15results about How to "Improve discovery" patented technology

The invention relates to the technical field of computer vision, and discloses a camouflage target detection method based on cascade frequency domain perception and refinement feature guidance, comprising the following steps: a computer device obtains a camouflage image and extracts a multi-scale feature map; performing frequency domain separation by using a frequency domain sensing module, generating low-frequency and high-frequency characteristic sub-bands, and performing cross-hierarchy aggregation from bottom to top through a cascaded frequency domain sensor to obtain frequency fusion characteristics; receiving the frequency fusion features from top to bottom by using an aggregation guide decoding unit, performing guide refinement in combination with the deep decoding features, adapting to irregular edges by using a variable kernel convolution strategy in the refinement process, and generating a prediction feature map in combination with a partial convolution strategy; and outputting the predicted feature map as a detection result. According to the invention, through frequency domain information complementation and dynamic convolution refinement, the problem of difficult detection caused by highly similar texture of the camouflage target and the background is solved, and the positioning accuracy and edge segmentation integrity of the camouflage target are improved.

The application discloses a kind of method and device for security of intelligent seal machine, belong to intelligent seal machine field, wherein method includes: S1, receives and contains approval file of stamping application information;S2, through camera in seal cylinder and camera outside seal cylinder synchronous collection file image;S3, two-way image is fused after respectively being carried out distortion correction, generates complete file image;S4, complete file image is carried out OCR identification, and compared with application file content;S5, control electronic lock opens if consistent;S6, after electronic lock opens, real-time monitoring acceleration sensor data, if not detecting equipment moves, then execute stamping, if detecting movement, then control electronic lock closes and reports exception.This application realizes file full area identification by double camera joint, combines dynamic anti-moving monitoring, effectively eliminates the risk that seal area is tampered and cannot be identified, significantly improves the security of stamping.

This invention relates to the field of digital archives management technology, specifically to an artificial intelligence-based system for analyzing and warning of abnormal archive behavior. The invention uses a graph construction module to transform archive access logs into a relational topology graph; an attribute propagation module performs weighted transfer and accumulation of weights based on a graph traversal algorithm to calculate the cumulative relational index value of nodes; a connectivity retrieval module retrieves cross-regional links based on partition isolation rules; and an early warning output module generates early warning signals based on index values ​​and link status. This invention utilizes the flow mechanism of feature values ​​in a graph structure to aggregate minute risky behaviors within discrete long-term windows, accurately identifying complex abnormal patterns where a single operation may not violate regulations but the long-term cumulative risk exceeds the limit. This solves the problem that existing static statistical rules are unable to detect hidden logical violations, effectively reducing false alarms and false negatives.

The invention belongs to the technical field of aero-engine fuel oil and control, and particularly relates to an engine blockage judging method and processing method, electronic equipment and a storage medium. The method comprises the following steps: acquiring fuel nozzle data and fuel data of the engine; calculating ideal pressure reduction of each fuel nozzle according to the fuel nozzle data and the fuel data; according to the ideal pressure reduction, the pressure difference parameter of each fuel nozzle is determined; and in response to the fact that one pressure difference parameter is larger than a threshold value, it is judged that the fuel nozzle is blocked. According to the invention, quantitative characterization and real-time monitoring of the health state of the nozzle are realized, the discovery and positioning capability of early-stage blockage faults is obviously improved, and the safety and reliability of engine operation are enhanced.

This application discloses an automatic feature engineering method, device, medium, and product for predicting the remaining useful life of industrial equipment. It relates to the fields of intelligent operation and maintenance, fault prediction and health management of industrial equipment, and automatic feature engineering using machine learning. The method includes: acquiring an initial candidate feature set and labels; constructing a directed acyclic graph (DAG) for feature transformation using candidate features as nodes and recording derivation relationships; assigning nodes to multiple agents to establish a collaborative search decision-making process; encoding node states using a graph attention network and fusing parent-child relationships and context to generate action distributions; constructing prior policy guidance based on node feature types and feasible region constraints; sampling actions to update the graph structure and obtain a new candidate feature set; and using an evaluator to calculate performance increments as rewards, iteratively optimizing until the optimal feature set is output. This application can automatically construct feature combinations characterizing the degradation process of industrial equipment, significantly improving the accuracy and stability of remaining useful life prediction.

