51 results about "Triplet loss" patented technology

The invention discloses a cross-view image geographic positioning method based on dynamic threshold pseudo tag self-training learning, and the method specifically comprises the following steps: introducing a difficult sample feature mining method, dynamically adjusting the loss weight of a sample according to the change of similarity, and building a dynamic difficult sample triple loss model; the method comprises the following steps: dynamically adjusting a confidence threshold value of a sample by adopting an index moving average weighting method, iteratively training and screening an unlabeled sample, namely a pseudo label, establishing a pseudo label self-training mechanism of a dynamic threshold value, mining and utilizing non-paired data, and solving the problem of high manual labeling cost; a reference image most similar to a query image is found through image retrieval, and the offset of a query position is predicted. Experiments on CVUSA and CVACT data sets show that as the distance threshold increases, the accuracy of the cross-view image geographic positioning method based on dynamic threshold pseudo tag self-training learning presents a stable rising trend, and the cross-view image geographic positioning method based on dynamic threshold pseudo tag self-training learning is superior to other methods under the same threshold condition.

The invention provides a retrieval enhancement generation method and system in the dual-carbon field, and relates to the field of data processing. According to the method, multi-source unstructured data in the dual-carbon field is collected, after data preprocessing is carried out, a multi-granularity query problem set is formed, a dual-carbon field knowledge base is obtained, and a dual-carbon field-oriented embedding model CEMBING and a reordering model CReranker are constructed. The CEMBING model adopts a semantic partitioning method and a joint training strategy, so that semantic information of a dual-carbon field text can be effectively captured; the CReranker model adopts a negative example mining strategy and a triple loss function, candidate documents can be accurately sorted, the problems of knowledge limitation and insufficient timeliness of LLMs in the application of the dialogue system in the dual-carbon field are effectively solved, and the accuracy and efficiency of retrieval enhancement generation of the dialogue system in the dual-carbon field are improved.

The invention discloses a motor bearing fault detection system and method based on robust deep learning, and belongs to the technical field of mechanical fault detection and intelligent perception. Feature extraction is carried out on an original vibration signal with a label based on a supervised learning branch network, and the original vibration signal is used as a reference sample; the samples with the same fault category and different fault categories as the reference samples are positive samples and negative samples, and inter-class separation and intra-class aggregation relations in a triple loss optimization embedding feature space are introduced to generate embedding representation; based on an unsupervised learning branch network, encoding the original vibration signal after time domain and frequency domain artificial feature extraction, and introducing triple loss to carry out unsupervised embedding learning to generate high-level feature embedding representation; and the embedded representations output by the two branch networks are fused, dual loss of triple loss and center loss is introduced for training, and a bearing fault detection model after training is completed is used for bearing fault detection.

The invention discloses a satellite signal authentication method and device based on a complex valued neural network and a storage medium. The satellite signal authentication method comprises the following steps: acquiring a complex valued data sequence of a satellite signal to be authenticated; inputting the complex-valued data sequence of the to-be-authenticated satellite signal into a trained complex-valued neural network model to enable the trained complex-valued neural network model to output the radio frequency feature vector of the to-be-authenticated satellite signal, the trained complex-valued neural network model being obtained by training based on a preset triple loss function and a back propagation algorithm; calculating a target similarity score between the radio frequency feature vector of the satellite signal to be authenticated and the corresponding target anchor point sample, wherein the score is used for representing an average angular distance between the radio frequency feature vector and the target anchor point sample; and classifying and authenticating the satellite signal to be authenticated based on a preset score threshold and the target similarity score. According to the method, the recognition complexity can be reduced while the recognition precision of the satellite signal is improved.

