69 results about "Preprocessing algorithm" patented technology

The invention relates to the technical field of computer vision and edge computing, and discloses a multi-computing-power collaborative video structured analysis method, which comprises the following steps of: acquiring a plurality of paths of original video streams, and synchronously decoding the plurality of paths of original video streams; gradient analysis is adopted to evaluate the decoded multiple paths of original video streams, and a preprocessing algorithm combination is selected based on a quality feature vector; establishing a data transmission channel, and adopting an asynchronous communication protocol and a multi-level cache system; performing content complexity analysis, and performing model matching according to the content complexity analysis to obtain a scheduling decision; carrying out distributed deployment, and carrying out pipeline reasoning on the deployed model; pipeline reasoning results are fused, and consistency constraint verification is carried out on fusion results; optimizing system configuration parameters according to a video structured analysis result to realize video structured analysis; the real-time performance, the accuracy and the system stability of video analysis are improved.

A method for analyzing datasets for vulnerabilities to adversarial attacks includes storing, by a processing server, a dataset comprising multiple data types in a database; preprocessing the dataset using data preprocessing algorithms based on the data types; performing data featurization to identify dataset features; generating, for each data type, a victim model based on the dataset; determining performance metrics of the victim models; generating attack datasets corresponding to each data type and including the performance metrics and dataset features; executing adversarial attacks, including membership inference, property inference and reconstruction attacks, on the victim models; determining adversarial attack success metrics, including accuracy, precision and F1 score; updating the attack datasets to include the adversarial attack success metrics; and generating, for each updated attack dataset, a vulnerability metric comprising a library of algorithms and models specific to the combination of adversarial attack and data type.

The invention discloses a radiation source data intelligent distillation method and application thereof, and belongs to the field of communication, unmanned aerial vehicles and artificial intelligence. According to the method, an unmanned aerial vehicle data set is collected and established, a time-frequency domain comprehensive anti-interference feature extraction method based on real and virtual part separation and a CV-ResNet classification model are constructed, a data preprocessing algorithm and the classification model are subjected to module integrated distillation, and after the validity of the algorithm is verified, the distillation model is applied to an unmanned aerial vehicle load. Through integrated distillation, soft target output inverse operation, graph feature operation, EGCN graph feature representation and a feature point edge loss function, algorithm embedded application is realized, the recognition efficiency and the attack countermeasure success rate are improved, the load weight of the unmanned aerial vehicle is reduced, and the endurance and comprehensive efficiency are improved.

The invention provides a transformer abnormity monitoring method and system based on artificial intelligence, relates to the technical field of transformer state monitoring, and solves the technical problems of poor scene adaptability, multi-source data isolation and weak early warning in the prior art. The method comprises the following steps: acquiring real-time data and historical data of a transformer; preprocessing the real-time data through a preprocessing algorithm to obtain standard data; constructing an AI anomaly monitoring model based on historical data; wherein the AI anomaly monitoring model comprises a dynamic baseline module, an anomaly detection module and an anomaly classification module; and inputting the standard data into the AI anomaly monitoring model to generate an anomaly alarm. The method is used in the transformer abnormity monitoring process.

The invention discloses a stand-oriented adaptive spectrum diagnosis method and system, and belongs to the technical field of hyperspectral remote sensing. The method comprises the following steps: firstly, dividing hyperspectral data into a plurality of classified spectrum subsets by adopting unsupervised clustering; then, aiming at each subset, adaptively optimizing a spectrum preprocessing algorithm, and extracting an optimal characteristic wave band by adopting a hybrid strategy of combining a competitive adaptive reweighting algorithm (CARS) and a continuous projection algorithm (SPA); and finally, a dynamic weighted ensemble learning model is constructed, a meta-learner of the dynamic weighted ensemble learning model not only fuses prediction results of a plurality of heterogeneous base learners, but also introduces spectral features and category identifiers of samples to make decisions, so that dynamic weighting of the base learners is realized, and a final diagnosis result is output. According to the method, the technical problems of fixed feature extraction process and poor robustness of a single model in the prior art are solved, and the precision, robustness and generalization ability of the forest stand spectrum diagnosis model are remarkably improved.

