54results about How to "Reduce the intra-class distance" patented technology

The invention discloses a clustering and reclassifying face recognition method, which comprises the steps of acquiring a training sample; carrying out equalization processing on the training sample; carrying out Gabor texture feature extraction on face images, and acquiring a feature vector corresponding to each face image after feature extraction; carrying out dimension reduction on acquired Gabor texture features of each face image to acquire feature vectors after dimension reduction; carrying out a clustering operation until distance convergence so as to complete clustering; classifying all of the clustered feature vectors to acquire a plurality of subclasses, calculating to determine each vector mean value, and calculating to acquire a within-class distance and an among-class distance; carrying out feature extraction and preprocessing on face images of a target to be recognized, acquiring a feature vector after projection transformation, and calculating the distance between the acquired feature vector and the feature vectors in each subclass sequentially so as to acquire the similarity; and determining identity information of the target to be recognized. The method disclosed by the invention can shorten the among-class distance so as to reduce an error in the acquisition process, and the accuracy of face recognition is improved.

The invention discloses a human face expression recognition method based on Curvelet transform and sparse learning. The method comprises the following steps: 1, inputting a human face expression image, carrying out the preprocessing of the human face expression image, and cutting and obtaining an eye region and a mouth region from the human face expression image after processing; 2, extracting the human face expression features through Curvelet transform, carrying out the Curvelet transform and feature extraction of the human face expression image after preprocessing, the eye region and the mouth region, carrying out the serial fusion of the three features, and obtaining fusion features; 3, carrying out the classification recognition based on the sparse learning, and respectively employing SRC for classification and recognition of the human face Curvelet features and fusion features; or respectively employing FDDL for classification and recognition of the human face Curvelet features and fusion features. The Curvelet transform employed in the method is a multi-scale geometric analysis tool, and can extract the multi-scale and multi-direction features. Meanwhile, the method employs a local region fusion method, and enables the fusion features to be better in imaging representing capability and feature discrimination capability.

The present invention discloses a method and device of vehicle reidentification based on multiple objective function deep learning. The method comprises: employing first-class vehicle training data generated based on a vehicle image data set and a preset Softmax objective function to train the preset network parameters in a deep learning network frame, obtaining an optimal depth learning network frame, employing second-class vehicle training data generated based on the vehicle image data and a preset Triplet Loss ternary objective function to train the optimal network parameters in the depth learning network frame, obtaining a target depth learning network frame, and performing reidentification of the vehicle image to be identified according to the target depth learning network frame. Through adoption of the target depth learning network frame, the image features with high robustness and high reliability can be obtained to facilitate vehicle reidentification and effectively satisfy the requirement of the vehicle reidentification in the monitoring video.

The invention discloses a face recognition method, system and device based on centralized coordination learning, and the method comprises the following steps: obtaining a to-be-recognized face image,carrying out the face detection of the face image, and obtaining a first face image; After alignment processing is carried out on the first face image, a second face image with a preset size is obtained; inputting the second face image into a preset face recognition model based on centralized coordination learning for feature extraction, and obtaining a face feature vector of the second face image; and calculating cosine similarity by combining the face feature vector and a preset face database, and obtaining a face recognition result according to the cosine similarity. According to the invention, a face recognition model based on centralized coordination learning is adopted to carry out feature extraction on the face image, each feature is pulled to an original point and is respectively put into all quadrants, the inter-class distance is larger, the classification efficiency and recognition accuracy of the face are improved, and the method can be widely applied to the technical fieldof face recognition.

The invention belongs to the technical field of emotion recognition, particularly relates to a speech emotion recognition method and system based on ternary loss, and aims at solving the technical problem of accurately recognizing confusable emotion categories. For this purpose, the speech emotion recognition method comprises the following steps: framing the speech data to be measured so as to obtain a speech sequence of the specific length; temporal coding is performed based on the preset emotional temporal coding network and according to the speech sequence so as to obtain the emotional feature vectors corresponding to the speech sequence; and predicting the emotion category corresponding to the emotion feature vectors based on the preset speech emotion classifier and according to multiple preset real emotion categories. According to the speech emotion recognition method, the confusable speech emotion categories can be greatly recognized, and the method can be executed and implemented by the speech emotion recognition system.

The invention relates to a few-sample learning method based on a sample-level attention network. The method comprises the steps of respectively inputting a sample of a support set and a sample of a query set into a feature extraction module, and obtaining a corresponding support set feature vector and a query set feature vector; inputting the support set feature vector corresponding to each classinto a sample-level attention network module to obtain a class prototype of each class; calculating the distance between the query set feature vector and each class prototype to obtain the probabilitydistribution of the class to which the query set feature vector belongs; jointly training the feature extraction module and the sample-level attention network module by adopting cross entropy loss ofthe query set and classification loss of the support set, and obtaining a gradient updating network through back propagation. According to the method, a new target task can be solved only by learninga large number of similar learning tasks, and for the target task, model updating is not needed any more.

