CN-120375439-B - Face recognition method and device based on biological characteristics and space-time fusion

Abstract

The application discloses a face recognition method and device based on biological characteristics and space-time fusion. The face recognition method based on the biological characteristics and the space-time fusion comprises the steps of obtaining PPG signals and face image information through a camera device, obtaining access network environment information, obtaining a trained face recognition model, and inputting the PPG signals, the face image information and the access network environment information into the face recognition model so as to obtain a recognition result. The application provides dynamic fusion of various biological characteristics (such as face recognition, PPG signals and the like) so as to improve the accuracy and the safety of identity authentication. By dynamically monitoring and analyzing the changes of different biological characteristics, the falsification and the falsification of static biological characteristic data can be effectively prevented.

Inventors

• AN HONG

Assignees

• 国政通科技有限公司
• 国政信通(北京)科技有限公司

Dates

Publication Date

20260512

Application Date

20250402

Claims (5)

1. 1. The face recognition method based on the fusion of the biological characteristics and the space time is characterized by comprising the following steps of: obtaining PPG signals and face image information through a camera device; Acquiring access network environment information; acquiring a trained face recognition model; Inputting the PPG signal, the face image information and the access network environment information into the face recognition model so as to obtain a recognition result; the face recognition model is a CNN-converter mixed architecture face recognition model; The CNN-converter mixed architecture face recognition model comprises the following components: The input layer is used for acquiring the PPG signal, the face image information and the access network environment information; The CNN feature extraction module is used for extracting face image features of the face image information; The LSTM sequence processing module is used for acquiring a PPG characteristic according to the PPG signal and acquiring a network environment information characteristic according to access network environment information; The transducer global dependence modeling module is used for receiving the face image characteristics, the PPG characteristics and the network environment information characteristics, acquiring global dependence through a multi-head self-attention mechanism and generating global characteristics; The feature fusion module is used for fusing the face image features, the PPG features, the network environment information features and the global features to obtain fusion features; and the classification module is used for acquiring the fusion characteristics and acquiring the identification result according to the fusion characteristics.
2. 2. The face recognition method based on the fusion of biological characteristics and space time according to claim 1, wherein the face recognition method based on the fusion of biological characteristics and space time further comprises: And training the CNN-converter mixed architecture face recognition model.
3. 3. The method for face recognition based on biometric and spatiotemporal fusion of claim 2, wherein the training the CNN-transform hybrid architecture face recognition model comprises: acquiring training data; acquiring the CNN-converter mixed architecture face recognition model; And training the CNN-converter mixed architecture face recognition model through training data.
4. 4. The face recognition method based on the fusion of biological characteristics and space time as claimed in claim 3, wherein the CNN-fransformer hybrid architecture face recognition model adopts the following loss function: Wherein, the method comprises the steps of, L represents a loss function, x represents a feature vector output by a model, is a high-dimensional vector, and comprises features extracted from input data, y represents a real label, C represents the number of categories, Representing the ith element in the true tag vector y, if the sample belongs to the ith class =1, Otherwise =0、 A weight vector representing class i, Representing weight vectors Dot product with feature vector x for measuring similarity between them, Representing weight vectors L2 norm of (2), An L2 norm representing the feature vector x, Representing a positive number less than 10, for preventing the denominator from being zero, ensuring the stability of calculation, Representing a positive coefficient for controlling the influence of the adaptive weight term on the overall loss function, As a constraint, ensure that only the class of non-authentic tags is considered, if the sample belongs to the ith class =0, Otherwise =1、 The square of the L2 norm of the weight vector W, i.e. the sum of squares of all weight elements, β, represents a positive coefficient.
5. 5. The utility model provides a face identification device based on biological feature and space-time fusion, its characterized in that, face identification device based on biological feature and space-time fusion includes: The information acquisition module is used for acquiring face image information, PPG signals and access network environment information transmitted by the camera device; The face recognition model acquisition module is used for acquiring a trained face recognition model; the recognition result acquisition module is used for inputting the PPG signal, the face image information and the access network environment information into the face recognition model so as to acquire a recognition result; the face recognition model is a CNN-converter mixed architecture face recognition model; The CNN-converter mixed architecture face recognition model comprises the following components: The input layer is used for acquiring the PPG signal, the face image information and the access network environment information; The CNN feature extraction module is used for extracting face image features of the face image information; The LSTM sequence processing module is used for acquiring a PPG characteristic according to the PPG signal and acquiring a network environment information characteristic according to access network environment information; The transducer global dependence modeling module is used for receiving the face image characteristics, the PPG characteristics and the network environment information characteristics, acquiring global dependence through a multi-head self-attention mechanism and generating global characteristics; The feature fusion module is used for fusing the face image features, the PPG features, the network environment information features and the global features to obtain fusion features; and the classification module is used for acquiring the fusion characteristics and acquiring the identification result according to the fusion characteristics.