The invention discloses an infrared dim target detection method with a low signal-to-noise ratio, and relates to the technical field of image processing. The method comprises the following steps: firstly, continuously acquiring an infrared faint target image sequence, setting a motion feature set of a target, estimating the position of the target in each frame according to the motion feature set of the target, and superposing multiple frames of obtained images according to the motion feature set; an image group obtained through superposition is processed to obtain a candidate target saliency map, the saliency map is segmented to extract candidate targets, unified motion feature attributes are allocated to pixels of the obtained candidate targets, and gray values are allocated from the corresponding image group according to the unified feature attributes; distributing the mean value of the image group in the pixel direction to the remaining pixels to complete the accurate information enhancement of the target; and detecting and outputting the enhanced image by adopting an airspace detection algorithm. According to the method, the detection of a plurality of faint targets with low signal-to-noise ratios can be realized, and the discovery capability of an infrared detection system on a long-distance target is improved.

The application discloses a camouflage information acquisition method and system based on a dynamic reference field, which performs time, space and observation geometry alignment on multi-source observation data of a region to be acquired, generates a current multi-modal observation tensor, an environment compensation field, an observation geometry field and a grid index set; generates a dynamic reference field and a reference confidence field based on a last-time trusted observation library, and calculates a camouflage residual field and a cross-modal inconsistency field; further generates an acquisition priority field, determines an acquisition grid, an acquisition view angle and a repeated acquisition number; generates a camouflage information sample chain and a pollution prevention update gate according to a repeated acquisition result, and only allows a trusted result to be written back to form a current trusted observation library, which is used for dynamic reference updating at a next time, so that the accuracy, pertinence and pollution resistance of the camouflage information acquisition are improved.

The invention discloses a sample enhancement method and system based on multi-source data, belongs to the technical field of data processing, optimizes recognition of sample data based on the multi-source data, and provides recognition based on multi-dimensional behaviors, recognition based on attention relationships and recognition based on attention point locations. According to the method, suspected sample data mining based on multi-dimensional features and multi-dimensional relationships is realized, weighted integration of multiple sub-models is constructed in an ensemble learning form, the discovering capability of suspected sample data is improved, an adaptive iterative optimization mechanism of the model is constructed, and the precision ratio of the sub-models is checked regularly, so that the weighted integration mechanism is optimized automatically, and the discovering capability of the suspected sample data is improved. The self-adaptive related service sample discovery capability is realized, the border intelligent prevention and control capability is greatly improved, and more effective sample data expansion is provided.

The application relates to a multi-dimensional behavior unknown network threat detection method, system, device and medium, the method comprising the following steps: acquiring multi-source data and historical normal behavior data, correlating and modeling the multi-source data to obtain a structured security view; calculating multi-dimensional behavior characteristics based on the structured security view, and establishing a behavior baseline based on the historical normal behavior data; performing abnormal deviation analysis according to the multi-dimensional behavior characteristics and the behavior baseline to obtain abnormal segments and abnormal degree scores of the abnormal segments; correlating upstream and downstream evidence of the abnormal segments and reconstructing attack paths to obtain attack chain evaluation results; determining key asset influence degrees and vulnerability exploitation correlations based on the multi-source data, and fusing the abnormal degree scores, the attack chain evaluation results, the key asset influence degrees and the vulnerability exploitation correlations to calculate threat scores, and then outputting alarm levels according to the threat scores. The application has the effect of improving detection accuracy.

This application discloses an adaptive optimization method and system for thermal coupling parameters of automotive backlight panels, relating to the field of parameter simulation optimization technology. The specific implementation scheme is as follows: A full-envelope operating condition script is compiled; a thermal model and thermal coupling parameter set are imported; thermal simulations are automatically run in batches sequentially using the full-envelope operating condition script; raw datasets for each operating condition segment are collected; risky operating condition segments are screened; thermal coupling parameter perturbation groups are applied to each risky operating condition segment; high-risk operating condition segments are identified; parameter inversion is performed on each high-risk operating condition segment; and the thermal coupling parameter update strategy for each high-risk operating condition segment is determined, along with the optimal thermal coupling parameter version. This invention can improve the prediction accuracy of the thermal model under complex or extreme operating conditions, thereby avoiding problems such as underestimated junction temperature peaks, distorted hotspot locations, and drifting temperature rise rates. This provides a scientific and reliable data basis for the structural design, safety margin setting, and temperature control strategy optimization of automotive backlight panel radiators.