The invention belongs to the technical field of medical signal processing and deep learning, and particularly relates to a ballistocardiogram signal data enhancement method and system based on BiLSTM-GAN, and the method comprises the steps: S1, carrying out the Butterworth band-pass filtering of a preset frequency band on an original ballistocardiogram signal, and outputting a first signal; s2, performing down-sampling processing on the first signal to a preset sampling rate, and outputting a second signal; s3, segmenting the second signal into segments with a preset length, labeling categories, and outputting BCG signal segments subjected to band-pass filtering; s4, constructing a BiLSTM-GAN model on the basis of the BCG signal fragments subjected to the band-pass filtering; s5, taking the BCG signal segments subjected to band-pass filtering as training data, and training the BiLSTM-GAN model based on a triple loss function including reconstruction loss, supervision loss and adversarial loss; s6, generating an enhanced BCG signal fragment; and S7, generating equivalent enhanced BCG signal fragments, and outputting a balanced data set. The technical problem of sample scarcity caused by class imbalance in medical BCG signals can be solved.

Systems and methods are disclosed for image classification of electro-optical images and synthetic aperture radar images using training techniques that can include appearance labeling and triplet mining to train a neural network system. The training data can include image pairs of electro-optical images and synthetic radar aperture images. The training data can include anchor, positive, and negative images. The neural network can be trained using triplet loss and cross-entropy loss. The trained neural network can be used for object classification such as automatic target recognition of aerial images.

The invention discloses a rolling bearing fault diagnosis method and system based on comparison decoupling single-source domain generalization, relates to the field of rolling bearing fault diagnosis, and aims to solve the problem that in the prior art, a method for effectively solving the problem that the model generalization performance is reduced due to insufficient diversity of single-source data in bearing fault diagnosis does not exist. The method is technically characterized by comprising the following steps: step 1, adopting rolling bearing time domain data, and performing two data enhancement processing on the rolling bearing time domain data to obtain two groups of enhanced data; step 2, the model construction is based on a causal decoupling network and a comparative learning BYOL framework, the causal decoupling network is used for extracting causal features and non-causal features of data, and the rank of the causal features is constrained through adaptive threshold weighted nuclear norm regularization; the contrast learning BYOL framework is used for pulling in causal features extracted from the two groups of enhanced data to obtain causal domain invariant features of each sample; step 3, using a prototype Gaussian triple loss function to constrain causal domain invariant features, and improving intra-class compactness and inter-class separability of training samples; and step 4, finally inputting a target domain sample to the trained comparison decoupling network model to realize diagnosis.

Aspects of the subject disclosure may include, for example, generating a natural language processing model by training an automatic speech recognition (ASR) encoder with manual transcription. The training is performed by correcting and adjusting relevant factors of the ASR encoder based on determined triplet loss, classification loss and Kullback-Leibler divergence loss. In response to an ASR utterance, the trained natural language processing model generates a predicted intent associated with the ASR utterance with improved accuracy. Other embodiments are disclosed.

The application discloses a multi-source domain adaptation EEG emotion state classification method based on knowledge distillation. First, data is acquired for band pass filtering, and independent component analysis technology is used to remove artifacts. Second, the electroencephalogram feature is extracted through the differential entropy method, and the three-dimensional electroencephalogram time sequence is converted into a two-dimensional sample matrix. Then, the training set and the test set are respectively demarcated under two task scenarios, and it is ensured that they do not coincide. The application adopts the pseudo-label triplet loss based on marginal sampling combined with the maximum mean difference. The application learns knowledge from different source domains to maximize the use of multiple single-source models and realize a more powerful model with less time consumption. Finally, the classification accuracy is used to evaluate the performance of the model under two task scenarios. The application combines the triplet loss and the maximum mean difference, which can not only realize unbiased alignment between each pair of source domains and the target domain at the domain level, but also consider the correlation at the data pair level.

This invention relates to a small-sample fault diagnosis method based on multi-scale feature learning and domain adaptive optimization, comprising: acquiring historical heterogeneous fault data of several mechanical devices as a source domain dataset, acquiring historical heterogeneous fault data of a target mechanical device as a target domain dataset, and defining label spaces for the source and target domain datasets respectively; randomly sampling and combining the source and target domain datasets to construct a triplet dataset and a domain adaptive dataset, inputting them into a dual-branch feature extraction subnetwork, measuring the triplet feature embedding distance, and iteratively training using a joint loss function composed of triplet loss and domain adaptive loss to obtain a dual-branch fault diagnosis model; collecting current heterogeneous fault data of the target mechanical device and inputting it into the dual-branch fault diagnosis model, outputting the fault classification of the vibration signal of the target mechanical device. This invention can solve the problem of model performance degradation caused by small sample data conditions and differences in data distribution.

