26results about How to "Improve detection robustness" patented technology

The invention relates to the technical field of ship traffic management and marine monitoring, and discloses a multi-source data-based false AIS signal detection method, which comprises the following steps of: firstly, extracting a historical track to construct a coverage grid mapping table containing radar detection attributes; receiving real-time data and dividing the real-time data into an AIS to-be-detected group and a radar truth value group; non-radar blind area targets are screened according to the mapping table; retrieving an associated target in the radar truth value group to execute existence reverse verification; a dynamic time warping algorithm is adopted to align time sequences and calculate feature differences, and fine comparison verification is executed; constructing a neighborhood flow field model, and executing flow field consistency verification according to the motion vector deviation; and outputting a false AIS signal list in combination with multi-stage verification results. According to the method, through physical coverage modeling and environmental dynamics constraint, high-simulation-degree false signals in a radar blind area can be effectively identified, so that the false alarm rate is reduced.

The invention discloses a hot-line work range prediction method based on an improved YOLOv8 algorithm, and relates to the technical field of hot-line work, and the method comprises the steps: obtaining the multi-modal data of a hot-line work site, and dividing the multi-modal data into a visible light data region and an infrared data region according to the physical band attribute of the data; a feature index channel based on improved YOLOv8 is established for the visible light data partition, and a target feature index table is output by using a self-adaptive deformable convolution and attention mechanism model; determining a key feature region of the visible light data partition based on the target feature index table, and introducing a space-time adjoint prediction model to predict the real temperature field distribution of the infrared data partition; performing collision detection on the real temperature field distribution and the key feature area by using a space collision detection algorithm, and judging whether the equipment space bounding box is intersected with the thermal energy field outline or not to obtain a boundary judgment result; according to the invention, the recognition precision and reliability of the hot-line work boundary are improved through multi-modal data complementation.

The application provides a method and device for detecting a very small defect, equipment and a storage medium. The method is based on a pre-trained improved neural network. The improved neural network is trained by using a combined loss function including a Focal Loss classification loss function and a GIoU Loss bounding box regression loss function. The improved neural network comprises a backbone network, a bidirectional feature pyramid network, an attention enhancement module and a detection head. The method comprises the following steps: inputting a target image to be detected into the pre-trained improved neural network; performing feature extraction on the target image based on the backbone network to obtain a multi-scale feature map; performing fusion processing on the multi-scale feature map based on the bidirectional feature pyramid network to obtain an enhanced feature map; performing feature recalibration on the enhanced feature map by the attention enhancement module to obtain an optimized feature map; and performing processing on the optimized feature map by the detection head to output category information and position information of the very small defect in the target image.

The invention discloses a deep counterfeiting detection method based on double-flow fusion and adaptive feature enhancement, and belongs to the technical field of computer vision and digital media security. According to the method, a double-flow feature extraction network is constructed, a high-low feature adaptive enhancement module (HLFAE) is adopted in a spatial flow to decompose multi-scale texture features, and micro texture expression of a forged area is enhanced in combination with expansion convolution and a channel attention mechanism; introducing a high-frequency sub-band capable of learning a discrete wavelet transform self-adaptive decomposition image into a frequency domain flow, and amplifying frequency domain artifact features through a convolutional network; a multi-modal enhanced feature attention module is designed to dynamically fuse space and frequency domain features, significant features are weighted through a multi-scale convolution kernel and a double attention mechanism, and feature interaction consistency is improved based on a cross-modal contrast enhancement (CMCE) module. According to the method, the highest AUC value of 99.63% is achieved on data sets such as FaceForce + + and Celeb-DFv2, the cross-domain generalization ability and the anti-disturbance robustness are remarkably improved, and the method is suitable for financial risk control and media content auditing.

This invention discloses a method for detecting the appearance of planar objects. First, a set of detection parameters, including imaging parameters, acquisition strategies, and threshold rules, is obtained based on the object information. Then, camera imaging calibration and pixel-to-actual-size ratio calibration are performed, establishing a coordinate mapping between pixels and the stage. Initial images are acquired to obtain object pose parameters and perform rotational alignment. Multi-view or rotational scanning is performed for synchronous image acquisition according to the imaging parameters and acquisition strategy. Image quality is evaluated frame by frame; if unqualified, parameters are adjusted and images are re-acquired. For qualified images, illumination normalization and standardization preprocessing are performed to extract the final region of interest, generate defect candidate regions, and extract grayscale, shape, and texture features for identification and classification. The defect category, confidence level, and actual size are output. Finally, a qualified or unqualified result is determined based on preset threshold rules. This invention eliminates misjudgments caused by low-quality images at the source, significantly enhances detection robustness, and has good adaptability to various objects.