The invention discloses a core drilling method automatic recording and defect intelligent identification method based on deep learning, and relates to the technical field of core sample defect intelligent discrimination. Core sample photos shot on site are acquired, and a deep learning model is used for performing a series of processing; comprising the steps of determining boundary points, extracting angular points, preprocessing photos, extracting return information, identifying defects and the like, and finally generating core sample catalog information and a drilling comprehensive histogram. The method is technically characterized in that a corner point recognition model, a return extraction model, a defect extraction model and a defect classification model are combined with algorithms such as a rotary clamping shell and image preprocessing, and automatic analysis and recognition of a core sample photo are achieved; according to the method, various defect types and degrees can be accurately identified, an efficient, accurate and visual catalog result and visual display are provided for engineering detection, and the detection efficiency and reliability of a core drilling method are remarkably improved.

The invention discloses a cognitive function intervention method and system based on multi-mode space-time dynamic decoding biological feedback, and belongs to the technical field of artificial intelligence algorithms and noninvasive nerve regulation. According to the method, brain nerve activity signals and autonomic nerve cooperation signals are captured in real time, the signals are subjected to standardization processing by adopting a multi-modal signal cascade preprocessing algorithm, and cognitive feature extraction of fusion signals and dynamic distribution of multi-modal signal feature weights are realized in combination with a circulating double-flow self-encoding model; and thus, the individual cognitive state is accurately evaluated. And based on the evaluated cognitive state, dynamically identifying and quantifying into self-perceived visual and auditory stimuli, and constructing an iterative closed-loop feedback cognitive nerve plasticity regulation loop to realize individualized targeted regulation of the cognitive function. The method is suitable for non-invasive individualized intervention of the cognitive function of the Parkinson's disease patient, and has remarkable clinical application potential and social value.

The application discloses a kind of qualitative detection method, equipment and medium of the surface micro linear defect of magnetic tile, it is related to the nondestructive testing technical field of magnetic material.The scattering characteristics of different defects are highlighted by dynamic adjustment of light source azimuth, a multi-directional illumination imaging system is constructed, which includes low-angle linear light source and high-resolution polarization camera.In the middle, in addition to the light source, the camera moves along the axial direction through a stepper motor, and dynamic image acquisition is performed.A preprocessing algorithm based on direction gradient histogram enhancement is designed, and a deep convolutional neural network is used to analyze the multi-scale feature fusion of the defect area.A lightweight classification model based on transfer learning is developed to accurately distinguish micro-cracks and pseudo-defects.The supporting equipment integrates a light source control module, an image acquisition unit, and an embedded processing platform, supporting real-time dynamic detection.The application breaks through the detection limit of sub-millimeter linear defects by traditional machine vision, and the miss rate is reduced to below 0.5%.

The application discloses a kind of coal mine goaf monitoring method and system, it is related to coal mine goaf monitoring technical field, including the collection coal mine goaf subsidence characteristic data;Using sound processing technology, sound characteristic data are preprocessed, and subsidence monitoring data are denoising processed;Through weight analysis module, safety factor is calculated based on preprocessed data, and data is transmitted to terminal monitoring system in real time.The coal mine goaf monitoring method provided in the application uses sound processing technology and preprocessing algorithm, effectively preprocesses the sound characteristic data and other monitoring characteristic data collected, improves the accuracy and reliability of data;Sound processing technology includes noise elimination, sound enhancement, echo suppression and fast fourier transform, can effectively extract sound characteristics, provide strong support for subsequent analysis.The application achieves better results in reliability and analysis accuracy.

The invention provides a video stream lag intelligent optimization method and device based on AI prediction. The method focuses on the following core links: a data acquisition and feature engineering method covering multilayer network state monitoring, video content deep analysis, user behavior sequence modeling and a real-time data preprocessing algorithm; the AI prediction engine core algorithm comprises an improved Transform architecture, a cross-modal feature fusion mechanism and a multi-scale prediction model; the intelligent optimization decision-making mechanism relates to a reinforcement learning optimization framework, a self-adaptive buffer strategy and an intelligent code rate switching and predictive preloading method; the system architecture and engineering implementation comprises an end-to-end real-time processing architecture, a lightweight model deployment technology, a multi-platform compatible interface and a real-time feedback and tuning mechanism.