The invention discloses a multi-view three-dimensional model retrieval method and a multi-view three-dimensional model retrieval system based on pairing depth feature learning. The multi-view three-dimensional model retrieval method comprises the steps of: acquiring two-dimensional views of a to-be-retrieved three-dimensional model at different angles, and extracting an initial view descriptor ofeach two-dimensional view; aggregating the plurality of initial view descriptors to obtain a final view descriptor; extracting potential features and category features of the final view descriptor respectively; performing weighted combination on the potential features and the category features to form a shape descriptor; and performing similarity calculation on the obtained shape descriptor and ashape descriptor of the three-dimensional model in a database to realize retrieval of the multi-view three-dimensional model. According to the multi-view three-dimensional model retrieval method, a multi-view three-dimensional model retrieval framework GPDFL is provided, potential features and category features of the model are fused, and the feature recognition capability and the model retrievalperformance can be improved.

The invention provides a vehicle re-identification method, a device and a system, which relate to the technical field of vehicle identification. The method comprises the following steps: obtaining a group of vehicle pictures to be identified; for each vehicle picture, the vehicle direction feature and the vehicle feature vector being extracted from the vehicle picture respectively; merging the directional feature and the vehicle feature vector into the fusion feature information of the vehicle; the fusion feature information of each vehicle being re-recognized, and the same vehicle being determined. The vehicle weight identification method, device and system of the invention greatly improve the accuracy of the vehicle weight identification and further improve the identification efficiencyby fusing the directional features of the vehicle and the vehicle features.

The invention discloses a three-dimensional face recognition method based on Bayesian multivariate distribution characteristic extraction. The method comprises three-dimensional data preprocessing, characteristic extraction and identification classification. The method has the advantages that the defect of large computation amount in the prior art is overcome; a three-dimensional face depth map is used for identifying, so that the computation amount can be reduced, and the identification efficiency is increased; the problem of insufficient training samples in single sample identification can be solved, and the training samples are added with a partitioning method; a characteristic extraction method based on Bayesian analysis is provided on the basis, so that the obtained characteristics have a minimum intra-class distance and a maximum inter-class distance, namely, optimal separability is realized; moreover, a classification method based on Mahalanobis distance is adopted, so that optimal identification classification is realized. As proved by experiment data, the method disclosed by the invention has a good three-dimensional face recognition result.

The invention relates to a network traffic classification system and method based on depth discrimination features. The network traffic classification system comprises a preprocessing module and a model learning module, wherein the preprocessing module uses network flows with different lengths generated by different applications as input, and each network flow is expressed as a flow matrix with afixed size so as to meet the input format requirement of a convolutional neural network (CNN); and the model learning module trains the deep convolutional neural network by taking the flow matrix obtained by the preprocessing module as input under the supervision of a target function formed by a metric learning regularization item and a cross entropy loss item, so that the neural network can learnthe input flow matrix to obtain more discriminative feature representation, and the classification result is more accurate.

The invention discloses an image similarity calculation method based on an improved Soft-Max loss function. An activating function of the improved Soft-Max layer in an image identification network isan improved Soft-Max activation function. In a reverse propagation process, the improved Soft-Max loss function is utilized for updating a network weight. Compared with a traditional Soft-Max loss function, the improved Soft-Max loss function has an advantage of increasing a decision edge which is obtained through learning of an image identification network. In a testing period, a trained image identification model is utilized for performing characteristic vector extraction on two testing images, and cosine similarity between the characteristic vector is calculated. Compared with an image similarity threshold, if the cosine similarity is higher than or equal with the image similarity threshold, a fact that the two images are in the same kind is determined; and if the cosine similarity is lower than the image similarity threshold, a fact that the two images are in different kinds is determined.

The invention discloses a small-sample face recognition method combining sparse representation and a convolutional neural network, and the method comprises the following steps: firstly carrying out the preprocessing of a face image, carrying out the face alignment and five-sense-organ positioning according to face key points, and cutting the face image into four local regions; extacting local features and overall features with higher discrimination by using a convolutional neural network, and constructing a block feature dictionary in combination with a sparse representation algorithm, so as to achieve the effect of sample enhancement; adding sparse representation constraints and cosine similarity to redefine a loss function of the convolutional neural network so as to reduce the intra-class distance between the features and expand the inter-class distance; and finally, carrying out face recognition by adopting enhanced sparse representation classification. The method is high in recognition performance, and has certain robustness for shielding changes in a small sample problem.

The invention relates to a non-contact palm vein recognition method based on an improved residual network. The non-contact palm vein recognition method comprises the following steps: 1) collecting twoinfrared images of the same palm; 2) positioning an ROI area of a palm; 3) carrying out palm vein registration; 4) carrying out palm vein verification; and 5) carrying out palm vein verification judgment: setting an identification threshold T, calculating the distance between the feature vector and the registration template, judging that palm vein verification succeeds if the distance is smallerthan the set identification threshold T, and otherwise, judging that verification fails. According to the improved residual network structure provided by the invention, texture features of different scales of an input sample can be extracted during training, so that the trained model has the capability of extracting multi-scale texture information, and the extracted feature vector better expressesthe information of the input sample; and various scaling and translation problems of an input sample can be solved to a certain extent.