The invention discloses a model optimization method, electronic equipment and a computer readable storage medium, and relates to the technical field of knock detection. The method comprises the following steps: acquiring a knocking signal sample set which comprises a plurality of anchor point samples, a plurality of positive class samples and a plurality of negative class samples; according to the envelope curve of each anchor point sample and each negative sample, determining the anchor point similarity of each negative sample, and based on the anchor point similarity of each negative sample, determining the triple loss boundary of each negative sample, the larger the anchor point similarity is, the larger the triple loss boundary is; inputting the knocking signal sample set into a to-be-optimized feature extraction model to obtain feature vectors of each anchor point sample, each positive class sample and each negative class sample; and optimizing a feature extraction model based on the feature vector of each anchor point sample, the feature vector of each positive class sample, the feature vector of each negative class sample and the triple loss boundary. According to the method, the distinguishing capability of a knocking detection algorithm on high-similarity interference signals can be improved in a complex real environment.

The application discloses a target detection anti-defense method, comprising the following steps: a multi-scale feature extraction module is used to obtain a detection feature representation of a to-be-detected image; a double-path attention mechanism is constructed to generate attention signals from an original image and the detection feature, and the comprehensive perception ability for spatial position disturbance is improved; a dynamic prediction module is constructed to perform weighted calculation on a plurality of convolution kernel parameters under the guidance of a spatial attention map, and different convolution parameters are adaptively allocated for different spatial positions; a network optimization mechanism based on a multi-loss function is constructed, wherein a dense triplet loss is used to ensure the discrimination ability of the spatial attention map for the local regions of two kinds of samples, and a target detection loss enables the model to have the prediction ability for target positions and categories; a target detection anti-defense model is trained, the target detection model is optimized in an end-to-end mode by using the network optimization mechanism of the multi-loss function, and target detection anti-defense is realized based on the trained target detection model.

The invention provides a software defect detection method based on Word2Vec and an auto-encoder triple network, which belongs to the field of software defect detection and comprises the following steps: acquiring an original code snippet and adding a label; processing an original code snippet with a label by utilizing Word2Vec, and weighting to obtain an embedded vector identifier of the code snippet; learning features of normal codes and defect codes through an auto-encoder; and constructing sample data containing anchor point samples, positive samples and negative samples to train the triple network, and optimizing an embedding space through a triple loss function to obtain a defect detection model. And calculating an embedding vector and a classification distance of the new code snippets based on the defect detection model, and judging whether the code snippets are defect codes or not. While the complexity of feature engineering is reduced, the problems of data imbalance and no sample detection are effectively solved, and the method has wide practical application value.

This invention relates to the field of logistics management technology, and discloses a cross-modal pharmaceutical logistics retrieval method, system, terminal, and storage medium. The method includes: based on an encoder, introducing a feature fusion mechanism to extract visual representations of textual information within the fused image, and achieving semantic enhancement through a cross-modal attention mechanism guided by a tag graph; designing a semantic neighborhood-aware contrastive hash code, a tag distribution-aware semantic alignment loss, and a cosine triplet loss to enhance the discriminative power of the hash code. This invention can improve the semantic alignment of multimodal features, enhance the model's generalization ability, and significantly improve the retrieval accuracy of pharmaceutical logistics.