The invention discloses an intelligent prediction method of a human resource system operation management budget system. Comprising an intelligent data collecting and processing module which is connected with each data source through an API interface, automatically collects employee attendance records and salary information, and performs cleaning and standardization processing on data through a data preprocessing algorithm to ensure the accuracy of the data; the multi-dimensional data analysis and modeling module realizes accurate prediction of human resource management budget by combining a random forest algorithm and a near-end strategy optimization PPO algorithm; the random forest algorithm is used for preliminarily analyzing and extracting data features, and the PPO algorithm is used for further optimizing a prediction strategy, so that the accuracy of model prediction in a dynamic environment is improved; the dynamic budget adjustment mechanism automatically adjusts the budget scheme according to market changes and real-time data, the human resource management efficiency and decision scientificity can be effectively improved, the operation cost is reduced, the data safety is guaranteed, and reliable management tools and technical supports are provided for enterprises.

This invention provides an intelligent analysis system for photoelectric detection spectral data based on deep learning, belonging to the fields of photoelectric detection, spectral data analysis, and deep learning application technology. The system includes core units such as a spectral data acquisition unit, a preprocessing unit, a feature extraction unit, and a deep learning analysis unit, as well as auxiliary units for anomaly detection and data storage, forming a closed-loop architecture of acquisition-processing-analysis-output-optimization. It improves data quality through step-by-step preprocessing algorithms, reduces computational load by adaptively selecting key features, deeply mines spectral patterns using a model that integrates convolutional neural networks and attention mechanisms, achieves rapid scenario adaptation through transfer learning, and dynamically optimizes model parameters based on a feedback mechanism. The system solves the problems of low accuracy, poor adaptability, and long training cycles in traditional spectral analysis, achieving accurate, efficient, and intelligent spectral data analysis, and is suitable for various photoelectric detection scenarios such as qualitative and quantitative analysis of substances and environmental monitoring.

Classification and preprocessing algorithms are applied to at least one reference document and a natural language command provided as input by a user to a development environment for computer programming. Approximate matching of keywords in user commands with data in the reference document is performed with a fuzzy hash map and used to develop information metrics that produce a ranking of candidate actions. Selection criteria are then used to select an action from multiple candidate actions to perform the desired operation in the development environment.

The present application belongs to the technical field of image processing, and in particular to an improved 3A image preprocessing algorithm based on DM8127. Through the research on the automatic focusing algorithm, the definition evaluation function is analyzed and compared, the search mode for finding the optimal value is improved, the search times are reduced, and better search effect is achieved. Combined with the influence of the aperture on the depth of field blur and the influence of the gain on the noise, a new automatic exposure algorithm is proposed based on the quality evaluation function, and the best comprehensive image quality balance between the depth of field blur and the noise is achieved. In the automatic white balance part, the image is divided, the gray area is found through the YUV component, and the parameters are adjusted on the DM8127 platform to make the result converge, which achieves better effect than the commonly used automatic white balance algorithm.

The invention relates to the technical field of hydrological modeling, in particular to a topographic data priority queue optimization system and method.In the system, a data grouping storage unit is used for obtaining elevation values corresponding to all grid units in depression processing of a digital elevation model, and the grid units are grouped and stored according to the elevation values through a hash table; and mapping the grid units with the same elevation value to the same hash bucket. According to the method and the device, the Hash table and the second data structure are combined to construct the mixed priority queue structure, so that the nodes are grouped and stored according to the priorities through the Hash table, the orderliness among different priorities is maintained through the second data structure, and node memories are pre-allocated in cooperation with a memory pool pre-allocation strategy. According to the method, the performance bottleneck of a priority queue in a traditional depression preprocessing algorithm is effectively broken through, the depression processing efficiency under a large-scale DEM data set is remarkably improved, and a reliable algorithm optimization scheme is provided for processing mass topographic data.

