CN-121981695-A - Attendance checking identification management method, system, readable storage medium and computer

CN121981695ACN 121981695 ACN121981695 ACN 121981695ACN-121981695-A

Abstract

The invention provides an attendance checking identification management method, an attendance checking identification management system, a readable storage medium and a computer, wherein the attendance checking identification management method comprises the steps of respectively carrying out data processing on facial video stream data and phrase audio data of staff to obtain a key frame image sequence and an audio spectrogram; introducing a multiscale spatial attention mechanism and a competitive multiscale fusion algorithm into an expression recognition network model to obtain an expression recognition optimization model, carrying out multiscale training on an audio filter, optimizing the audio filter after multiscale training through a weighted fusion strategy to obtain an audio recognition optimization model, carrying out self-adaptive weighted fusion on the expression recognition optimization model and the audio recognition optimization model to obtain an attendance recognition model, introducing an attendance rule into the attendance recognition model to obtain an attendance management model, and respectively inputting a key frame image sequence and an audio spectrogram into the attendance management model to realize attendance management of staff. The invention supports various attendance systems and realizes personalized and intelligent attendance management.

Inventors

Kuang Lingling
Fan Qihuan
TAN XIAOBIN
LI ZHUO
WANG YAXIONG
ZHANG CHAO
LI KAI
ZHANG HAOREN
ZHOU ZIHUA
JIANG YING

Assignees

江西江铃集团新能源汽车有限公司

Dates

Publication Date: 20260505
Application Date: 20260409

Claims (10)

1. The attendance identification management method is characterized by comprising the following steps of: When an employee approaches an attendance point, synchronously acquiring face video stream data and phrase audio data of the employee, and respectively carrying out data processing on the face video stream data and the phrase audio data to obtain a corresponding key frame image sequence and an audio spectrogram; constructing an expression recognition network model, and introducing a multiscale spatial attention mechanism and a competitive multiscale fusion algorithm into the expression recognition network model to obtain an expression recognition optimization model; Constructing an audio filter, performing multi-scale training on the audio filter, and optimizing the audio filter after multi-scale training through a weighted fusion strategy to obtain an audio recognition optimization model; Performing self-adaptive weighted fusion on the expression recognition optimization model and the audio recognition optimization model to obtain an attendance recognition model, and introducing a preset attendance rule into the attendance recognition model to obtain an attendance management model; and respectively inputting the key frame image sequence and the audio spectrogram into the attendance management model to realize the attendance management of the staff.
2. The attendance recognition management method according to claim 1, wherein the step of performing data processing on the face video stream data and the phrase audio data to obtain a corresponding key frame image sequence and audio spectrogram respectively includes: Performing face detection and key point positioning on the face video stream data by utilizing a multitask convolutional neural network, performing face alignment according to key points, and performing normalization processing on all face areas to obtain preliminary processing data; performing data enhancement processing on the preliminary processing data to obtain a corresponding key frame image sequence; Pre-emphasis processing is carried out on the phrase voice audio data, and framing is carried out according to a preset framing rule so as to obtain corresponding framing audio data; And carrying out Hamming window processing on the framing audio data, and converting the preprocessed time domain signals into standardized spectrograms to obtain the audio spectrograms.
3. The attendance recognition management method according to claim 1, wherein the steps of constructing an expression recognition network model, and introducing a multiscale spatial attention mechanism and a competitive multiscale fusion algorithm into the expression recognition network model to obtain an expression recognition optimization model comprise: Constructing an expression recognition network model, and optimizing the attention mechanism of the expression recognition network model by utilizing a multi-scale space attention mechanism so as to generate space attention weights with different scales; fusing the spatial attention weights of different scales by using a competitive multi-scale fusion algorithm to obtain an expression recognition optimization model, wherein the loss function of the expression recognition optimization model is as follows: In the formula, Representing a standard cross-entropy loss, Represents the balance coefficient of the balance-wheel, Indicating a loss of focus and, The expression class number is represented by the expression class number, Represent the first The weight of the individual target class is determined, Representation model pair number The predicted probability of the individual target class(s), Representing the focus parameter.
4. The attendance recognition management method according to claim 1, wherein the steps of constructing an audio filter, performing multi-scale training on the audio filter, and optimizing the multi-scale trained audio filter by a weighted fusion strategy to obtain an audio recognition optimization model comprise: constructing a plurality of audio filters based on the size of a window of a scale, and dynamically optimizing each audio filter by utilizing a gradient descent algorithm; And fusing the dynamically optimized audio filters through a weighted fusion strategy, and adaptively calculating the weight of each audio filter according to the variance of each scale feature to obtain an audio recognition optimization model.
5. The attendance identification management method as claimed in claim 1, wherein the step of introducing a preset attendance rule into the attendance identification model to obtain an attendance management model comprises: Generating a corresponding rule template library according to typesetting data of staff, and combining a corresponding attendance system with the rule template library to generate a corresponding attendance rule, wherein the attendance rule comprises a scheduling rule, and the scheduling rule comprises a fixed scheduling system, a periodic scheduling system and an elastic working system; And introducing the attendance rule into the attendance identification model to obtain a corresponding attendance management model.
6. An attendance identification management system, comprising: The data acquisition module is used for synchronously acquiring face video stream data and phrase audio data when the staff approaches to the attendance point, and respectively carrying out data processing on the face video stream data and the phrase audio data so as to obtain a corresponding key frame image sequence and an audio spectrogram; the first model construction module is used for constructing an expression recognition network model, and introducing a multi-scale space attention mechanism and a competitive multi-scale fusion algorithm into the expression recognition network model to obtain an expression recognition optimization model; The second model construction module is used for constructing an audio filter, performing multi-scale training on the audio filter, and optimizing the audio filter after multi-scale training through a weighted fusion strategy to obtain an audio recognition optimization model; The model optimization module is used for carrying out self-adaptive weighted fusion on the expression recognition optimization model and the audio recognition optimization model to obtain an attendance recognition model, and introducing a preset attendance rule into the attendance recognition model to obtain an attendance management model; And the attendance management module is used for respectively inputting the key frame image sequence and the audio spectrogram into the attendance management model so as to realize the attendance management of the staff.
7. The attendance identification management system according to claim 6, wherein the data acquisition module is specifically configured to: Performing face detection and key point positioning on the face video stream data by utilizing a multitask convolutional neural network, performing face alignment according to key points, and performing normalization processing on all face areas to obtain preliminary processing data; performing data enhancement processing on the preliminary processing data to obtain a corresponding key frame image sequence; Pre-emphasis processing is carried out on the phrase voice audio data, and framing is carried out according to a preset framing rule so as to obtain corresponding framing audio data; And carrying out Hamming window processing on the framing audio data, and converting the preprocessed time domain signals into standardized spectrograms to obtain the audio spectrograms.
8. The attendance identification management system of claim 6, wherein the first model building module is specifically configured to: Constructing an expression recognition network model, and optimizing the attention mechanism of the expression recognition network model by utilizing a multi-scale space attention mechanism so as to generate space attention weights with different scales; fusing the spatial attention weights of different scales by using a competitive multi-scale fusion algorithm to obtain an expression recognition optimization model, wherein the loss function of the expression recognition optimization model is as follows: In the formula, Representing a standard cross-entropy loss, Represents the balance coefficient of the balance-wheel, Indicating a loss of focus and, The expression class number is represented by the expression class number, Represent the first The weight of the individual target class is determined, Representation model pair number The predicted probability of the individual target class(s), Representing the focus parameter.
9. A readable storage medium having stored thereon a computer program, wherein the program when executed by a processor implements the attendance identification management method as claimed in any one of claims 1 to 5.
10. A computer comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the attendance recognition management method as claimed in any one of claims 1 to 5 when executing the computer program.