The application provides a kind of small sample cross-domain fault diagnosis method based on multi-level feature alignment, comprising: collecting rotating machinery monitoring data under source domain and target domain, constructing cross-domain small sample fault diagnosis task data set of target domain sample scarcity;Diffusion model combined with multi-head self-attention mechanism and normalization strategy is constructed, and enhanced sample with target domain feature distribution is generated;Build a shared feature extraction network that integrates source domain, target domain and generated samples, and realize unified modeling of diagnostic features;Design a joint loss function, including source domain classification loss, double MMD distribution alignment loss, cross-domain triplet loss and enhanced consistency regularization term, construct a multi-level feature alignment mechanism, and realize fault feature transfer and accurate diagnosis under target domain small sample condition.The method of the application can effectively transfer the diagnostic knowledge of the source domain under the condition of the scarcity of data in the target domain, enhance the discriminability of the fault features and the cross-domain adaptability of the model, and is suitable for intelligent diagnosis tasks of rotating machinery under complex working conditions.

The embodiment of the invention discloses a verification code identification method and device based on a large model and related equipment. The method comprises the following steps: acquiring different types of sample verification codes; preprocessing different types of sample verification codes, and extracting multi-modal features of the sample verification codes; obtaining a verification strategy and an interference feature of each sample verification code, inputting each multi-modal feature and the corresponding verification strategy and interference feature into the initial multi-modal large model for identification training, and identifying the sample verification code based on an output result of the initial multi-modal large model and a corresponding real verification code type identifier. Carrying out loss calculation according to the triple loss function to obtain model loss, carrying out back propagation according to the model loss, and optimizing model parameters of the multi-modal large model to obtain an optimized multi-modal large model; and obtaining a target verification code needing to be identified at present, and outputting the target verification code to the multi-modal large model for identification to obtain an identification result. According to the method, the recognition accuracy and success rate of the cross-type verification code are improved.

The application discloses an entity alignment method based on multi-modal collaborative representation learning. On enhanced data, the initial semantic information of text and images is extracted based on a BERT model and a deep residual network, and the text and image features are projected into the same semantic space. The triplet loss loss is combined to make the text and image positive samples more similar and the text and image negative samples more different in the space. After training, the feature extraction and similarity calculation are performed on the unlabeled text and image data, the high-confidence entity alignment result is added to the seed data set, the model is iteratively updated, and the alignment of all text and images in the multi-modal data set is completed. The method uses a multi-modal representation learning method based on a pre-training model to optimize entity representation, does not need to manually construct entity features, simultaneously adopts an iterative data updating and network training process, greatly reduces the requirement for the amount of manually labeled data in the initial seed data set, saves a large amount of manpower and cost, and can obtain more accurate feature representation and alignment result.

The training method for the multimedia feature extraction model in this application includes: determining multiple training samples from multiple multimedia resources, each training sample including an anchor multimedia resource, a first multimedia resource, and a second multimedia resource; the anchor multimedia resource is any one of the multiple multimedia resources; the first object set corresponding to the first multimedia resource intersects with the anchor object set corresponding to the anchor multimedia resource; the second object set corresponding to the second multimedia resource does not intersect with the anchor object set; sorting the multiple training samples to obtain a target sample matrix; inputting the target sample matrix into a preset multimedia feature extraction model for feature extraction processing to obtain a multimedia feature matrix; determining triplet loss information based on the multimedia feature matrix; and training the preset multimedia feature extraction model based on the triplet loss information to obtain the target multimedia feature extraction model. Introducing object information during the generation of training samples on the multimedia side can improve the accuracy of the target multimedia feature extraction model.

The purpose of the present disclosure is to provide a gait recognition method based on complex frequency domain convolution and state space equation, comprising: forming an input image sequence by at least 2 image frames; converting the input image sequence from the spatial domain to the frequency domain, so that the original gait sequence input is converted into a feature with complex frequency domain information and time sequence dependence; acquiring key spatial information and space-time features according to the feature with complex frequency domain information and time sequence dependence, respectively, and performing splicing and fusion to obtain a fused feature; performing a maximum pooling operation on the time dimension through a maximum pooling layer to filter out the most representative spatial feature, and then condensing complex information into the most representative feature representation through an HPP layer for classification processing; outputting a prediction of the gait category according to the space-time feature; and training through a classifier combined with a triplet loss and a cross-entropy loss to finally realize the recognition of the gait.

