2002results about How to "Improve generalization ability" patented technology

The invention relates to the technical field of knowledge graph completion, in particular to a large model knowledge graph completion method and system based on causal guidance. The method comprises the following steps: acquiring a target knowledge graph and an input triple to be complemented, performing structured analysis on an input triple relationship, extracting key topological characteristics, constructing a structured mediation variable, mapping the structured mediation variable into a structure guide prefix, injecting the structure guide prefix into large model input, and constructing a double-path inference model to generate an inference prediction result; and meanwhile, a gradient sensing dynamic loss balance mechanism is introduced, the loss weight is adaptively adjusted according to reasoning feedback, and finally a more accurate and stable knowledge graph completion result is output. According to the method, the controllability, interpretability and training stability of the reasoning process can be enhanced while the knowledge graph completion precision is improved.

The invention discloses a small target detection method based on a Fourier mixed attention mechanism. An improved small target image detection network model based on RT-DETR is researched and designed, and a Basic Block module in a backbone network is replaced by a self-developed Fourier mixed attention enhancement module (FTABlock). The module is composed of a Fourier transform attention module (FTAModule) and a hybrid dynamic convolution feedforward network (DKMixFFN). The FFTModule calculates the attention weight in the frequency domain through Fourier transform, and strengthens small target feature expression in combination with position coding and multi-head attention; the DKMixFFN adopts a dynamic convolution kernel and a multi-scale dynamic convolution layer to realize adaptive modeling of multi-scale features. According to the method, under the synergistic effect of frequency domain attention and dynamic convolution, the small target feature extraction and perception capability of the RT-DETR model is effectively enhanced, and the small target detection precision is improved.

The invention discloses a deep tunnel surrounding rock mechanical parameter inversion method based on a three-dimensional brittle failure zone contour and sparrow optimization algorithm-information vector machine (SSA-IVM) combined optimization algorithm. The engineering technical problem that due to the fact that excavation instantaneous displacement is difficult to monitor, deep tunnel surrounding rock mechanical parameters are difficult to reasonably obtain through displacement back analysis is solved. The method comprises the following steps: firstly, constructing a tunnel FLAC3D numerical simulation model with the same ground stress condition at the occurrence position of a three-dimensional brittle failure zone; secondly, taking an absolute error between the total number of computational grid units in the actually measured brittle failure zone and the total number of computational grid units entering a plastic state in the range of the actually measured brittle failure zone after calculation of the FLAC3D numerical model as an optimization objective function; and then, by taking the tunnel surrounding rock mechanical parameters as optimization variables and taking a target function reaching a global minimum value as a target, performing global optimization by combining a tunnel FLAC3D numerical model and adopting an SSA-IVM joint optimization algorithm, thereby obtaining reasonable tunnel surrounding rock mechanical parameters.

The invention discloses an intelligent storage multi-AGV scheduling method and device and a medium, and relates to the technical field of automatic scheduling, and the method comprises the steps: collecting the state and environment perception data of AGVs, and carrying out the standardization processing; constructing a dynamic conflict scheduling model, inputting the standardized AGV state and environment perception data into the dynamic conflict scheduling model for task allocation and path planning, and outputting a preliminary task allocation and path planning scheme; and integrating task execution feedback data with historical task data, optimizing parameters of the dynamic conflict scheduling model, and outputting an optimized task allocation and path planning scheme. According to the method, the task execution feedback data and the historical task data are integrated, and parameter optimization is performed on the dynamic conflict scheduling model based on the scheduling knowledge graph and the meta-gradient descent strategy, so that continuous self-evolution and precision improvement of the scheduling strategy are realized.

