dimension of the target object trajectory of each candidate file, performing feature fusion based on the target object features of each preset dimension of the target object trajectory of each candidate file and the target object features of the target dimension that meets the feature difference conditions, and determining whether the candidate file is invalid based on the feature-fused candidate file. This solves the problems of existing invalid file identification methods, which are prone to misjudgment and omission or waste of computational resources. It not only enables the determination of the validity of target object files and accurately identifies invalid files, but also saves computational resources.

[0018] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a method for determining invalid files provided in Embodiment 1 of the present invention;

[0021] Figure 2 This is a flowchart of a method for determining invalid files provided in Embodiment 2 of the present invention;

[0022] Figure 3 This is a schematic diagram of the structure of an invalid file determination device provided in Embodiment 3 of the present invention;

[0023] Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices. In the embodiments of this application, certain software, components, models, and other existing industry solutions may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solutions of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0026] The acquisition, transmission, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0027] Example 1

[0028] Figure 1 The flowchart below illustrates a method for determining invalid files according to Embodiment 1 of the present invention. This embodiment is applicable to identifying invalid files in a target file. The method can be executed by an invalid file determination device, which can be implemented in hardware and / or software and can be configured in a server.

[0029] like Figure 1 As shown, the method includes:

[0030] S110. Identify candidate files from the files of each target object.

[0031] The target object can refer to a face, a human body, a vehicle, equipment, etc. This embodiment does not limit the type of target object. The target object file refers to a file that aggregates the video frame images corresponding to the trajectory of the target object that is considered to belong to the same target object. Each target object trajectory corresponds to a set of video frame images, and the number of video frame images corresponding to different target object trajectories can be the same or different.

[0032] Understandably, if a target object is occluded, tilted, or has low brightness due to environmental factors, it is easily misidentified as the same target object, causing the video frames corresponding to its trajectory to be grouped into the same target object file. In this case, the target object trajectory in the target object file may have multiple sources and locations, belonging to multiple different target objects, which will interfere with subsequent in-depth information mining of the target object file. Therefore, such target object files are invalid and should be identified and removed. Candidate files are those that have undergone preliminary screening of target object files and may be invalid.

[0033] The candidate file includes at least two target object trajectories, and each target object trajectory corresponds to a target object feature of a preset dimension. For example, the candidate file may include 100 target object trajectories, and each target object trajectory corresponds to a target object feature of 512 dimensions. Then, the target object features of the candidate file can construct a 100×512 target object feature matrix.

[0034] In this embodiment, candidate files are selected from each target object file to determine which files are invalid. Specifically, candidate files can be filtered at preset time intervals, or when a new target object trajectory is detected; this embodiment does not impose any restrictions on this.

[0035] In an optional embodiment, all target object files can be directly used as candidate files, and invalid files can be identified for each target object file. This setting can avoid missing the identification of invalid files.

[0036] In another optional embodiment, among each target object file, target object files with a higher probability of being invalid files can be selected as candidate files based on the number of target object trajectories, the quality of target object trajectories, etc. This setting can save computing power and improve the efficiency of invalid file identification.

[0037] Furthermore, identifying candidate files from each target file may include:

[0038] A1. If the number of target object trajectories in the target object file is greater than or equal to a preset number threshold, then determine the trajectory quality of each target object trajectory.

[0039] Trajectory quality is used to represent the recognizability of the target object in the target object trajectory;

[0040] A2. If it is determined that the trajectory quality of each target object's trajectory meets the preset trajectory quality conditions, then the target object file is used as a candidate file.

[0041] Understandably, the same target object typically doesn't appear frequently within a certain time period. Conversely, invalid files often contain a large number of target object trajectories because the captured target objects share similar characteristics such as occlusion, angle deviation, and low brightness. Therefore, in this embodiment, candidate files can be initially screened based on the number of target object trajectories within the target object file.

[0042] In this embodiment, the number of target object files and candidate files can be counted and filtered at regular intervals. Alternatively, the number of each target object file can be counted in real time, and further processing can be performed when the number of a certain target object file is greater than or equal to a preset threshold.

[0043] Track quality represents the recognizability of a target object within its trajectory. Track quality is directly proportional to the recognizability of the target object within its trajectory; the higher the recognizability, the better the track quality. Understandably, if the recognizability of the target object is high and the track quality is good, the target object's trajectory is less likely to be incorrectly aggregated into invalid files. Therefore, in this embodiment, track quality can be used to filter target object files with a large number of target object trajectories from each target object file, and then the track quality can be used to determine whether to include them as candidate files.

[0044] In a specific example, trajectory quality can be represented by a score of the target object when the target object trajectory is obtained by processing consecutive video frame images using a target object recognition algorithm. For instance, the score can be represented by the similarity between the identified target object and a known standard target object when the target object recognition algorithm performs target object recognition on the video frame images. The average similarity of the target objects corresponding to each video frame image in the target object trajectory is calculated as the trajectory quality of the target object trajectory.