Description

Face recognition method and device based on biological characteristics and space-time fusion Technical Field The application relates to the technical field of image recognition, in particular to a face recognition method based on biological characteristics and space-time fusion and a face recognition device based on the biological characteristics and the space-time fusion. Background In the prior art, for face recognition, image processing and machine learning algorithms are mainly relied on. The technical scheme generally comprises the following steps: face image acquisition, namely acquiring face images through equipment such as a camera and the like. Preprocessing, namely performing preprocessing operations such as graying, denoising, normalization and the like on the acquired face image so as to improve the accuracy of subsequent processing. And extracting information capable of representing individual features, such as facial contours, textures, key points and the like from the preprocessed face image by utilizing an algorithm. And (3) feature matching and recognition, namely comparing the extracted face features with feature templates in a database, calculating similarity scores, and deciding according to the scores to determine identities. The disadvantages of the prior art are: In complex environments (such as light changes, occlusion, angle changes, etc.), the accuracy of face recognition can be greatly affected due to the fact that only single-mode image information is relied on. Unfused dynamic multi-modal biological features (e.g., facial expressions, head gestures, etc.) and spatiotemporal context information (e.g., changes between successive frames, changes in the position of a face in an image, etc.), result in an identification system that is not robust enough in the face of a complex scene. Disclosure of Invention The invention aims to provide a face recognition method based on biological characteristics and space-time fusion to at least solve one technical problem. In one aspect of the invention, a face recognition method based on biological characteristics and space-time fusion is provided, and the face recognition method based on biological characteristics and space-time fusion comprises the following steps: obtaining PPG signals and face image information through a camera device; Acquiring access network environment information; acquiring a trained face recognition model; And inputting the PPG signal, the face image information and the access network environment information into the face recognition model so as to obtain a recognition result. Optionally, the face recognition model is a CNN-transducer mixed architecture face recognition model. Optionally, the CNN-transducer hybrid architecture face recognition model includes: The input layer is used for acquiring the PPG signal, the face image information and the access network environment information; The CNN feature extraction module is used for extracting face image features of the face image information; The LSTM sequence processing module is used for acquiring a PPG characteristic according to the PPG signal and acquiring a network environment information characteristic according to access network environment information; The transducer global dependence modeling module is used for receiving the face image characteristics, the PPG characteristics and the network environment information characteristics, acquiring global dependence through a multi-head self-attention mechanism and generating global characteristics; The feature fusion module is used for fusing the face image features, the PPG features, the network environment information features and the global features to obtain fusion features; and the classification module is used for acquiring the fusion characteristics and acquiring the identification result according to the fusion characteristics. Optionally, the face recognition method based on the fusion of biological characteristics and space time further comprises the following steps: And training the CNN-converter mixed architecture face recognition model. Optionally, the training the CNN-transducer hybrid architecture face recognition model includes: acquiring training data; acquiring the CNN-converter mixed architecture face recognition model; And training the CNN-converter mixed architecture face recognition model through training data. Optionally, the CNN-transducer hybrid architecture face recognition model employs the following loss function: Wherein, the L represents a loss function, x represents a feature vector output by a model, and is usually a high-dimensional vector, and the feature vector comprises features extracted from input data, y represents a real label, C represents the number of categories, y i represents the ith element in the real label vector y, y i =1 if a sample belongs to the ith category, and y i＝0、Wi represents a weight vector of the category i,The dot product of the weight vector W i and the fe