Description

Attendance checking identification management method, system, readable storage medium and computer Technical Field The present invention relates to the field of data processing technologies, and in particular, to an attendance checking identification management method, an attendance checking identification management system, a readable storage medium and a computer. Background Along with the improvement of enterprise management informatization degree, attendance management is taken as an important component of human resource management, and the accuracy and the instantaneity of the attendance management directly influence the operation efficiency and the management level of enterprises. The traditional attendance management such as a card punching machine, fingerprint identification, an IC card and the like has the problems of easy card punching, low identification rate, large environmental influence and the like. In recent years, with the development of face recognition technology, the advantages of non-contact, strong convenience and the like are paid attention to the aspect of attendance management. However, the existing attendance checking and identifying method with the face identification technology has obviously reduced identifying performance under the conditions of illumination change, shielding, expression change and the like, is easy to be attacked by counterfeiting of photos, videos and the like, lacks flexible rule configuration capability, and is difficult to adapt to various scheduling systems of enterprises. Disclosure of Invention Accordingly, an object of the present invention is to provide an attendance checking identification management method, system, readable storage medium and computer, which at least solve the above-mentioned shortcomings. The invention provides an attendance checking identification management method, which comprises the following steps: When an employee approaches an attendance point, synchronously acquiring face video stream data and phrase audio data of the employee, and respectively carrying out data processing on the face video stream data and the phrase audio data to obtain a corresponding key frame image sequence and an audio spectrogram; constructing an expression recognition network model, and introducing a multiscale spatial attention mechanism and a competitive multiscale fusion algorithm into the expression recognition network model to obtain an expression recognition optimization model; Constructing an audio filter, performing multi-scale training on the audio filter, and optimizing the audio filter after multi-scale training through a weighted fusion strategy to obtain an audio recognition optimization model; Performing self-adaptive weighted fusion on the expression recognition optimization model and the audio recognition optimization model to obtain an attendance recognition model, and introducing a preset attendance rule into the attendance recognition model to obtain an attendance management model; and respectively inputting the key frame image sequence and the audio spectrogram into the attendance management model to realize the attendance management of the staff. Further, the step of performing data processing on the face video stream data and the phrase audio data to obtain a corresponding key frame image sequence and audio spectrogram includes: Performing face detection and key point positioning on the face video stream data by utilizing a multitask convolutional neural network, performing face alignment according to key points, and performing normalization processing on all face areas to obtain preliminary processing data; performing data enhancement processing on the preliminary processing data to obtain a corresponding key frame image sequence; Pre-emphasis processing is carried out on the phrase voice audio data, and framing is carried out according to a preset framing rule so as to obtain corresponding framing audio data; And carrying out Hamming window processing on the framing audio data, and converting the preprocessed time domain signals into standardized spectrograms to obtain the audio spectrograms. Further, the steps of constructing an expression recognition network model, and introducing a multiscale spatial attention mechanism and a competitive multiscale fusion algorithm into the expression recognition network model to obtain an expression recognition optimization model comprise: Constructing an expression recognition network model, and optimizing the attention mechanism of the expression recognition network model by utilizing a multi-scale space attention mechanism so as to generate space attention weights with different scales; fusing the spatial attention weights of different scales by using a competitive multi-scale fusion algorithm to obtain an expression recognition optimization model, wherein the loss function of the expression recognition optimization model is as follows: In the formula, Representing a standard cross-entro