[0045] In another specific example, trajectory quality can also be represented by the completeness and / or sharpness of each target object in the target object trajectory. For instance, for a target object identified in a frame of the target object trajectory, a known matching standard target object is determined, and both the standard target object and the identified target object are projected onto the same scale. The completeness of the identified target object is calculated based on the ratio of its pixel area to the projected pixel area of ​​the standard target object. The average completeness of each target object in the target object trajectory is taken as the trajectory quality. Alternatively, the sharpness of each target object region in the target object trajectory can be calculated using methods such as image frequency domain transformation, image information entropy, or the Laplacian algorithm, and the average sharpness of each target object in the target object trajectory is taken as the trajectory quality.

[0046] In another specific example, after calculating the similarity, completeness, and sharpness of the target object in each frame of the target object's trajectory as described above, a weighted sum of these three values ​​can be performed to obtain a score for each frame of the target object. The average score of each frame's target object is then calculated as the trajectory quality.

[0047] The above example is merely an illustration of the trajectory quality calculation for the target object's trajectory. This embodiment does not limit the specific method for determining trajectory quality.

[0048] In this embodiment, the trajectory quality of each target object trajectory meets the preset trajectory quality conditions, indicating that the overall trajectory quality of each target object trajectory in the target object file is poor. Therefore, in this embodiment, the target object file with a large number of target object trajectories and a generally poor trajectory quality is selected as a candidate file for subsequent invalid file judgment.

[0049] In a specific example, a trajectory quality threshold can be preset. If the average trajectory quality of each target object trajectory is less than or equal to the trajectory quality threshold, then the trajectory quality of each target object trajectory is determined to meet the preset trajectory quality condition.

[0050] In another specific example, a trajectory quality threshold can be preset. Among the trajectory of each target object in the target object file, the number of target object trajectories with a trajectory quality less than or equal to the trajectory quality threshold can be determined. If the ratio of the above number to the total number of all target object trajectories is greater than or equal to a preset ratio threshold, then the trajectory quality of each target object trajectory is determined to meet the preset trajectory quality condition.

[0051] In another specific example, different trajectory quality intervals can be predetermined, and a trajectory quality threshold can be determined. For instance, if trajectory quality is represented by values ​​from 0 to 100, the trajectory quality intervals can be set to [0, 25), (25, 50), [50, 75), and [75, 100), with a trajectory quality threshold set to 75. The number of target object trajectories within each trajectory quality interval is counted, and the number of target object trajectories whose trajectory quality falls within the trajectory quality interval below the trajectory quality threshold is calculated. If the ratio of this number to the total number of all target object trajectories is greater than or equal to a preset percentage threshold, then the trajectory quality of each target object trajectory is determined to meet the preset trajectory quality condition.

[0052] Similarly, the above examples are merely specific examples of trajectory quality conditions, and this embodiment does not limit the specific judgment process of trajectory quality conditions.

[0053] Furthermore, identifying candidate files from each target file may also include:

[0054] S1. Determine the number of target object trajectories in the target object file, and the trajectory quality of each target object trajectory;

[0055] S2. Calculate the trajectory quality score of the target object file based on the trajectory quality of each target object's trajectory.

[0056] S3. Calculate the file score of the target object file based on the number of target object trajectories, the quantity weight, the trajectory quality score, and the score weight.

[0057] S4. Based on the file scores of each target file, determine the candidate files among the target files.

[0058] The process for determining the trajectory quality is the same as in the above embodiments, and will not be repeated here.

[0059] Based on the trajectory quality of each target object's trajectory, a trajectory quality score for the target object file is calculated. In a specific example, the trajectory quality score for the target object file can be determined based on the average trajectory quality of each target object's trajectory. Specifically, the average trajectory quality can be used directly as the trajectory quality score, or a range of average trajectory quality values ​​can be set, with different ranges corresponding to different trajectory quality scores.

[0060] In another specific example, the same method as in the above embodiments can be used to calculate the ratio of the number of target object trajectories with trajectory quality less than or equal to the trajectory quality threshold to the total number of all target object trajectories, or the ratio of the number of target object trajectories with trajectory quality falling within the trajectory quality range less than the trajectory quality threshold to the total number of all target object trajectories. Based on the ratio calculated above and the mapping relationship between the pre-set ratio range and the trajectory quality score, the trajectory quality score corresponding to the ratio calculated above is determined.

[0061] This embodiment does not limit the specific process of calculating the trajectory quality score of the target object file based on the trajectory of each target object.

[0062] In this embodiment, a weighted calculation is performed based on the number of target object trajectories, the quantity weight, the trajectory quality score, and the score weight of the target object file to obtain the file score. The lower the file score, the higher the probability that the target object file is an invalid file.

[0063] It should be noted that the more target object tracks a file has, the higher the probability that the file is invalid; that is, the number of target object tracks is inversely proportional to the file score. Conversely, the lower the track quality score, the higher the probability that the target object file is invalid; that is, the track quality score is directly proportional to the file score. Therefore, when weighting the number of target object tracks and the track quality score, the weight of the number of target object tracks can be set to a negative value, or the reciprocal of the number of target object tracks can be taken before weighting it with the track quality score.

[0064] In this embodiment, candidate files are determined from each target object file based on its file score. This can be achieved by sorting the target object files by their file scores from smallest to largest, and then selecting the first preset number of target object files as candidate files in ascending order of file score. Alternatively, the number of candidate files can be calculated by multiplying the total number of target object files by a preset ratio, and then selecting the corresponding number of candidate files in ascending order of file score.