The invention aims to provide a weighted nuclear norm weak and small target detection method based on low-rank clustering improvement, and the method comprises the following steps: S1, carrying out the preprocessing of a weak and small target image to be detected, and obtaining a preprocessed image; s2, dividing the preprocessed image into a plurality of image blocks; s3, classifying the similar image blocks into one group according to the image features to obtain a plurality of image block groups; s4, respectively constructing a weighted norm model for each image block group, and establishing a constraint model of each image block group based on the weighted norm model; s5, constructing a Lagrange function based on the constraint model of each image block group; s6, performing solution inversion on the Lagrange function of each image block group, and performing rearrangement to obtain a background tensor sequence and a target tensor sequence; and S7, performing clustering processing on the background tensor sequence and the target tensor sequence according to the sequence order to obtain a final background tensor B and a final target tensor T. The objective of the invention is to improve the robustness of background low rank.

This invention discloses a classroom scene object detection method based on multi-scale feature enhancement and semantic alignment, implemented according to the following steps: Step 1, obtain a classroom scene object detection dataset; Step 2, improve the YOLOv11 one-stage object detection network to obtain a student classroom scene object detection model; Step 3, train the student classroom scene object detection model using the classroom scene object detection dataset obtained in Step 1 to obtain a trained student classroom scene object detection model; Step 4, input the student classroom scene image to be detected into the trained student classroom scene object detection model for detection. This invention solves the problem in existing technologies where insufficient feature extraction and feature fusion mechanisms make it difficult to meet the accuracy requirements of actual classroom scene object detection.

The invention provides a pipeline oil leakage detection method and system based on three-mode physical-visual synergy, and relates to the technical field of pipeline safety detection. The method comprises the following steps: acquiring a visible light image, an infrared image and a laser echo feature sequence of a to-be-detected pipeline area, performing spatial pixel-level alignment and normalization processing, and inputting the processed image into a pipeline oil leakage detection model comprising a shared parameter backbone network, a one-dimensional convolution branch, a cross-modal interactive fusion module and a Transformer decoder; the backbone network extracts visible light and infrared image features, and the one-dimensional convolution branch extracts laser spectrum features; the cross-modal interaction fusion module fuses three-modal features through three-directional attention and constructs a feature pyramid; and the Transform decoder decodes the feature pyramid and outputs an oil leakage detection result. According to the invention, through introducing laser physical characteristics and visual modal deep fusion, material attribute discrimination and visual positioning cooperative detection of an oil leakage target are realized, and the accuracy and robustness of pipeline oil leakage detection in a complex environment are improved.

The invention relates to the technical field of data processing, in particular to a cigarette packet paperboard production data processing method and system based on intellectualization, and the method comprises the steps: obtaining technological parameters in a cigarette packet paperboard production process, inputting the technological parameters into a preset quality prediction model, and obtaining predicted quality indexes; calculating the process composite fluctuation degree of the process parameters; calculating a long-term deterioration index of the cigarette packet paperboard; calculating an adaptive reliability coefficient of the quality prediction model; calculating a short-term quality risk index, wherein the short-term quality risk index is in positive correlation with the product of the adaptive reliability coefficient and the quality index; and when the short-term quality risk index exceeds a preset risk threshold value, indicating that the quality defect is about to occur, and sending out a shutdown signal. According to the invention, self-adaptive quality monitoring and early warning in the cigarette packet paperboard production process are realized, the rejection rate and the production operation cost are reduced, and the accuracy of the monitoring result is improved.