The invention discloses a pedestrian re-identification method based on local high-frequency features and hybrid metric learning. A local frequency feature representation method is adopted, a color andtexture feature set of a target image is extracted under illumination and visual angle change conditions, a sliding window is adopted to describe local details and extract image features, the maximumvalue of features appearing locally at high frequency is taken as a feature value, and multi-scale feature descriptor cascading is obtained. After subspace dimension reduction, weight coefficients ofa metric learning matrix and a hybrid metric learning matrix thereof are obtained according to the posterior probability of sample occurrence, and finally a similarity degree as a basis for pedestrian re-identification is obtained. According to the invention, the pedestrian appearance features can be used to identify images related to the given pedestrian in a multi-camera monitoring scene, and the method has good application values in the fields of intelligent monitoring, intelligent security, criminal investigation, pedestrian retrieval, pedestrian tracking, behavior analysis and the like.

The invention discloses a deep hash method based on metric learning, relates to the field of computer vision and image processing, and solves the problems that a comparison loss function of an existing deep hash method can only enable feature vectors of images of the same category before quantization to be close as much as possible, but cannot encourage the same symbol; the values of different types of images before quantization are far away as far as possible, but the symbols cannot be encouraged to be opposite; finally, the quantized hash code is poor in discriminability, and misjudgment andother problems are caused. According to the invention, a hash comparison loss function is constructed; sign bit constraint is carried out on the real numerical value feature vector before quantization, so that the hash code of the representative image obtained after the real numerical value feature vector before quantization is quantized by a sign function is more accurate, and the sign is constrained through two control functions of fsim (fi.fj) and fdiff (fi.fj), other parts in the expression are used for enabling the feature values of the same category of images to be close and the featurevalues of different categories of images to be far. According to the method, the classification precision is effectively improved, and the misjudgment rate is reduced.

The invention provides a face recognition method. The face recognition method comprises the following steps: S1, obtaining face training test resources from a public data set; S2, preprocessing the image; S3, inputting the processed face data set into a convolutional neural network of a novel network structure of an improved NVM module to obtain a trained lightweight convolutional neural network;S4, inputting the face images in the test set into the trained lightweight convolutional neural network model, and judging whether the lightweight model can be accurately verified to effectively classify the face data set or not.

The invention provides a DNN group gesture identification method based on optimized gesture database distribution, wherein the method belongs to the field of a computer. The method comprises the steps of (1), acquiring gestures and forming a gesture data set; (2), re-classifying the gesture set by means of optimized gesture database distribution for obtaining a plurality of sub-databases; (3), performing learning training on a DNN model on each sub-database, and obtaining DNN structures; (4), inputting a to-be-identified gesture, acquiring the to-be-identified gesture by means of Kinect equipment, realizing gesture segmentation by means of a background subtracting method, and separating human hands from a background; (5), respectively transmitting the to-be-identified gesture to each DNN structure for identification, and calculating an output error E of each DNN identification result by means of a formula of E=(anticipated output)-(network response); and (6), returning an output result which corresponds with a least output error E, and determining the output result as the identified gesture.

The invention provides an end-to-end text-independent voiceprint recognition method and system. The method comprises the following steps of capturing important narrow-band loudspeaker characteristics of an original voice sample by designing a filter based on a Sinc function; then, using a time delay neural network (TDNN) and a gated loop unit (GRU) for generating a hybrid neural network structure of complementary speaker information of different levels, adopting a multi-level pooling strategy, adding an attention mechanism to a pooling layer, and extracting feature information which most represents the frame level and the utterance level of a speaker from a time delay neural network layer and a gated loop unit layer, performing regularization processing on the speaker vector extraction layer; carrying out training through an AM-softmax loss function; and finally, realizing an end-to-end text-independent voiceprint recognition process through similarity calculation of the embedded model and the recognition model. Therefore, the accuracy and applicability of end-to-end text-independent voiceprint recognition are improved.

The embodiment of the invention provides a fine-grained image classification model processing method, and relates to the technical field of image classification. The fine-grained image classificationmodel processing method comprises the steps of generating first data comprising at least two samples according to an obtained data set; processing sample sorting of the first data according to a preset rule to obtain at least two groups of second data; extracting a sample of the second data according to a preset rule to form a sample pair; processing the sample pair through a preset Euler model toobtain a data loss value; and processing a preset first fine-grained image classification model according to the data loss value to obtain a second fine-grained image classification model. Accordingto the invention, the class spacing can be reduced and the intra-class distance can be increased, so that the fine-grained image classification model can improve the fine-grained image classificationperformance.