The application discloses a power grid load monitoring and optimization method based on deep learning, comprising: obtaining historical power equipment operation data and historical soft fault event data, and performing data preprocessing; calculating the control masking index of the voltage fluctuation and the adjustment disturbance of the voltage regulation module in each time window; extracting a multi-dimensional feature vector to predict the probability value of the existence of a soft fault of the current power equipment; extracting morphological feature data and action feature data through a standardized intra-day load curve data set; constructing a soft fault identification model to identify the soft fault type and soft fault level of the current power equipment, and executing a corresponding power equipment optimization adjustment strategy; and evaluating the effectiveness of the power equipment optimization adjustment strategy, and adjusting the power equipment optimization adjustment strategy based on the effectiveness. The application significantly improves the accuracy of soft fault identification, improves the safety, stability and intelligent level of power grid operation, and enhances the refinement and reliability of the whole life cycle operation management of power equipment.

This utility model discloses a lifting column module, relating to the field of fire safety technology. The module features a lifting column supported by a base, with a motor-driven lead screw that raises and lowers the column via a nut. A guiding reinforcement device and a sealing structure ensure stability and reliability. A base plate is installed at the top of the column to secure a dual-spectrum camera and fire extinguishing nozzle, and also includes power and communication sockets. An industrial drive board is installed at the bottom of the base to control the motor's operation. This module allows for flexible height adjustment of the dual-spectrum camera through the lifting column, effectively expanding the monitoring range, eliminating blind spots, improving the fire inspection robot's ability to detect fire hazards and its accuracy in early warning, and enhancing its adaptability and practicality in complex environments.

The present application relates to the technical field of power facility safety protection, and more particularly to an intelligent early warning and positioning system for unmanned aerial vehicle intrusion in all frequency bands of power transmission facilities, comprising: an all-frequency detection module for scanning a preset wideband radio signal to capture an unmanned aerial vehicle communication signal; an intelligent early warning and analysis module for receiving the signal, identifying the unmanned aerial vehicle and judging whether it enters a preset protection range, and issuing a warning information when it is judged to enter; a precise positioning and tracking module for calculating and outputting the three-dimensional coordinates and flight trajectory of the unmanned aerial vehicle based on communication signal data from multiple monitoring nodes after the warning information is issued. The present application realizes all-frequency monitoring, intelligent early warning and high-precision positioning and tracking of unmanned aerial vehicles around power facilities, effectively improving the low-altitude security and protection capability.

The invention belongs to the technical field of network security, and discloses a blackmail virus dynamic detection system based on a multi-modal large model, which comprises a data acquisition module used for dynamically capturing a multi-dimensional behavior data flow caused by running of a to-be-detected program in a monitored computing environment; and the dynamic analysis module is connected with the data acquisition module and is used for loading and running the to-be-detected program in a controlled isolation analysis environment and receiving the multi-dimensional behavior data flow from the data acquisition module. According to the scheme, by integrating multi-dimensional dynamic data such as file operation, network behaviors and system calling and constructing the unified multi-modal feature vector, more comprehensive and richer analysis materials are provided for the model, and recognition of a complex and cross-modal malicious mode is facilitated; a pre-trained multi-modal large model is used for deep correlation analysis and pattern recognition, and the detection ability and generalization ability of novel and variant ransomware viruses are remarkably improved.

The application provides an interpreter-based script full-path verification method and device based on an executor rewrite, a medium and a program product. The method comprises the following steps: obtaining an interpreter-based script and performing analysis to generate an abstract syntax tree; configuring a verification executor as an interpretation execution subject of the abstract syntax tree, and pointing an interpretation execution flow of the abstract syntax tree to the verification executor for traversal-type interpretation execution; in the traversal-type interpretation execution process, an override execution rule is adopted, so that a plurality of execution paths corresponding to a syntax node are executed, and when an exception occurs, the remaining execution paths are continuously verified; meanwhile, in the traversal-type interpretation execution process, a called script access object is synchronously identified and white list verification processing is performed, so that the script accessible object range is safely verified while the full-path execution verification is completed. The application cooperatively verifies the execution logic and access permission of the interpreter-based script, and improves the integrity, safety and reliability of the script verification.