[0065] In this embodiment, based on the number of target object trajectories and the trajectory quality of each target object trajectory, a preliminary screening is performed in each target object file, and candidate files are selected for subsequent identification of invalid files. This improves the efficiency of invalid file identification and saves computing resources while ensuring that invalid files are not missed.

[0066] S120. Among the preset dimensions of the trajectory of each target object in the candidate archive, determine at least two target dimensions that satisfy the feature difference condition.

[0067] Among these, satisfying the feature difference condition means that there is a significant difference in the features of the target objects in this dimension among different target objects. The target dimension refers to the dimension that meets the feature difference condition selected from all preset dimensions, and the number of target dimensions is less than the number of preset dimensions.

[0068] Taking a face as an example, the trajectories of different target objects may be incorrectly aggregated into invalid files due to facial occlusion (such as wearing masks). In this case, the differences in target object features corresponding to the facial occlusion areas are small between different target objects. It is precisely because the target object features of different target objects are relatively similar in certain dimensions, and there are many such similar dimensions, that different target objects ultimately appear to have high similarity, and their trajectories are incorrectly aggregated into the same target object file. Therefore, the technical solution of this embodiment ignores the dimensions where the target object features of different target object trajectories are relatively similar in each preset dimension, and selects each target dimension that meets the feature difference condition, thereby enhancing the overall difference of the target object trajectories.

[0069] Taking the above candidate files as an example, which include 100 target object trajectories and each target object trajectory corresponds to 512-dimensional target object features, the target object features of the candidate files can be constructed into a 100×512 target object feature matrix. After selecting 64 target dimensions that meet the feature difference conditions, the target object features of the candidate files are represented as a 100×64 target object feature matrix.

[0070] In an optional embodiment, among the preset dimensions of each target object trajectory in the candidate archives, at least two target dimensions that satisfy the feature difference condition are determined. This can be achieved using Principal Component Analysis (PCA) technology. PCA technology projects the original preset dimensions and target object features into a new space through linear transformation, retaining the principal components with larger variance values, thereby achieving dimensionality reduction of the target object features.

[0071] In another optional embodiment, determining at least two target dimensions that satisfy the feature difference condition from the preset dimensions of each target object trajectory in the candidate archives can also be achieved using Linear Discriminant Analysis (LDA). LDA is a supervised linear feature extraction and dimensionality reduction method that improves classification performance while reducing dimensionality, and the target object features of the dimensionality-reduced target dimensions have better interpretability.

[0072] In another optional embodiment, among the preset dimensions of each target object trajectory in the candidate archives, at least two target dimensions that satisfy the feature difference condition are determined. This can also be achieved by training a deep learning model. Specifically, sample target object trajectories that are aggregated into the same target object archive are determined, and the target objects to which different sample target object trajectories belong are labeled. A pre-set deep learning model is trained using the sample target object trajectories to obtain a target dimension extraction model. The target object trajectories of each target object in the candidate archives are input into the target dimension extraction model to obtain the target dimension and target object features output by the target dimension extraction model.

[0073] In this embodiment, the process of selecting target dimensions that meet the feature difference conditions from each preset dimension is a process of dimensionality reduction of the target object features. By reducing the dimensionality of the target object features of the target object trajectory, the difference between the target object features of the target object trajectory is increased.

[0074] S130. Based on the target object features of each target object trajectory in the candidate files in the preset dimension, and the target object features in the target dimension, perform target object feature fusion.

[0075] In this embodiment, target object features are fused based on the original preset dimensions of each target object features and the target dimensions of each target object features extracted based on feature differences.

[0076] In this embodiment, for each target object feature of a preset dimension of the target object trajectory in the candidate file, if it is determined that a certain dimension is a target dimension extracted based on feature differences, that is, a dimension with large feature differences, then the original target object features of that dimension and the target object features of the extracted target dimension that matches that dimension are fused. If it is determined that a certain dimension is not a target dimension extracted based on feature differences, then the original target object features of that dimension are retained.

[0077] Furthermore, the target object feature fusion is performed on the original target object features of a certain dimension and the extracted target object features of the target dimension that match that dimension. This can be done by concatenating the original target object features of that dimension and the extracted target object features of the target dimension that match that dimension, superimposing the original target object features of that dimension and the extracted target object features of the target dimension that match that dimension, or by performing a weighted summation of the original target object features of that dimension and the extracted target object features of the target dimension that match that dimension. This embodiment does not limit the specific method of target object feature fusion.

[0078] Furthermore, when performing a weighted summation of the original target object features in a certain dimension and the extracted target object features matching that dimension, a fixed weight can be assigned to each original target object feature. The weights of each target object feature in the target dimension can be set based on the magnitude of the feature differences within that dimension; the greater the feature differences, the larger the weight can be. Specifically, the magnitude of the feature variance value in the target dimension represents the magnitude of the feature differences, and different weights can be assigned to different feature variance value ranges. The feature variance value of the target dimension is calculated based on the target object features of each target object trajectory in that target dimension.