The invention belongs to the technical field of artificial intelligence, and particularly relates to a GUI-Agent trajectory data generation method and system based on multi-agent collaborative reasoning. The method comprises the following steps: S1, capturing and analyzing a page state and a structure, and constructing an initial graph; s2, analyzing semantics of interactive elements based on VLM and generating an initial action instruction set; s3, executing the instruction, recording state change, automatically deducing a target intention, forming an initial track, and storing the initial track in a track pool; s4, for the current track, predicting and executing the next action according to the target and the page state of the current track, and updating the track; s5, unexplored elements are recognized, new actions and intentions are generated in combination with the VLM, and a new track is created and added into a track pool; s6, evaluating whether the action result achieves a target or not, and deciding to terminate, adjust or continue to explore according to the target; and S7, circularly executing the steps S4 to S6, performing parallel iteration processing on the trajectory pool until a termination condition is met, and outputting a trajectory data set.

The invention relates to the technical field of intelligent traffic and Internet of Vehicles, in particular to a traffic and air pollution bidirectional coupling flow prediction method based on double-path dynamic fusion. Comprising the following steps: collecting traffic flow and air pollutant concentration data, and carrying out space-time alignment and reversible instance normalization; the data is divided into two branches, the first branch extracts time-dependent features through gated convolution and probability sparse self-attention, and the second branch obtains variable interaction features through dimension remodeling, context extraction and reversible coupling transformation; bidirectional feature interaction is carried out through cross attention, weights are dynamically generated based on channel attention, and residual connection is carried out after weighted fusion; and performing linear mapping and inverse normalization on the fused features to obtain a traffic flow predicted value. According to the method, the prediction precision and robustness in a pollution sensitive scene are remarkably improved.

The invention discloses a dynamic space-time diagram flow prediction method and system based on course learning, and relates to the technical field of supply chain logistics data analysis, and the method comprises the steps: building a space-time matrix based on historical multi-source data, generating a dynamic adjacent matrix through learning, and carrying out the smooth fusion through combining a static diagram, and forming a dynamic diagram structure. And then space and time features are respectively extracted by using a graph convolutional network and a gating loop unit, and deep interaction and fusion are realized through a bidirectional cross attention mechanism. A multi-dimensional difficulty estimator is innovatively introduced, the prediction difficulty of each training sample is quantified from three dimensions of space, time and time-space coupling, the selection sequence of the training samples is dynamically adjusted based on an adaptive course scheduler, and progressive learning is realized. And finally, feature representation is obtained through global pooling, and multi-step traffic prediction is realized by adopting a parallel independent decoder, so that error accumulation is avoided. According to the invention, prediction precision and model training efficiency in a complex supply chain logistics scene are effectively improved.

The invention discloses a medical image segmentation method, system and device based on multi-attention and multi-scale fusion, and relates to the technical field of image segmentation, and the method comprises the steps: constructing an MAMF-Net model which comprises an encoder and a decoder which are in multi-layer jump connection; the encoder adopts a hybrid architecture of convolution and Transform, and is integrated with a self-adaptive expansion convolution method; the decoder integrates dual-channel attention gating and a multi-scale global channel feature enhancement method to enhance features transmitted by jump connection, and combines features extracted by the encoder to fuse and reconstruct a segmentation result; training the MAMF-Net model by adopting the historical medical image sample set to obtain a medical image segmentation model; and obtaining any medical image to be identified and inputting the medical image to the medical image segmentation model, and determining a corresponding segmentation result. The problem of insufficient fusion of global semantics and local details in medical image segmentation is solved.

The invention relates to an intelligent agent continuous training and effect evaluation closed-loop method and system fusing work order feedback, and a medium, and relates to the technical field of deep learning. The agent continuous training and effect evaluation closed-loop method comprises the following steps: receiving and analyzing work order feedback information, and recording the number of error information and information storage duration in combination with a preset structure sample label; dynamically judging the number of error information and the information storage duration, triggering a retraining rule of an original agent model, distributing candidate information weights, and screening a target sample set; directionally training and optimizing the original agent model according to a retraining rule in combination with the target sample set, and generating a new agent model; constructing a historical work order problem set, comparing and verifying a new agent model in combination with an original agent model, generating an agent evaluation table, and correcting a candidate information weight; by accurately focusing the weak link, the continuous and efficient enhancement of the model capability is realized, so that the answer accuracy of the intelligent agent is remarkably improved, and the AI illusion is effectively inhibited.