The invention discloses a drain valve fault online monitoring method and system based on voiceprint analysis, and relates to the technical field of industrial equipment state monitoring, and the method comprises the steps: employing a blind source separation algorithm based on REPET to carry out the preprocessing of a voiceprint signal, and removing the interference of environment noise; and voiceprint features are extracted by using an MFCC method, and fault diagnosis is performed by using a GMM model. According to the invention, real-time and accurate monitoring of the fault of the drain valve is realized, the accuracy rate of voice print abnormity recognition of faults such as inner leakage and blockage of the drain valve is high, the signal processing and recognition time is short, the misjudgment rate is low, the efficiency and reliability of fault diagnosis of the drain valve are effectively improved, and a powerful guarantee is provided for safe and stable operation of a steam turbine. For complex field environments such as noisy thermal power plants, narrow space and the like, the method shows excellent adaptability. The advanced voiceprint preprocessing algorithm can effectively filter environmental noise and accurately focus on the voiceprint features of the drain valve.

The application discloses a kind of pre-processing algorithms for remote sensing image coding.The algorithm is mainly to extract image texture feature parameters, extract image texture feature parameters, according to the texture feature parameter bypasses the prediction mode and the division mode with smaller probability.The algorithm mainly includes image texture feature extraction;Fast rate distortion cost estimation;Intra-frame division prediction;Determine candidate prediction mode and division set.The present application, especially for hardware difficult to realize the H.265 encoding algorithm of remote sensing image.The algorithm saves hardware resources, reduces hardware implementation complexity and difficulty under the premise of trying to guarantee image compression quality, effectively utilizes hardware storage resources, and also reduces the time delay of compression and decoding.

This invention discloses an energy-saving motor speed data acquisition method and system based on cloud computing, belonging to the field of data acquisition technology. The method includes: acquiring factory production plans, personnel shift schedules, and historical motor operation data; generating production load level information based on the factory production plans, personnel shift schedules, and historical motor operation data; encoding the production load level information to obtain rhythm commands; and sending the rhythm commands to a local data acquisition terminal, which adjusts data acquisition parameters according to the rhythm commands and performs motor speed data acquisition processing based on the data acquisition parameters to obtain motor speed data. The data acquisition parameters include operating frequency, data sampling rate, and preprocessing algorithm complexity. This invention can combine production load level information with rhythm commands to adjust data acquisition parameters to achieve motor speed data acquisition, reducing the power consumption and cost of acquisition equipment.

The application discloses a near-infrared quantitative model construction method and device and a storage medium, wherein the method comprises the following steps: acquiring near-infrared spectrum data, and constructing a near-infrared quantitative model training sample set; acquiring a plurality of near-infrared spectrum data preprocessing algorithms, variable selection algorithms and modeling algorithms and the value range of parameters thereof, and constructing a particle swarm search space composed of the parameters and the near-infrared spectrum data preprocessing algorithm sequence; based on the near-infrared quantitative model training sample set, the parameters of each algorithm and the preprocessing algorithm sequence are optimized in the particle swarm search space by using a particle swarm optimization algorithm; and the global optimal solution obtained by selection is used as the parameters of each algorithm and the preprocessing algorithm sequence, near-infrared spectrum data is preprocessed, variable selection and modeling are performed, and a near-infrared quantitative model is obtained. The modeling requirement is reduced, and a high-quality quantitative model can be simply and efficiently obtained on the basis of the data modeling experience of researchers.

The invention discloses a Bayer image parallel preprocessing method and a Bayer image parallel preprocessing system, which belong to the technical field of image processing, and adopt a parallel mode to carry out multiple preprocessing operations on a Bayer image so as to reduce the delay of the whole Bayer image preprocessing part. Image fusion is added in the Bayer preprocessing unit, weight parameters of all preprocessing modules are increased, weight adjustment can be carried out on all Bayer preprocessing algorithms through the communication control unit so as to control the algorithm composition of the image preprocessing unit, image fusion is carried out in an HSV color space, the gray scale and the color can be processed separately, and the image fusion efficiency is improved. The system comprises a Bayer image output platform, an image processing algorithm platform and a display.