[0079] In this embodiment, the target object features of the reduced target dimension are fused with the target object features of the original preset dimension. This can enhance the differences between the trajectories of each target object in the candidate file. By amplifying the potential differences between the target object trajectories, it can accurately determine whether the target object trajectories correspond to the same target object, thereby determining whether the candidate file is an invalid file.

[0080] S140. Based on the candidate files after the feature fusion of the target object, determine whether the candidate files are invalid files.

[0081] In this embodiment, the differences between target object trajectories are amplified by target object feature fusion. At this time, if each target object trajectory in the candidate file corresponds to multiple target objects, it indicates that the result of target object trajectory aggregation in the candidate file is unstable, and the candidate file is an invalid file with aggregation abnormality.

[0082] In an optional embodiment, based on the candidate files after target object feature fusion, it is determined whether the candidate files are invalid files. This can be done by calculating the similarity between the target objects corresponding to each pair of target object trajectories in the candidate files after target object feature fusion. If it is determined that the similarity between the target objects corresponding to each pair of target object trajectories is greater than or equal to a preset similarity threshold, then the candidate files are valid files; otherwise, they are invalid files.

[0083] In another optional embodiment, determining whether a candidate file is invalid based on the candidate files after feature fusion of the target object may further include:

[0084] C1. Re-cluster the candidate files after feature fusion of the target object to obtain the re-clustering result;

[0085] C2. Based on the re-clustering results, determine whether the candidate files are invalid files.

[0086] Clustering can be achieved through Euclidean distance-based clustering algorithms, Hidden Markov Models, Gaussian Mixture Models, density-based clustering algorithms such as DBSCAN (Density-Based Spatial Clustering of Applications with Noise), and training deep learning models. This embodiment does not limit the specific methods and processes of clustering.

[0087] In an optional embodiment, determining whether a candidate file is invalid based on the re-clustering result can be done by classifying the candidate file as invalid if the re-clustering result shows at least two clusters. After target object feature extraction, target object feature fusion, and re-clustering in the target dimension that satisfy the feature difference condition, the candidate file forms at least two new clusters. That is, the target object trajectory in the candidate file corresponds to at least two different target objects, and the number of different target objects corresponding to each target object trajectory is the number of clusters obtained after re-clustering. This indicates that the aggregation result of the candidate file is unstable and contains aggregation errors, thus classifying it as an invalid file.

[0088] In another optional embodiment, determining whether a candidate file is invalid based on the re-clustering results can be done as follows: if the re-clustering result is determined to be at least three clusters, then the candidate file is determined to be invalid; or, if the re-clustering result is determined to be two clusters, and each cluster includes at least two target object trajectories, then the candidate file is determined to be invalid. Based on the above embodiments, this embodiment adds a check on the number of target object trajectories in each cluster. It is understood that if, after re-clustering, the file is divided into two clusters, and one cluster includes only one target object trajectory (i.e., all other target object trajectories correspond to the same target object), then, in addition to target object file archiving errors, there may also be cases of target object identification anomalies. In this case, a target object identification anomaly warning can be issued for the candidate file, and the candidate file is determined to be invalid based on the feedback result of the target object identification anomaly warning.

[0089] In this embodiment, starting from increasing the differences between the trajectories of each target object, the candidate file is determined as invalid based on the re-clustering results of the trajectories of each target object in the candidate file, thereby improving the accuracy and applicability of invalid file identification.

[0090] It should be noted that this embodiment uses the example of fusing the target object features of each target object in the target dimension with the target object features of the original preset dimension, and then re-clustering them. Alternatively, after extracting the target object features of each target object in the target dimension with significant feature differences, re-clustering can be performed based on the target object trajectories in the candidate files according to the target object features of each target object in the target dimension, and then the re-clustering result can be used to determine whether the candidate files are invalid. Alternatively, after extracting the target object features of each target dimension with significant feature differences, the target object features other than those in the target dimension can remain unchanged in the original preset dimension. Feature enhancement processing can be performed on the target object features of the target dimension, for example, through filters, feature amplification algorithms, etc., to amplify or enhance the target object features of the target dimension. After feature amplification or enhancement, re-clustering is performed, and then the re-clustering result can be used to determine whether the candidate files are invalid. This embodiment does not impose any limitations on this approach.

[0091] The technical solution of this invention involves filtering candidate files from various target object files, selecting target dimensions that meet feature difference conditions for each preset dimension of the target object trajectory of each candidate file, performing feature fusion based on the target object features of each preset dimension of the target object trajectory of each candidate file and the target object features of the target dimension that meets the feature difference conditions, and determining whether the candidate file is invalid based on the feature-fused candidate file. This solves the problems of existing invalid file identification methods, which are prone to misjudgment and omission or waste of computational resources. It not only enables the determination of the validity of target object files and accurately identifies invalid files, but also saves computational resources.

[0092] Example 2

[0093] Figure 2 This is a flowchart of a method for determining invalid files according to Embodiment 2 of the present invention. Based on the above embodiments, the present invention further specifies the process of screening candidate files, the process of determining the target object features of the target dimension that meet the feature difference conditions, and the process of target object feature fusion.

[0094] like Figure 2 As shown, the method includes:

[0095] S210. Identify candidate files from the files of each target object.

[0096] The process of selecting candidate files from each target file has been described in the above embodiments, and will not be repeated here.