A person re-identification method, a storage medium, and a terminal device are provided. In the method, a loss function used during model training is a preset distribution-based triplet loss function constraining a difference between a mean of a negative sample feature distance and a mean of a positive sample feature distance to be larger than a preset difference threshold; where the positive sample feature distance is a distance between a feature of a reference image, and a feature of a positive sample image, and the negative sample feature distance is a distance between the feature of the reference image and a feature of a negative sample image. In this manner, it can constrain the mean of the positive sample feature distance and that of the negative sample feature distance, thereby improving the accuracy of person re-identification results.

A kind of lightweight cross-view gait recognition method based on multilayer perceptron network, comprising the following steps: obtaining and preprocessing gait dataset including gait contour graph and gait skeleton graph;Gait recognition model including spatial compression enhancement module and space-time feature extraction module is constructed;Gait contour graph and gait skeleton graph are input into spatial compression enhancement module, compressed through multistage convolution, coded by pulse neural network LIF neuron and channel attention mechanism, to generate compressed and enhanced dual-mode fusion features;Dual-mode fusion features are input into space-time feature extraction module, to generate discriminative gait features through dynamic window space-time segmentation, frequency domain processing and multidimensional perceptron modeling;Gait recognition is carried out based on discriminative gait features;Gait recognition model is trained by jointly optimizing triplet loss and cross-entropy loss;The gait recognition model trained is tested.

The invention discloses a ConvNext backbone network and robust mixed matching-based lane detection method and system, and the method comprises the steps: firstly extracting multi-scale lane features through a ConvNext backbone network, and generating a high-semantic fusion feature map through a feature fusion network; feature enhancement is carried out by using a channel and a space attention mechanism; then, the parameters of the lane line are output through a lightweight Transform decoder; a composite matching strategy of LaneIoU and distance weighting is adopted, and optimal matching is achieved in combination with a Hungary algorithm; and finally, performing end-to-end training through a triple loss function. According to the method, the large-kernel deep convolution design of the ConvNext backbone network is combined with the lightweight Transform decoder, so that the model calculation efficiency is ensured, the lane detection precision in a complex scene is remarkably improved, and the robustness to interference factors such as illumination variation and shielding is effectively enhanced.

The invention relates to a model training method and system based on a cross entropy triple loss function. The method comprises the following steps: constructing a feature extraction network and a classification prediction network; fusing the feature extraction network and the classification prediction network to obtain a convolutional neural network classification model; constructing a cross entropy loss function and a triple loss function; constructing a cross entropy triple loss function based on the cross entropy loss function and the triple loss function; and updating model parameters of the convolutional neural network classification model based on the cross entropy triple loss function, so as to solve the problem that the model learns the characteristic of'effective classification but insufficient discrimination 'due to the fact that the convolutional neural network model is updated through the cross entropy loss function in the prior art. And the generalization ability and the classification reliability of the model in complex scenes (such as subdivided categories, sample noise and few samples) are influenced.

The application discloses a cross-resolution pedestrian re-identification method based on double-flow input feature reconstruction, and relates to the fields of computer vision and artificial intelligence. First, a double-flow input and feature reconstruction module is constructed to adaptively reconstruct the features of low-resolution images, and a feature degradation loss and a reconstruction loss are combined with a light residual decoder to learn the feature reconstruction capability; second, a weight-shared twin network structure is constructed in combination with a feature enhancement module to obtain deep invariant features of pedestrians; finally, a maximum average fusion pooling operation is used to pool the deep invariant features to obtain vectorized representation, and a cross-resolution triplet loss, a cross-resolution center loss and an identity loss are used to constrain the distribution of pedestrian features and improve the cross-resolution matching precision. The application is mainly applied to a video monitoring intelligent analysis system and has broad application prospects in the fields of intelligent security and the like.

The invention relates to the technical field of face recognition, and discloses a face recognition method based on deep learning, which is used for building a neural network model based on Retinface and Facenet to realize end-to-end face recognition. The Retinface uses an FPN feature pyramid network and an SSH network to enhance feature extraction, and can achieve pixel-level positioning of faces under various scale conditions, thereby achieving face detection. Facenet maps the face to the feature vector of the Euclidean space through CNN to obtain the feature vectors of the faces of different pictures, and the face recognition is realized through distance comparison. And meanwhile, Cross-Entry Loss and Triplet Loss are combined in the facenet network to serve as overall loss, network convergence is assisted, and the stability of the network is enhanced. By adopting the method, the face recognition detection speed can be increased, and meanwhile, the face recognition detection precision is improved.