The invention provides a medical equipment power distribution monitoring method and system based on artificial intelligence, relates to the technical field of medical equipment power management, and solves the technical problems that the detection precision is insufficient and potential risks are lack of intelligent prediction in the existing medical equipment power distribution monitoring process. The method comprises the following steps: acquiring power supply data and resolving the power supply data to obtain power supply parameters; pre-processing the power supply parameters through a pre-processing algorithm to obtain pre-processed data; performing harmonic analysis on the preprocessed data through an FFT algorithm to obtain harmonic parameters; performing anomaly detection on the preprocessed data and the harmonic parameters based on threshold comparison to obtain a real-time anomaly result; inputting the preprocessed data and the harmonic parameters into a pre-trained risk prediction model to obtain a prediction risk result; and performing weighted fusion decision on the real-time abnormal result and the prediction risk result to generate an abnormal alarm signal. The method is used in the medical equipment power management process.

The invention relates to the technical field of intelligent monitoring, in particular to a power plant production safety event dynamic monitoring system based on video analysis and an AI algorithm, and the system comprises a data acquisition module which is used for collecting video image data of each channel; the system comprises a data acquisition module used for acquiring original image data, a data processing module used for preprocessing the acquired original image data, an AI algorithm module used for performing target identification and classification by using a deep learning algorithm model according to the input processed data, and an algorithm training module used for training the model in the AI algorithm module by using marked training data provided by the data processing module. The AI algorithm module applies a deep learning algorithm model to perform feature extraction and pattern recognition on the input video and production data, so that the predicted security event bounding box and category probability can be accurately output, and the running state of equipment, the operation behavior of personnel and various indexes of the environment can be comprehensively and carefully monitored.

The invention discloses a kitchen open fire hidden danger early warning model based on YOLO lightweight, and relates to the technical field of artificial intelligence, and the model comprises the following steps: S1, obtaining an original monitoring image of a kitchen environment, employing a preprocessing algorithm fusing bilateral filtering and adaptive histogram equalization, carrying out the denoising and enhancement of the original monitoring image, and outputting a standardized image; and S2, based on the standardized image, applying a structured pruning strategy to a pre-trained YOLOv5 model, and removing redundant convolutional layers and channels in a backbone network of the pre-trained YOLOv5 model. According to the kitchen open fire hidden danger early warning model based on YOLO lightweight, through a preprocessing mode of fusing bilateral filtering and adaptive histogram equalization, image noise is removed, flame edges and key features are completely reserved at the same time, the anti-interference capability of flame recognition in a complex kitchen environment is greatly improved, and the method is suitable for popularization and application. Clear risk guidance is provided for the user, the hidden danger can be quickly handled, and the kitchen fire protection level is comprehensively improved.

The invention provides a noise deafness occupational disease risk detection system and method, and relates to the technical field of occupational health intelligent monitoring, and the system comprises a data collection module, a data preprocessing module and a risk prediction module. The method comprises the steps that noise data are collected based on a data collection module, the noise data comprise LAeq time sequence data, LCPeak data, 1 / 3 octave spectrum data, exposure duration and environmental parameters, the collected noise data are preprocessed based on a data preprocessing module, and a risk prediction model is constructed based on an improved CNN-GRU network structure. And performing noise deafness occupational disease risk prediction based on the risk prediction model. The industrial scene adaptability is improved based on hardware innovation, and accurate data are provided; the data validity is ensured based on a preprocessing algorithm; and finally, realizing accurate risk prediction based on a CNN-GRU model.

The invention discloses a multimedia interaction control system and method based on an OPS microcomputer, and the system comprises an input collection module which is used for collecting a user physical interaction signal through a multi-sensor fusion technology, and generating a standardized digital interaction data set based on a signal preprocessing algorithm; the core processing module is used for identifying an interaction behavior mode through a deep learning feature extraction algorithm based on the digital interaction data set and generating an interaction control parameter with a time sequence label; the content rendering module is used for calling a three-dimensional model resource library and executing real-time geometric transformation calculation based on the interaction control parameters to form an image frame sequence with a dynamic response characteristic; and the output driving module is used for driving heterogeneous display equipment through a multi-interface adaptive conversion technology based on the image frame sequence to generate a synchronous multimedia interaction display effect. And audio and video synchronous output is ensured through a frame synchronization technology, and a smooth multimedia interaction display effect is finally realized.