[0097] S220. Based on the target object features of each preset dimension of the target object trajectory in the candidate files, determine the feature values ​​corresponding to each preset dimension, and sort the feature values ​​corresponding to each preset dimension in descending order.

[0098] Among them, the feature values ​​corresponding to each preset dimension are used to represent the degree of contribution of each dimension in determining the feature differences between the target object trajectories.

[0099] In an optional embodiment, the feature values ​​corresponding to each preset dimension can be determined based on the variance of the target object features of each preset dimension. Specifically, the target object features of each preset dimension in the target object trajectory of the candidate file are grouped according to the dimension, and the target object features of each preset dimension in each target object trajectory are grouped into a group. For a group of target object features, the mean of all target object features in that dimension is calculated, and then the variance of each target object feature is calculated based on the mean. The average of the variances of all target object features in that dimension is taken as the feature value corresponding to that dimension.

[0100] In another optional embodiment, this embodiment takes calculating the feature values ​​of a preset dimension using principal component analysis and then reducing the dimensionality of the original preset dimension as an example.

[0101] Specifically, a candidate profile containing n-dimensional target object features and trajectories of m target objects can be represented by an m×n target object feature matrix. The standardization process for each target object feature in the m×n target object feature matrix can be expressed by the following formula: Where, x ij μ represents the target object feature in the j-th dimension of the trajectory of the i-th target object in the target object feature matrix X. j σ refers to the mean of the features of each target object in the j-th dimension of the target object trajectory. j x' refers to the standard deviation of the target object features in the j-th dimension of each target object trajectory. ij It refers to the characteristics of the target object after standardization.

[0102] The standardized feature matrix of the target object is represented by X'. The covariance matrix of X' is calculated using the following formula: We obtain the n×n covariance matrix S.

[0103] Perform eigenvalue decomposition on S and solve the characteristic equation |S-λI|=0, where λ is the eigenvalue and I is the identity matrix. Solving the characteristic equation yields n eigenvalues ​​λ1, λ2, ..., λ1. n The eigenvalues ​​are sorted in descending order to obtain λ1', λ2', ..., λ... n '.

[0104] S230. Plot the total variance percentage-principal component number curve with the number of principal components as the horizontal axis and the cumulative total variance percentage of the first principal components' eigenvalues ​​in descending order as the vertical axis.

[0105] The number of principal components ranges from 0 to the number of preset dimensions. Taking the target object feature matrix of the above candidate file as 100×512 as an example, the number of principal components is 0-512.

[0106] After sorting the eigenvalues ​​in descending order, we obtain the eigenvalues: λ1', λ2', ..., λ n '. Calculate the proportion of total variance for each eigenvalue in turn. The proportion of total variance can be calculated using the following formula:' T iThis represents the proportion of total variance for the i-th eigenvalue. For each principal component, calculate the cumulative value obtained by summing the proportions of total variance for the first i-th eigenvalues. Construct a curve of cumulative total variance proportion versus number of principal components, with the number of principal components on the horizontal axis and the cumulative value on the vertical axis.

[0107] For example, when the number of principal components on the horizontal axis is 1, the vertical axis is T1; when the number of principal components on the horizontal axis is 2, the vertical axis is T1+T2, and so on.

[0108] In this embodiment, the cumulative value of the total variance ratio of the eigenvalues ​​corresponding to the number of principal components is calculated, which can determine the number of dimensions with large feature differences based on the changes in the curve.

[0109] S240. Based on the total variance cumulative value-principal component number curve, determine the number of principal components that meet the curve slope condition as the target number.

[0110] The condition of satisfying the curve slope can be defined as the curve slope changing the most. Specifically, the slope of the curve corresponding to the cumulative total variance proportion of each principal component on the horizontal axis minus the number of principal components can be determined, and the rate of change of the slope at each point can be determined. If the slope change is found to be the most significant at a certain point on the curve, it indicates that the contribution rate of the first principal components to the total variance proportion is relatively large, and the feature differences corresponding to the first principal components from largest to smallest eigenvalue are relatively large. In this case, the horizontal coordinate of that point is taken as the target number.

[0111] In this embodiment, by plotting the cumulative total variance of each feature value as a function of the number of principal components, the number of principal components corresponding to the point with the largest change in the slope of the curve is determined as the target number. This can accurately determine the number of target dimensions to be retained after dimensionality reduction, improving the accuracy and universality of candidate file dimensionality reduction. It can achieve the dimensionality reduction effect while avoiding the loss of too much information.

[0112] It should be noted that in this embodiment, the relationship curve of the cumulative value of the proportion of the total variance of the eigenvalues with the number of principal components is drawn, and the target number is determined according to the slope change of the relationship curve as an example. In another optional embodiment, the number of principal components corresponding to the point where the cumulative value of the proportion of the total variance is greater than or equal to the preset cumulative value threshold can also be determined according to the relationship curve of the number of principal components and the cumulative value of the proportion of the total variance of the eigenvalues as the target number. Exemplarily, taking the example that there are 3 preset dimensions in total, and the total variance proportions of each preset dimension are 0.5, 0.3, and 0.2 respectively, the cumulative value of the total variance proportion of the first principal component is 0.5, the cumulative value of the total variance proportion of the first two principal components is 0.5 + 0.3 = 0.8, and the cumulative value of the total variance proportion of the first three principal components is 0.5 + 0.3 + 0.2 = 1.0. If the cumulative value threshold is set to 0.75, then the cumulative value of the total variance proportion of the first two principal components, 0.8, is greater than 0.75, and the target number is 2.