The invention provides a large language model generation code detection method and system, which is applied to the technical field of artificial intelligence, and comprises the following steps: obtaining a to-be-detected code; a to-be-detected code is input to a trained shared encoder, a code feature vector is obtained, the shared encoder is obtained through multi-target joint training, and the multi-target joint training is used for optimizing classification loss, domain confrontation loss, comparison loss and difficult sample loss at the same time; l2 normalization is carried out on the code feature vector, and the code feature vector is mapped to a hyperspherical space to obtain spherical embedding; the sphere is embedded and input into a sphere category classifier based on sphere logistic regression for classification processing, a detection result of the to-be-detected code output by the sphere category classifier is obtained, the detection result comprises AI generation and human writing, and a decision boundary of the sphere category classifier is an intersection line of a hyperplane and a hypersphere for classification. According to the invention, the AI generation code and the human compiled code can be accurately distinguished.

The invention discloses an anaphora disambiguation method and system based on big language model enhanced text and structured query language generation. The method comprises the following steps: receiving a natural language question of a user, and analyzing and generating structured table field information through mode information; identifying and rewriting fuzzy time expression in the problem; extracting ambiguous entities, and replacing the ambiguous entities with database standard values through semantic matching and fuzzy matching; synthesizing the information to generate a structured query language statement and executing the structured query language statement; and if the execution result is null, automatically triggering an anaphora disambiguation process, performing error correction and re-matching on field values in the statement through a dual matching mechanism, and generating and executing a corrected query statement. According to the method, the problems of fuzzy anaphora, indefinite time expression, low matching accuracy, resource waste and the like in a traditional SQL system are effectively solved through a strategy of combining pre-processing and post-processing, and the accuracy, robustness and execution efficiency of complex query are remarkably improved.

The invention discloses a generative model redrawing attack-oriented image robust watermark traceability method and system and a computer readable storage medium, and belongs to the field of digital information security and artificial intelligence content governance. The method comprises the following steps: in a watermark embedding stage, performing error correction coding and digital signature processing on traceability information containing identity information and a timestamp to generate a load to be embedded, and embedding the load to be embedded and a synchronization template for geometric synchronization into a host image in a function separation manner by using a deep neural network; in the training stage, combined optimization is carried out on the watermark embedding and extracting process by introducing generative model redrawing and microsimulation of image distortion attack, so that the robustness of the watermark under a complex attack condition is improved; in the extraction and verification stage, under the condition that an original image is not needed, geometric synchronous correction and blind extraction of a watermark load are carried out on an image to be analyzed, and authenticity confirmation of traceability information is completed through error correction decoding and digital signature verification. By adopting the technical scheme of the invention, the problems of insufficient traceability information robustness, difficult source confirmation and incomplete evidence chain in a generative model redrawing scene in the prior art are solved, and verifiable traceability and credible evidence generation of the image source information are realized.

The embodiment of the invention discloses a malignant load identification method, device and equipment, a medium and a program product, and relates to the technical field of power load monitoring. The method comprises the following steps: performing modal decomposition on an original power utilization sequential sequence to obtain a plurality of intrinsic mode components, and reconstructing intrinsic mode components which do not belong to noise components to obtain a target power utilization sequential sequence; performing feature extraction on the target power consumption time sequence to obtain target power consumption features, and inputting the target power consumption features into a pre-trained malignant load identification model for identification to obtain an identification result; the malignant load identification model is obtained by updating model parameters of a weak learner based on a natural gradient descent method and performing training optimization. The lightweight malignant load learning model obtained through training in the scheme can be deployed and operated on the intelligent electric meter, high-quality input features are obtained through multi-mode decomposition and reconstruction, the accuracy of malignant load recognition is improved, and accurate recognition of the malignant load based on the lightweight model is achieved.