The invention discloses a method, a system, equipment and a medium applied to agricultural pest and disease damage prediction, mainly relates to the technical field of agricultural pest and disease damage, and is used for solving the problems of insufficient data timeliness and integrity, lack of dynamic modeling capability and poor algorithm suitability of an existing scheme. Comprising the steps of obtaining agricultural multi-source data collected by preset multi-source equipment, and dynamically adjusting a data uploading interval of the preset multi-source equipment according to the agricultural multi-source data; determining a preset preprocessing algorithm of the agricultural multi-source data based on the specific type of the agricultural multi-source data, and obtaining preprocessed agricultural multi-source data; constructing a CNN-LSTM-Attention cascade architecture algorithm, and obtaining a trained cascade architecture algorithm by taking combined data of agricultural multi-source data, data source weight and causal relationship as a training sample; and inputting the collected agricultural multi-source data into the trained cascade architecture algorithm to obtain agricultural pest and disease prediction data.

The application discloses a plug-and-play voice confrontation sample defense method, first, the input voice is smoothed by Gaussian noise to generate a smooth voice sequence with subtle differences, then the smooth voice sequence is input into a multi-version preprocessing algorithm module, different implementation compression algorithms and enhancement algorithms are used to process the smooth voice to generate a multi-version voice sequence; then the generated multi-version voice sequence is input into a voice recognition system to obtain a corresponding voice transcription text sequence; finally, a mode voting weight allocation (MVWA) method is used to allocate weights to each text sequence, and then a recognizer output voting error reduction (ROVER) algorithm is used to estimate the benign transcription of the input voice according to the weights. Through the application, the aggressive confrontation sample can be restored to a benign sample, the safety performance of the voice recognition system is improved, and an effective restoration method is provided for the defense of the confrontation sample.

The invention belongs to the technical field of fingerprint lock intelligent control, and particularly relates to a high-adaptability fingerprint identification intelligent closet lock and a control method, the control method comprises a fingerprint identification method, and the fingerprint identification method is characterized in that an optical combination capacitance dual-mode fingerprint sensor is combined with environment adaptive switching logic; in cooperation with preprocessing algorithms such as adaptive filtering, directional diagram correction and detail enhancement, the image feature point extraction rate in a complex finger state is effectively improved, meanwhile, a first threshold is combined with a second threshold hierarchical judgment and different finger secondary verification mechanism, and personalized and environment adaptive threshold optimization is adopted for assistance. Under the condition that the safety standard of the error acceptance rate is kept, the passing rate of legal users is improved, common fake fingerprints are recognized through living body detection, and the passing rate is improved while fingerprint recognition is ensured to be passed.

This invention discloses a post-processing method for multi-level iterative collaborative representation of derained images, comprising the following steps: obtaining an initial image set, which includes several sets of derained images and corresponding rainless images, wherein the derained images are images obtained after a deraining preprocessing algorithm on rainy images, and the rainless images are images collected for the corresponding scene that do not contain rain streaks; constructing an algorithm matrix to remove mismatches; obtaining the images after removing mismatches, and using them to construct deraining network data, and obtaining the final derained image based on it. This invention combines the Hessian-Affine and SIFT algorithms, focusing on the strong correlation between feature point neighborhood information and feature points, resulting in higher feature matching accuracy during feature matching, and providing more refined assistance for image post-processing.

The invention discloses a foreign language control intelligent bed system. The system comprises a multi-language voice acquisition and preprocessing module, a semantic analysis and instruction mapping module, a bed body posture driving control module, a multi-dimensional safety monitoring module, an environment perception and self-adaptive adjustment module and a user state feedback and storage module. Through the multi-language pre-training model and the precise voice preprocessing algorithm, precise recognition and analysis of multiple common languages are achieved, the problem that an existing intelligent bed is single in language adaptation is solved, and the use requirements of users of different languages are met; a target attitude angle calculation model is constructed in combination with user body parameters and instruction requirements, multi-section attitude personalized adjustment of the bed body is achieved, and the precision and stability of attitude adjustment are improved in cooperation with a closed-loop control mechanism; a multi-dimensional safety monitoring system is constructed, pressure distribution, edge distance and motion state monitoring are integrated, rapid identification and protection of an abnormal state are realized through a safety threshold calculation model, and the use safety is greatly improved.