[0113] In yet another optional embodiment, the target number of the corresponding target dimension can also be determined for the preset dimension. For example, for a preset dimension of 512 dimensions, the target number can be set to 64, and for a preset dimension of 1024 dimensions, the target number can be set to 128, etc. The target number can also be set according to the trajectory quality of each target object trajectory in the candidate file. The worse the trajectory quality of each target object trajectory is, the fewer the target numbers can be set. The specific determination method of the target number in this embodiment is not limited.

[0114] S250. Take the preset dimensions corresponding to the first target number of eigenvalues as the target dimensions that meet the feature difference condition.

[0115] After determining the target number, select the first target number of eigenvalues from large to small, and take the preset dimensions corresponding to each eigenvalue as the target dimensions. If the target number is k, k < n, then the preset dimensions corresponding to λ1', λ2', …, λ k ' are the target dimensions.

[0116] Furthermore, in S220, when determining the eigenvalues corresponding to each preset dimension and sorting the eigenvalues corresponding to each preset dimension in descending order, the preset dimensions corresponding to each eigenvalue can be saved as the preset dimension index of each eigenvalue. After determining the first target number of eigenvalues, the target dimensions can be determined according to the preset dimension indexes corresponding to the first target number of eigenvalues.

[0117] It should be noted that this embodiment uses the example of sorting the feature values ​​corresponding to each preset dimension from largest to smallest and selecting the top target number of preset dimensions as target dimensions. Alternatively, the target number can be left unset or not determined, and each preset dimension whose feature value is greater than or equal to the preset feature value threshold can be selected as the target dimension based on the feature values ​​corresponding to each preset dimension and a preset feature value threshold. This embodiment does not limit the specific method for determining the target dimensions that meet the feature difference condition.

[0118] S260. Standardize the feature values ​​corresponding to each target dimension to obtain the standardized feature values ​​corresponding to each target dimension.

[0119] The purpose of standardization is to prevent abnormally large feature values ​​in certain dimensions from affecting the calculation of subsequent target dimension weights.

[0120] For λ1', λ2', ..., λ k Standardization can be achieved by using the average and standard deviation of the feature values ​​corresponding to each target dimension. Similar to the specific process of target object feature standardization described above, the specific process of feature value standardization will not be repeated in this embodiment.

[0121] S270. The ratio of the standardized feature value corresponding to the target dimension to the sum of all standardized feature values ​​is used as the weight of the target dimension.

[0122] The target dimension weight represents the role of the target dimension in explaining the differences between target object trajectories. The larger the target dimension weight, the greater the role of that target dimension in explaining the differences between target object trajectories. At the same time, the greater the difference in target object features between target objects in different target object trajectories along that target dimension.

[0123] Furthermore, after determining the weights of the target dimension, the corresponding preset dimension index and weights can be saved for subsequent feature fusion.

[0124] S280. Determine whether the current dimension of the trajectory of each target object in the candidate file matches the target dimension. If yes, execute S290; otherwise, execute S2100.

[0125] In this embodiment, the current dimension matches the target dimension; that is, there exists a target dimension, and its corresponding preset dimension index is the current dimension. For example, if there are 512 preset dimensions and 64 target dimensions, and the preset dimension index corresponding to a target dimension is 20, then when the 20th preset dimension is processed as the current dimension, it is determined that the current dimension matches the target dimension.

[0126] In this embodiment, if the current dimension matches a target dimension with significant feature differences, then feature fusion is performed on the target object features of the current dimension and the target object features of the target dimension that matches the current dimension. If the current dimension does not match a target dimension with significant feature differences, that is, if the current dimension is a dimension with relatively small feature differences, then the original target object features of the current dimension are retained unchanged.

[0127] In this embodiment, the target object features of dimensions with large feature differences and the target object features of dimensions with small feature differences are processed differently. The target object features of dimensions with large feature differences after dimensionality reduction are fused with the original target object features of that dimension, which further increases the difference of the target object features in that dimension, thereby improving the difference between the trajectories of different target objects after feature fusion.

[0128] S290. Based on the target object features of the current dimension, the target object features of the target dimension that matches the current dimension, and the weights of the target dimensions that match the current dimension, perform target object feature fusion.

[0129] This embodiment uses the weighted summation of the target object features in the current dimension and the target object features in the target dimension that match the current dimension as an example to illustrate the feature fusion method.

[0130] In an optional embodiment, an initial weight can be assigned to the target object features of the current dimension. This initial weight is used when fusing the original target object features of different dimensions with the target object features of the target dimension that matches the current dimension after dimensionality reduction. The sum of the product of the feature value of the target object feature of the current dimension and the initial weight, and the product of the feature value of the target object feature of the target dimension that matches the current dimension and the weight of that target dimension, is used as the feature value of the fused target object feature.