The invention belongs to the technical field of power transmission line modeling, and provides a power transmission line channel point cloud modeling method and system, and the technical scheme is that based on standardized point cloud data, global statistical features are extracted, channel scene categories are automatically judged, and adaptive feature weight vectors are generated according to the judged scene categories; performing weighted calculation on the similarity by using the feature weight vector, and performing superbody clustering segmentation on the standardized point cloud data to obtain an optimized superbody set formed by a plurality of superbodies with similar internal features; based on multi-dimensional feature statistics and context rules, screening the optimized superbody set, and rejecting small non-ground interference superbodies to obtain a ground candidate superbody set; and performing identification, merging and curved surface fitting on the ground candidate superbody set to generate a continuous power transmission line channel ground model. High-precision identification of ground points is realized, and a high-quality ground reference can be provided for safety analysis of a power transmission channel.

The invention discloses a multipath channel parameter reconstruction and beam coverage prediction method, and relates to the technical field of wireless communication. Aiming at the problems of low data utilization efficiency, limited environment modeling precision, strong coupling of measured data and antenna beam configuration and the like in the prior art, the method comprises the steps of configuring antenna beams and orientation parameters, collecting reference signal receiving strength, performing deterministic channel modeling simulation, establishing a mapping model, performing sparse recovery, constructing a channel propagation model, evaluating a channel map and the like. In combination with a coherence perception weighting network (CARE-Net) algorithm and ray tracing (RT) physical prior guidance, accurate reconstruction from a low-dimensional reference signal receiving strength (RSRP) measurement value to a high-dimensional multipath parameter is realized. The method has the advantages that the reconstructed multi-path parameters are decoupled from the antenna configuration, the method has the cross-beam / cross-configuration generalization prediction capability, the channel coverage performance under different antenna configurations can be accurately predicted, a reliable basis is provided for wireless network optimization, and the method is low in cost, high in precision and strong in generalization.

The invention relates to the technical field of load prediction, and provides a power load prediction method and system based on time sequence decomposition and an attention mechanism, and the method comprises the steps: carrying out the adaptive time sequence decomposition of an obtained original load sequence, calculating the sample entropy of each decomposed component, and carrying out the clustering; constructing a group of encoder and decoder networks for each piece of clustered data, performing parallel encoding to extract features, performing serial decoding reconstruction on the features from low frequency to high frequency, and outputting prediction data from low frequency to high frequency step by step; the weight is initialized based on the sample entropy, the trained weight is obtained through optimization in the encoder and decoder network training process, and the predicted value of the power load is obtained through weighted fusion. According to the method, adaptive time sequence decomposition, a weight mechanism guided by sample entropy and an attention-enhanced encoder-decoder structure are introduced, so that multi-component collaborative modeling and cross-scale dynamic prediction are realized, and the prediction accuracy and stability in complex load data and small sample scenes are effectively improved.

The invention relates to a heat supply prediction method and system based on spatial-temporal feature fusion deep learning, and the method comprises the following steps: S1, carrying out the collection and fusion of multi-source heterogeneous data, and constructing an integrated data set; s2, preprocessing the data; s3, constructing a graph structure model of the heat supply system, and constructing a weighted undirected graph; s4, constructing and executing forward calculation of the space-time double-flow deep network; s5, designing a composite loss function including mean square error loss and physical constraint loss, and performing joint optimization training on the space-time double-flow deep network; and S6, performing multi-step heat supply load prediction by using the trained model, outputting a heat supply load curve of each heat exchange station in a specified time period in the future, and integrating a prediction result with a heat supply scheduling system. The method has the advantages that the prediction precision is improved compared with that of a traditional machine learning model by capturing the spatial-temporal characteristics at the same time, and the advantages are more remarkable in extreme weather.