This invention relates to the fields of computer vision and industrial quality inspection technology, specifically to an industrial anomaly detection method and system based on frequency-space dual-domain masking and counterfactual enhancement. It is applicable to small-sample industrial visual defect detection on complex textured surfaces such as textiles, metal wire drawing, and wafer surfaces. The invention employs a dual-encoding architecture of frequency domain suppression and spatial domain focusing. Features are fused after dual-branch processing using frequency domain masking and spatial semantic masking. During the training phase, a counterfactual adversarial fine-tuning strategy is incorporated, generating counterfactual samples based on a small number of normal samples and optimizing them using a triplet loss function. During the inference phase, dynamic hierarchical quantization decoding is used to complete anomaly detection. This invention solves the problems of high false positive rates and difficulty in collecting defect samples in complex textured backgrounds, achieving high accuracy and low false positive rates under small sample conditions, while significantly reducing inference computational overhead, thus adapting to the deployment needs of industrial edge computing.

The invention discloses a robustness semi-supervised encrypted traffic classification method and system, and the method comprises the steps: constructing a byte-level graph structure based on point mutual information, and carrying out the modeling of a topological relation between bytes in traffic; capturing a local context dependency relationship between bytes by adopting a GNN-based flow graph encoder; an automatic feature selection algorithm based on an NSGA-II evolutionary algorithm is designed, artificial feature engineering is replaced, and a feature matrix with high discriminant power is generated; constructing a multi-scale parallel convolutional neural network module, and extracting data packet level features to obtain rich sequence information; a collaborative training mechanism is designed, interactive optimization of a graph encoder and a CNN module is realized by generating a pseudo tag and screening a high-confidence sample, and the ability of a model to learn from a heterogeneous view is enhanced by adopting a triple loss optimization strategy. According to the method, the byte-level graph structure and the data packet-level features can be effectively fused, so that the problem that the classification performance is limited due to the fact that the data packet-level features are ignored in a traditional method under the conditions that the encrypted traffic data labeling cost is high and labeled data are scarce is solved.

The present invention discloses a method and system for open-set identification of drone signals based on metric learning. By dual-channel monitoring of the 2.4G and 5.8G frequency bands, the recognition ability of different types of drone signals is significantly improved. A deep learning network model is used as an encoder to extract feature vectors, and the generalization ability of the model is enhanced through L2 regularization. At the same time, an improved triplet loss function is introduced to simplify the training process and accelerate model convergence. In addition, the present invention proposes an improved KNN unknown category detector with a dynamic update mechanism that can automatically adjust parameters, adapt to new data, and effectively identify samples of unknown categories. The open-set identification application method of the present invention can iteratively update the network model, classifier, and location category detector when a new category of drone signal is detected, ensuring the continuous optimization and adaptability of the system and improving the flexibility and accuracy of drone identification technology.

This invention discloses a robust gait recognition method and system based on 3D convolutional neural networks, involution operators, and Transformer multi-fusion. The method includes: performing data augmentation and spatiotemporal alignment preprocessing on binary silhouette sequences extracted from walking videos; extracting local spatiotemporal features through an improved 3D convolutional network fusing involution operators, frequency domain spectrum enhancement layers, and pseudo-3D residual modules; extracting global temporal features through an improved Vision Transformer incorporating viewpoint and walking state conditional embeddings; batch normalizing the two feature streams separately, concatenating them, and mapping them with a fully connected layer to obtain fused features; performing end-to-end training using a joint loss function including cross-entropy loss, triplet loss, and strongly constrained triplet loss; and during the inference phase, binarizing the fused features into a 512-byte irreversible template using a learnable step function and performing authentication using normalized Hamming distance.