This application provides a log detection method, apparatus, storage medium, and computer device. Addressing the challenges of existing security log analysis systems—namely, their inability to detect covert attacks with identical cross-entity semantics due to their "user / IP + timeline" perspective, their lack of behavioral semantic abstraction capabilities due to graph analysis granularity remaining at the log instance level, their high false positive rate due to ignoring differences in log source environments, and their inability to continuously adapt to new variant attacks due to the lack of a closed-loop feedback mechanism—this application proposes semantic behavior clustering, environment-weighted correction, large language model intent inference, cross-domain propagation prediction, and feedback evolution to achieve accurate and adaptive malicious behavior identification from logs.

The application discloses a kind of lead seal thin wall structure's ultrasonic detection signal processing method, device and computer program product, method includes: obtaining the ultrasonic A scanning signal covering the region to be measured;Based on the signal characteristics of the defect-free area in A scanning signal, calculate adaptive detection threshold;To pre-process signal repeatedly executes iterative signal decomposition processing, until it satisfies the convergence condition based on the adaptive detection threshold.The iterative signal decomposition processing includes: based on the standard reference echo model of pre-set, at least two candidate echo components are positioned and constructed in current signal to be processed;Select a candidate echo component as effective echo component by competition mechanism;From current signal to be processed, the effective echo component is subtracted, and new signal to be processed is generated.The application is combined by statistical noise decoupling and competitive nonlinear iterative stripping, can effectively separate multiple defect echoes of time domain serious aliasing under strong noise background, improve the axial resolution of thin wall structure.

The application relates to the field of equipment state monitoring, and particularly discloses an oil depot pump set abnormality identification system and method, which comprises a multi-type data acquisition module, an adaptive threshold adjustment module, a multi-modal deep learning fusion identification module, an abnormality trend prediction module and a correlation decision support module; the multi-type data acquisition module is used for acquiring vibration signals, physical state signals of an oil depot pump set and characteristic parameters of an operation environment and a load where the oil depot pump set is located, converting the vibration signals into two-dimensional time-varying spectrum features, converting the physical state signals into standard time sequence data, and transmitting the standard time sequence data to the multi-modal deep learning fusion identification module; the technical scheme can reduce false alarm and missed alarm probabilities, realize deep fusion of data body features of multi-type sensors, and can guide maintenance decisions according to state monitoring.

The application discloses a breath methane and hydrogen synchronous detection method and system based on an improved gas sensor, and relates to the biomedical detection field.The method comprises the following steps: using the improved gas sensor to synchronously collect a breath flow signal and a sensor original electric signal at a set sampling rate; identifying a breath alveolar air plateau according to the breath flow signal, and intercepting a corresponding sensor original signal segment; performing filter denoising and baseline correction preprocessing on the sensor original signal segment of the plateau; based on the hydrogen and methane response time difference amplified by the gas diffusion delay structure of the improved gas sensor, extracting a first characteristic quantity related to the hydrogen concentration and a second characteristic quantity related to the methane concentration from the preprocessed plateau signal segment; and calculating and outputting the hydrogen concentration value and the methane concentration value according to the first characteristic quantity and the second characteristic quantity.The application can synchronously detect breath methane and hydrogen based on the sensor diffusion delay structure, and has the advantages of low cost, portability and high detection reliability.

This invention provides a panel yield detection method and system based on machine learning and image processing, belonging to the field of yield detection technology. The method flow is as follows: edge detection is performed on the lamp-lit image of the display panel to be tested to obtain an edge image; connected component analysis is performed on the edge image to extract candidate regions of the panel; the maximum bounding rectangle of the candidate regions of the panel is constructed, and the geometric features of the maximum bounding rectangle are calculated; the geometric features are input into an isolated forest model to calculate the size anomaly score, and the existence of size anomalies is determined based on the anomaly score; color features are extracted from the candidate regions of the panel, and the existence of color anomalies is determined based on the color features; the results of the size anomaly and color anomaly determinations are combined to output the defect detection result of the display panel to be tested. This invention transforms the abnormal display problem into geometric integrity detection and color consistency detection, combining traditional image processing with the isolated forest algorithm, eliminating the need for a large number of anomaly sample annotations, and exhibiting strong adaptability and high accuracy.

This invention belongs to the field of medical image analysis technology and relates to a lesion detection method and system based on multi-scan interactive deformable Mamba, including: 1. image patch embedding; 2. multi-scale feature extraction and interaction in the parallel scanning backbone network; 3. feature pyramid optimization based on dynamic weighted scan fusion; 4. multi-scale fusion. This invention effectively solves the characterization problems of morphological heterogeneity of oral cancer lesions and fine structural features of lung nodules through adaptive scanning network and deformable scanning mechanism, achieving a dual breakthrough in detection accuracy and computational efficiency, and providing an efficient and reliable solution for multi-cancer medical image analysis.