[0131] In another optional embodiment, the weights of the current dimension and the target dimension that matches the current dimension can be set, and the weights of the target object features of the current dimension can be determined based on the preset weights and the weights of the target dimensions that match the current dimension.

[0132] S2100: Retain the characteristics of the target object in the current dimension.

[0133] S2110. After the feature fusion of the target object, the candidate files are re-clustered to obtain the re-clustering result.

[0134] S2120. Based on the re-clustering results, determine whether the candidate files are invalid files.

[0135] The specific process of re-clustering the candidate files after feature fusion of the target object, and determining whether the candidate files are invalid files based on the re-clustering results, has been described in the above embodiments and will not be repeated here.

[0136] The technical solution of this embodiment selects candidate files from each target object file, improving the efficiency of invalid file identification and saving computational resources while ensuring that invalid files are not missed. Based on principal component analysis, the feature values ​​of each preset dimension of the target object trajectory of the candidate files are determined. The feature values ​​are sorted from largest to smallest, and a curve is plotted to show the cumulative ratio of the number of principal components to the total variance of the feature values. The number of targets is determined based on the slope of the curve. The dimensions corresponding to the top number of feature values ​​are selected as target dimensions according to the order of feature values ​​from largest to smallest, which can accurately filter out the dimensions with large differences in target object features between different target object trajectories. The weight of each target dimension is determined according to the feature values ​​of each target dimension. During feature fusion, the target object features of the extracted target dimensions are weighted and fused with the target object features of the original preset dimensions. The original preset dimensions that were not extracted during dimensionality reduction are still retained. Through target feature fusion, the differences between different target object trajectories in dimensions with large differences in target object features are further improved, thereby improving the potential differences between different target object trajectories. The candidate files, whose trajectories show improved diversity, are re-clustered. If the re-clustering results in multiple new files, it indicates that the clustering results of the candidate files are unstable, and the candidate files are invalid. This embodiment's technical solution, starting from the differences between the target object trajectories in the candidate files, has good applicability to different candidate files, accurately identifying invalid files and facilitating their subsequent removal. This makes the in-depth information mining based on the target object files more accurate.

[0137] Example 3

[0138] Figure 3 This is a schematic diagram of a device for determining invalid files according to Embodiment 3 of the present invention. Figure 3 As shown, the device includes:

[0139] The candidate file determination module 310 is used to determine candidate files in each target object file, wherein the candidate file includes at least two target object trajectories, and each target object trajectory corresponds to a target object feature of a preset dimension;

[0140] The target dimension determination module 320 is used to determine at least two target dimensions that satisfy the feature difference condition from the preset dimensions of the trajectory of each target object in the candidate file.

[0141] The target object feature fusion module 330 is used to perform target object feature fusion based on the target object features of each target object trajectory in the candidate file in a preset dimension, and the target object features in the target dimension.

[0142] The invalid file judgment module 340 is used to determine whether a candidate file is invalid based on the candidate files after the feature fusion of the target object.

[0143] The technical solution of this invention involves filtering candidate files from various target object files, selecting target dimensions that meet feature difference conditions for each preset dimension of the target object trajectory of each candidate file, performing feature fusion based on the target object features of each preset dimension of the target object trajectory of each candidate file and the target object features of the target dimension that meets the feature difference conditions, and determining whether the candidate file is invalid based on the feature-fused candidate file. This solves the problems of existing invalid file identification methods, which are prone to misjudgment and omission or waste of computational resources. It not only enables the determination of the validity of target object files and accurately identifies invalid files, but also saves computational resources.

[0144] Based on the above embodiments, optionally, the candidate file determination module 310 includes:

[0145] The target object trajectory quantity determination unit is used to determine the trajectory quality of each target object trajectory if the number of target object trajectories in the target object file is greater than or equal to a preset quantity threshold.

[0146] Trajectory quality is used to represent the recognizability of the target object in the target object trajectory;

[0147] The trajectory quality judgment unit is used to select the target object file as a candidate file if the trajectory quality of each target object trajectory meets the preset trajectory quality conditions.

[0148] Based on the above embodiments, optionally, the target dimension determination module 320 includes:

[0149] The feature value sorting unit is used to determine the feature value corresponding to each preset dimension based on the target object features of each preset dimension of the target object trajectory of each candidate file, and sort the feature values ​​corresponding to each preset dimension in descending order.

[0150] The target quantity determination unit is used to determine the target quantity.

[0151] The target dimension target object feature determination unit is used to take the preset dimension corresponding to the number of feature values ​​of the first target as the target dimension that satisfies the feature difference condition.

[0152] Based on the above embodiments, optionally, the target quantity determination unit is specifically used for:

[0153] Plot the total variance percentage-principal component number curve with the number of principal components as the horizontal axis and the cumulative total variance percentage of the first principal components' eigenvalues ​​in descending order as the vertical axis.

[0154] Based on the cumulative total variance percentage versus principal component count curve, the number of principal components that meet the curve slope condition is determined as the target number.

[0155] Based on the above embodiments, optionally, the target object feature fusion module 330 includes:

[0156] The target object feature fusion unit is used to perform target object feature fusion on the target object features of each preset dimension of the target object trajectory of each candidate file. If it is determined that the current dimension matches the target dimension, the target object features of the current dimension, the target object features of the target dimension that matches the current dimension, and the weight of the target dimension that matches the current dimension.