The invention provides an emotion recognition method based on electroencephalogram feature fusion and a double-stage attention mechanism, and the method comprises the following steps: A, electroencephalogram signal processing: carrying out the preprocessing of an electroencephalogram signal; and B, double-stage attention feature fusion: in each selected frequency band, adopting a double-stage attention mechanism to fuse the electroencephalogram features, and generating fusion features for emotion classification. And C, double-branch feature extraction: performing double-branch 3D convolution processing on the fused features, extracting multi-scale space-spectral time features, and splicing the multi-scale space-spectral time features along a channel dimension to form uniform features. And D, classification and output: inputting the unified features into a classifier, and generating an emotion category prediction result through a flattening layer and a full connection layer. According to the method, the difference entropy, the power spectrum density and the difference entropy asymmetry feature are fused through unified three-dimensional feature representation, a double-stage attention mechanism is introduced, and high-accuracy emotion recognition is achieved.

The invention provides a new energy automobile endurance prediction method based on working condition identification and prediction, and belongs to the technical field of new energy automobile endurance mileage estimation. According to the method, sliding window statistical characteristics are constructed through speed, current, voltage and other signals, typical working conditions of cities / suburbs / highways and the like are identified, the working condition of a next time window is predicted, and energy consumption priori which can be updated in real time along with changes of road conditions is formed. Secondly, a rolling capacity estimation model is provided, the available capacity of the battery is continuously evaluated based on historical charging and discharging data, and dynamic changes of the SOH of the battery along with time and use conditions are recorded; according to the method, working condition prior, rolling SOH and time sequence energy consumption characteristics are fused, a TCN-BiLSTM multi-model cooperation framework is constructed, short-term dynamic and long-term attenuation trends are considered, high-precision and generalizable endurance mileage prediction under complex working conditions and battery state fluctuation is realized, reliable support is provided for intelligent energy consumption management and journey planning of vehicles, and the method is suitable for popularization and application. The method can also be applied to an intelligent power distribution system with a high reliability requirement, and provides a core solution for technology promotion service in related fields.

The invention discloses a multi-task electromagnetic model based on a hybrid expert network, and belongs to a wireless communication technology. The model comprises a preprocessing module, a feature extraction module, a task output module and a pre-training-fine tuning learning strategy. The preprocessing module carries out standardization processing on the multi-source electromagnetic signals; the feature extraction module is based on a Transform structure, introduces a hybrid expert network to replace part of a traditional feedforward neural network, and dynamically selects an expert sub-network through a task specific routing mechanism; the task output module configures a special structure according to different task targets; in the pre-training stage, a mask auto-encoder is used for pre-training large-scale label-free data, and a downstream task is subjected to full-amount fine adjustment through small-scale label data. According to the method, multi-task collaborative learning and differential expression are realized, and the recognition performance, robustness and processing efficiency of the model in a complex electromagnetic environment are improved.

The invention relates to a network prediction model for identifying a new cancer gene, a model construction method and application, and belongs to the field of biostatistics. According to the method, genomics, transcriptomics and proteomics data are integrated, a dynamic tensor twinning graph neural network is constructed, a hybrid multilayer random block model is utilized to perform tensor decomposition to extract global and local community features, causal and non-causal information is separated, causal feature mutual information is maximized, and non-causal interference is inhibited, so that the dynamic tensor twinning graph neural network is obtained. The model interpretation force is enhanced; through a dynamic community perception algorithm and a twin graph neural network, inter-layer graph dissimilarity is learned by using a graph similarity function, accurate detection of network structure mutation nodes in a high-frequency dynamic scene is realized, and an independent state transition and co-evolution mode is synchronously positioned. And the biological functions of the candidate genes are verified by combining KEGG pathway enrichment analysis and an independent database. Compared with an existing optimal algorithm, the mutation detection accuracy and the calculation efficiency are remarkably improved, and the AUROC and the AUPRC are both higher than those of an existing advanced recognition method.