This invention discloses an abnormal transaction detection method based on dynamic adaptive online contrastive learning. The method includes: using a learnable data augmentation module to adaptively generate a feature mask based on the importance distribution of node features, and outputting an augmented view; inputting the original node features and the augmented view into a spatiotemporal graph neural network, using a directed graph aggregation layer, a structure-attribute gating mechanism, and a time evolution module to capture the spatiotemporal dynamics of the transaction graph, and outputting spatiotemporal features; constructing a semi-supervised consistency learning framework based on a mean-based teacher architecture, and training the student model using a combination of supervision loss, consistency loss, and contrastive loss; during the testing phase, using a drift detector to determine whether feature distribution drift has occurred at the current time step, using an evaluator to filter high-confidence pseudo-label samples to fine-tune the teacher model online, and outputting the abnormal transaction prediction probability for each node to complete the abnormal detection.

A small target detection method, system, device and medium based on a local attention feature correlation mechanism, the method comprising: data preprocessing, model construction, model training, small target detection; the system, device and medium are used to realize a small target detection method based on a local attention feature correlation mechanism; the application extracts spatial features and semantic features through a residual network, performs feature fusion through a feature pyramid mechanism, reconstructs features based on a local attention feature correlation mechanism, trains a model using a Hungarian matching algorithm, obtains an optimal matching model suitable for a data set, can ensure high accuracy with fewer iteration times, and has the characteristics of low network operation complexity, low training cost, easy model migration, and high model detection performance.

The application discloses an emergency rescue real-time human body detection method and equipment based on timing motion feature enhancement, which comprises the following steps: acquiring visible light, infrared image sequences and environmental parameters (including smoke concentration and light intensity) of a rescue area in real time, and performing spatial registration preprocessing; performing frame difference processing on the two types of image sequences respectively, generating binary motion masks, calculating optical flow amplitude, and obtaining a multi-scale motion energy field according to adaptive fusion of the smoke concentration; extracting human body micro-motion frequency band energy through time-frequency conversion, generating a frequency domain dynamic attention mask, and combining the multi-scale motion energy field to extract a significant motion target area; constructing a double-branch neural network, extracting timing motion features and multi-modal appearance features, dynamically distributing weights and fusing, and outputting a human body bounding box and a detection confidence; calculating the environmental complexity according to the environmental parameters, determining a dynamic confidence threshold, verifying the detection result in combination with the average temperature of the human body bounding box area, and generating an alarm information if the condition is met. The application aims at solving the problems of high missed detection rate and unstable recognition of traditional methods in complex environments.

The present application relates to a kind of detection method and device of traffic signal light state under vibration condition, solve the technical problem that signal light state identification is unreliable under vibration condition.Method includes: the image preprocessing of signal light video under vibration condition is carried out to overcome input interference and form the image sequence of detection;Through target detection model, signal light in input image is carried out target detection to obtain positioning area and detection confidence, and the statistical characteristic analysis of RGB channel is carried out to signal light by positioning area;According to detection confidence and statistical characteristic, the fusion probability determination of signal light state is carried out;According to the state time sequence of signal light, signal light interframe state is smoothed, and state conversion constraint is established to verify signal light state and output.Multiple-source information probability fusion can improve detection robustness, and modal feature fusion of cross-appearance information and structure information can effectively suppress noise, and improve the sensing performance in complex scene.

The application discloses a kind of concrete hidden crack ultrasonic thermal excitation system optimal working point self-adapting configuration method, comprising: by constructing the parameter space including initial coupling pressure and working frequency, acquisition system is under different parameter combination Multiple physical field response data;Based on transducer active power response or thermal image temperature difference, construct excitation efficiency evaluation index, lock initial optimal working point by identifying the extreme characteristics of index response function;Using Hessian matrix to the neighborhood of optimal point carries out second-order sensitivity analysis, constructs tolerance ellipse model to quantify parameter control boundary;While based on the power-phase decoupling mechanism of voltage and current Real-time monitoring frequency and pressure drift.The application can be matched with optimal working parameters adaptively for different stiffness loading device, scientifically set engineering tolerance, improve the energy conversion efficiency and detection robustness of ultrasonic thermal excitation system.