[0157] The target object feature determination unit is used to retain the target object features of the current dimension if it is determined that the current dimension does not match the target dimension.

[0158] Optionally, based on the above embodiments, the apparatus further includes:

[0159] The feature value standardization module is used to standardize the feature values ​​corresponding to each target dimension to obtain the standardized feature values ​​corresponding to each target dimension.

[0160] The target object feature weight determination module is used to take the ratio of the standardized feature value corresponding to the target dimension to the sum of all standardized feature values ​​as the weight of the target dimension.

[0161] Based on the above embodiments, optionally, the invalid file determination module 340 includes:

[0162] The re-clustering unit is used to re-cluster the candidate files after feature fusion of the target object to obtain the re-clustering result;

[0163] The invalid file determination unit is used to determine whether a candidate file is invalid based on the re-clustering results.

[0164] The invalid file determination device provided in the embodiments of the present invention can execute the invalid file determination method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0165] Example 4

[0166] Figure 4 A schematic diagram of an electronic device 10 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0167] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 may also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0168] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0169] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the method for determining invalid files.

[0170] In some embodiments, the method for determining invalid files may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the method for determining invalid files described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to execute the method for determining invalid files by any other suitable means (e.g., by means of firmware).

[0171] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0172] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0173] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0174] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0175] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0176] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0177] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0178] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for determining invalid files, characterized in that, include: Candidate files are determined from each target object file, wherein each candidate file includes at least two target object trajectories, and each target object trajectory corresponds to a target object feature of a preset dimension. Among the preset dimensions of the trajectory of each target object in the candidate archive, at least two target dimensions that satisfy the feature difference condition are determined. Based on the target object features of each target object trajectory in the candidate files in the preset dimension, and the target object features in the target dimension, target object feature fusion is performed. Based on the candidate files after fusing the features of the target object, determine whether the candidate files are invalid files.

2. The method according to claim 1, characterized in that, Candidate files are identified from each target file, including: If the number of target object trajectories in the target object file is greater than or equal to a preset number threshold, then the trajectory quality of each target object trajectory is determined. Trajectory quality is used to represent the recognizability of the target object in the target object trajectory; If the trajectory quality of each target object's trajectory is determined to meet the preset trajectory quality conditions, then the target object file is selected as a candidate file.

3. The method according to claim 1, characterized in that, Among the preset dimensions of the target object trajectories in the candidate files, at least two target dimensions that satisfy the feature difference condition are determined, including: Based on the target object features of each preset dimension of the trajectory of each target object in the candidate files, determine the feature values ​​corresponding to each preset dimension, and sort the feature values ​​corresponding to each preset dimension in descending order. Determine the target quantity; The preset dimension corresponding to the number of feature values ​​of the first target is used as the target dimension to satisfy the feature difference condition.

4. The method according to claim 3, characterized in that, Determine the target quantity, including: Plot the total variance percentage-principal component number curve with the number of principal components as the horizontal axis and the cumulative total variance percentage of the first principal components' eigenvalues ​​in descending order as the vertical axis. Based on the cumulative total variance percentage versus principal component count curve, the number of principal components that meet the curve slope condition is determined as the target number.

5. The method according to claim 3, characterized in that, Based on the target object features of each target object trajectory in the candidate files, and the target object features of the target dimension, target object feature fusion is performed, including: For each preset dimension of the target object features of the candidate file's target object trajectory, if it is determined that the current dimension matches the target dimension, then target object features are fused based on the target object features of the current dimension, the target object features of the target dimension that matches the current dimension, and the weight of the target dimension that matches the current dimension. If it is determined that the current dimension does not match the target dimension, then the target object features of the current dimension are retained.

6. The method according to claim 5, characterized in that, The process of determining the weights of the target dimension includes: The feature values ​​corresponding to each target dimension are standardized to obtain the standardized feature values ​​corresponding to each target dimension. The ratio of the standardized feature value corresponding to the target dimension to the sum of all standardized feature values ​​is used as the weight of the target dimension.

7. The method according to claim 1, characterized in that, Based on the candidate files after feature fusion of the target object, determine whether the candidate files are invalid files, including: The candidate files after feature fusion of the target object are re-clustered to obtain the re-clustering result; Based on the re-clustering results, determine whether the candidate files are invalid files.

8. A device for determining invalid files, characterized in that, include: The candidate file determination module is used to determine candidate files from each target object file. Each candidate file includes at least two target object trajectories, and each target object trajectory corresponds to a target object feature of a preset dimension. The target dimension determination module is used to determine at least two target dimensions that satisfy the feature difference condition from the preset dimensions of the trajectory of each target object in the candidate file. The target object feature fusion module is used to perform target object feature fusion based on the target object features of each target object trajectory in the candidate file in a preset dimension, as well as the target object features in the target dimension. The invalid file determination module is used to determine whether a candidate file is invalid based on the candidate files after the feature fusion of the target object.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for determining invalid files as described in any one of claims 1-7.

10. A storage medium for storing computer-executable instructions, characterized in that, The computer-executable instructions, when executed by a computer processor, are used to perform the method for determining invalid files as described in any one of claims 1-7.

Images