The invention relates to a federated learning-based magnetocardiogram analysis method, device and equipment and a medium, and relates to the technical field of magnetocardiogram analysis, and the method comprises the following steps: aiming at a preset arrhythmia analysis task, configuring a corresponding independent task head on a frozen shared feature backbone network, and caching a corresponding reference logic value; and configuring a corresponding new task head for the new task, and inserting a low-rank adapter module into the frozen shared feature backbone network so as to perform joint optimization on the inserted low-rank adapter module, the new task head and the independent task head, updating parameters of the low-rank adapter module, parameters of the new task head and parameters of the independent task head, according to the method, the limitation of a data island is broken through, data resources of multiple centers are integrated, massive unlabeled data can be fully mined, the model feature extraction capability is improved, the situation that new and old data are used for retraining the model is avoided, and the safety of the federal continuous learning model is improved. And the old category diagnosis capability is ensured.

The invention provides a multi-path information retrieval method and system based on a language model and uncertainty evaluation. The multi-path information retrieval method comprises the following steps: receiving a user query; performing semantic analysis on user query by using a pre-trained language model so as to output a quantitative uncertainty index in an unsupervised manner; dynamically determining a quantity ratio for controlling query generation and a balance parameter for controlling result sorting based on the index; based on the user query and the strategy parameters, generating and executing two groups of differential queries at least comprising exploratory queries and utilizable queries to obtain multiple paths of retrieval results; and combining multiple paths of results, and carrying out dynamic balance reordering between the correlation and diversity of the results according to balance parameters by adopting an ordering mechanism. According to the method, the retrieval strategy can be adaptively adjusted according to the uncertainty of query, and the coverage degree and diversity of results are remarkably improved.

The invention discloses a photovoltaic power prediction method based on improved empirical mode decomposition and an optimized long short-term memory network, and the method comprises the steps: firstly carrying out the preprocessing of abnormal value elimination, missing value filling, normalization and the like of photovoltaic power and related meteorological data, and improving the data quality; then, an improved empirical mode decomposition (EE-ANEMD) algorithm is adopted to decompose the preprocessed power sequence into a multi-scale intrinsic mode function component and a residual term, and high-frequency noise, intermediate-frequency fluctuation and a low-frequency trend are effectively separated; global optimization is carried out on the hidden layer unit number, the initial learning rate and the maximum number of training times of the LSTM network through an improved sparrow search algorithm (ISSA), finally, the optimized LSTM is utilized to carry out training prediction on each component, and results are fused and subjected to reverse normalization to obtain a final value. Experiments show that the test set RMSE of the method is reduced compared with that of a single LSTM, the mid-term prediction precision is remarkably improved, and reliable technical support is provided for power system dispatching, new energy consumption planning and photovoltaic power station operation and maintenance.

The invention provides a memristor neural network cloud edge collaboration method and a memristor neural network cloud edge collaboration system based on multi-mode sensing. The method belongs to the field of artificial intelligence multi-modal data processing. The method mainly comprises the following steps: distributing multi-modal data subjected to synchronous acquisition and time-space alignment to an FPGA (Field Programmable Gate Array) and a memristor array for parallel processing; splicing the processed multi-modal data in a channel dimension to generate a multi-channel fusion feature tensor; a memristor array is used for carrying out storage and calculation integrated calculation on the multi-channel fusion feature tensor, and an advanced feature map is output; receiving a differential update packet issued by the cloud, and selectively updating the weight in the memristor array according to the update packet; and carrying out in-situ fine tuning on partial weights in the memristor array based on locally acquired data. According to the method and the constructed system, the memristor array can be effectively compensated, the calculation precision is improved, and rapid self-adaption of a local scene and continuous and efficient iteration of global knowledge are achieved.