The invention discloses an improved YOLOv8-based hydraulic cylinder structure size identification method and system, and the method comprises the steps: constructing an improved YOLOv8 target detection model: embedding a direction sensitive convolution module in front of a spatial pyramid rapid pooling module of a backbone network, introducing an external attention mechanism into a cross-stage partial fusion module, and carrying out the detection of the target detection model; a multi-scale attention aggregation module is integrated in the neck network; performing multi-size cooperative detection on a hydraulic cylinder drawing by using the trained model, identifying various installation size areas, and generating an identification image; segmenting each size region, extracting a structure-installation size association sub-graph, identifying a size text in the structure-installation size association sub-graph through an optical character identification model, and generating structured association data; checking, matching and associating the associated data based on a predefined installation size mapping function; and fusing the recognition and verification results, performing multi-level verification based on the geometric constraint of the hydraulic cylinder structure and a tolerance accumulation rule, and generating a natural language output result.

The invention relates to an information security technology, in particular to an intelligent contract vulnerability detection method and system based on comparative learning and globe fine tuning, and the method comprises the steps: extracting a corresponding embedded vector from a to-be-detected contract through a pre-trained embedder; performing granular ball clustering on a data set formed by the embedded vectors of all the to-be-detected smart contracts, namely splitting the data set into a plurality of sub-granular balls; and inputting the grain ball center of each grain ball into a classifier formed by a two-layer and multi-layer sensor to predict whether the smart contract in the grain ball has vulnerabilities or not. According to the method, on the basis of improving the robustness of intelligent contract vulnerability detection in a noise environment, the sample quality and the adaptability to different types and scales of intelligent contracts are considered, the vulnerability detection requirements of large-scale intelligent contracts can be efficiently met, and reliable technical support is provided for block chain platform intelligent contract security auditing.

This invention provides a vehicle blind spot identification method and system based on millimeter-wave radar. The method includes transmitting a frequency-modulated continuous wave signal to the vehicle blind spot using an onboard millimeter-wave MIMO array radar and receiving echo signals from the vehicle blind spot to obtain raw radar echo data; performing polar coordinate projection imaging processing on the raw radar echo data to obtain an initial polar coordinate domain image; performing array error compensation on the initial polar coordinate domain image to obtain a compensated image; and performing blind spot target identification on the compensated image to obtain blind spot target parameters. This invention can significantly reduce the computational load, effectively correct array errors, improve detection robustness under multi-target and clutter edge conditions, and ultimately achieve low-cost, high-reliability blind spot warning.

The application discloses a multi-modal depression detection method based on space-time frequency domain enhancement and mutual attention mechanism, and specifically comprises the following steps: step 1, preprocessing of multi-modal perception data of depression subjects; step 2, joint space-time frequency domain feature mining; step 3, frequency spectrum gating guided speech enhancement; step 4, landmark guided visual semantic reconstruction; step 5, three-way parallel mutual attention cross-modal fusion; step 6, spatial enhancement and classification discrimination; step 7, sample set decision statistics and global index evaluation. The method solves the problems of weak facial and speech subtle feature capturing ability, insufficient video and audio cross-modal correlation mining and original data susceptible to background noise interference.

The application relates to the technical field of target detection, in particular to an infrared-visible light target detection method based on a dual-domain cooperative YOLO, which comprises the following steps: acquiring an infrared image and a visible light image; inputting the infrared image and the visible light image into a DS-YOLO model based on a dual-domain cooperative YOLO; and outputting a target detection result through the DS-YOLO model, wherein the DS-YOLO model is built based on a YOLOv8 architecture, a feature pyramid interaction network is used to replace an original neck network of the YOLOv8, a frequency domain cooperative alignment module is designed for shallow features, a spatial expansion fusion module is designed for deep features, and a dual-domain cooperative fusion architecture is designed for intermediate features. The application effectively enhances cross-modal complementarity, improves detection precision and robustness under complex and changeable conditions, reduces multi-modal feature detail loss and misplacement to a certain extent, and improves the accuracy of target detection.