The invention relates to the technical field of vehicle trajectory analysis, in particular to a vehicle trajectory anomaly detection method and system based on deep learning, and the method comprises the following steps: obtaining trajectory coordinates and boundary distance to generate symbol offset, constructing a road offset continuous field through interpolation and spatial embedding, calculating the ratio of instantaneous speed to reference speed, and obtaining a vehicle trajectory anomaly detection result. And generating an unbalance degree parameter, executing vector dimension stretching, extracting a trajectory embedding vector and an abnormal score by using a long short-term memory network, executing square summation and difference operation, and generating a trajectory energy gradient value. According to the method, migration characteristics are mapped to a continuous field to achieve track and road geometric constraint association, unbalance degree parameters are used for remarking directional quantity scales to enhance speed dynamic perception, energy gradient modulation scoring is cooperated, a confidence probability curved surface is constructed by using trilinear interpolation, and a curved surface peak value is positioned to output a detection result. Positioning noise is suppressed; and abnormal precision under sparse sampling is improved.

The invention provides a target identification method and system based on multi-source information fusion, and relates to the technical field of low-altitude target detection. The method comprises the following steps: acquiring target radar track data, interception equipment track data and position area information data; based on target radar track data, extracting a first feature in an RCS form dimension, and extracting a second feature in a motion dimension; based on the target radar track data and the track data of the monitoring equipment, determining a frequency spectrum monitoring correlation factor and regional position information features; performing feature fusion on the first feature, the second feature, the spectrum interception correlation factor and the regional position information feature to obtain a target feature, and identifying a target type; and identifying a target threat level based on the target type and the target radar track data. The method and the device are used in a target identification process based on multi-source information fusion, and the technical problem that the target threat degree cannot be accurately identified in a complex environment in the prior art is solved.

The invention discloses a wafer defect detection method and system based on an attention guidance network, and belongs to the technical field of wafer defect detection, and the method comprises the following steps: 1, adjusting light supplement according to a detection demand, and then collecting a wafer image; 2, performing spectrum enhancement preprocessing on the acquired wafer image; and 3, extracting the features of the enhanced image through a multi-scale distributed feature extraction backbone network, and fusing the extracted features through an attention-guided feature pyramid network. Complex background interference of the wafer is effectively suppressed, and defect characteristics are remarkably enhanced; constructing a feature extraction mechanism capable of capturing local details and long-range context information at the same time; balanced and accurate detection of multi-scale defects is realized; the physical priori of the defect is embedded into the network in a learnable manner, so that the learning efficiency and generalization are improved; while ultrahigh precision is ensured, low model complexity is maintained, and the real-time requirement of a production line is met.

The invention discloses a voiceprint recognition system attack resisting method and device, equipment and a medium, by constructing a shadow voice sample set and a substitution model, dependence on target speaker data and query capability is reduced, and privacy risk and query overhead are reduced; meanwhile, two types of substitution models are trained based on an information theory method, so that the mobility and stability of the models are improved, and the generalization ability is enhanced; besides, the attacker model adopts a specific architecture and combines an alternate training strategy, so that attack loss and prediction probability difference are optimized, imperceptibility and attack effectiveness are balanced, and tone quality reduction or attack failure is avoided; and finally, through systematic training and optimization, a more uniform evaluation caliber is expected to be provided, and more reliable guidance is provided for engineering landing and risk evaluation.

The invention belongs to the technical field of computer vision, and particularly discloses a monocular 3D object detection method for realizing depth enhancement based on a visual basic model, electronic equipment and a readable storage medium, and the method comprises the steps: S1, building a data set: employing a monocular camera to collect a pavement scene, and obtaining an RGB image in the pavement scene; s2, image preprocessing: preprocessing the RGB image for subsequent feature extraction and depth estimation; s3, performing feature extraction by adopting a dual-backbone network: performing visual semantic feature extraction on the preprocessed RGB image by using DINOv2; performing depth feature extraction on the preprocessed RGB image by using a DPT head; s4, generation of depth perception query points: inputting the visual semantic features and the depth features into a DETR network to generate the depth perception query points; and S5, target detection output: using an MLP-based detection head to obtain information of the category, the size, the center point position, the depth, the 3D size and the direction of the object.