284results about How to "Save computing power" patented technology

The invention discloses a preference alignment model training method based on limited preference identifier sorting, and the method comprises the steps: constructing a candidate preference data set based on an SFT data set; based on the candidate preference data set, constructing a limited preference identifier sorting data set; and training the SFT model by adopting the limited preference identifier sorting data set to obtain a preference alignment model. According to the method, the problems of instability, over-fitting and the like of a large language model in training can be avoided, a large amount of preference data does not need to be manually marked, and the computing power required for training the preference alignment model is greatly reduced.

According to the abstract syntax tree optimization method based on the ST language, the ST programming language is optimized through the compiler front-end technology, the code structure is optimized from the source, the compiling efficiency is improved, and the execution efficiency of the finally generated instruction is improved. An abstract syntax tree optimization method based on an ST language comprises the steps that 1, an antlr tool is used for conducting lexical analysis and grammatical analysis of the ST language, and a complete abstract syntax tree is created; (2) traversing the abstract syntax tree to carry out constant merging, calculating constant tree nodes in advance, and merging the constant tree nodes into a tree node; (3) traversing the abstract syntax tree to eliminate dead codes, and deleting useless tree nodes; and (4) traversing the abstract syntax tree to calculate characteristic values of the tree nodes, and eliminating sub-expressions according to the similarity of the tree nodes, thereby reducing redundant calculation and improving execution efficiency.

The invention discloses an intelligent operation and maintenance management method and system based on an artificial intelligence data center, and relates to the technical field of data processing, and the operation and maintenance management system comprises a data collection module, a data classification storage module, a data prediction module, a data processing module and a data evaluation module. The data acquisition module is used for acquiring different hardware equipment data and software data in the operation and maintenance management system in real time. The data classification storage module is used for classifying and storing the data in the operation and maintenance management system, and then the importance of different data is determined by utilizing grades; and then the data prediction module analyzes the classified data, so that the operation and maintenance management system can predict different classification data conveniently, after fault data occurs in the operation and maintenance management system, judgment is performed between the fault data and the classification, and the position of the fault data in the operation and maintenance management system is effectively and quickly positioned.

The invention discloses a multi-tenant permission configuration method, system and device based on an SaaS platform and a medium, and relates to the technical field of data processing, and the method comprises the steps that a multi-tenant information layer is acquired based on the SaaS platform, and a first permission set is acquired, and the first permission set comprises one or more different permission types; defining a weight index, and generating a first identifier and a first weight value; acquiring a newly added tenant based on the SaaS platform, matching role types and completing permission configuration; setting a verification time period, obtaining a second permission set of the newly added tenant in real time every one verification time period, and generating a second weight value corresponding to the first tenant; whether the second weight value is the same as the first weight value or not is judged, if not, tenant permission configuration is withdrawn, and if yes, a second identifier corresponding to the first tenant is generated; and judging whether the second identifier is the same as the first identifier or not, and if not, recovering the permission configuration of the tenant. The method has the advantages of accurate configuration, dynamic verification and high safety.

The invention provides a target aircraft carbon emission determination system, and relates to the technical field of aircraft carbon emission determination. The method comprises the steps that the flight duration T of a target aircraft on a target flight path, the take-off weight of the target aircraft and the flight height and the flight speed of the target aircraft in the level flight stage on the target flight path are acquired, and the flight duration T of the target aircraft on the target flight path is calculated according to the model of the target aircraft; determining a carbon emission calculation weight corresponding to the target aircraft from a preset carbon emission calculation weight mapping table so as to determine the carbon emissions of the climbing stage, the level flight stage and the landing stage, and further obtaining the corresponding carbon emission of the target aircraft flying on the target flight path; according to the invention, the determined carbon emission corresponding to the flight of the target aircraft on the target flight path is more accurate.

The invention discloses a coal gangue detection method based on a YOLOv8s model, and belongs to the field of coal gangue screening. In order to solve the problems of low efficiency, low precision and low speed in the existing coal gangue screening, the method comprises the following steps: acquiring a coal gangue target in real time by adopting an industrial camera, transmitting an acquisition result to a trained SKL-YOLOv8s network model, and outputting a coal gangue detection result; the network architecture of the SKL-YOLOv8s model comprises that a backbone network adopts KW dynamic convolution, an LSK module and a C2f module of YOLOv8s, a neck network adopts an SML-Neck structure, a head network adopts a detection head of YOLOv8, and a network loss algorithm adopts an MPDIOU algorithm. The device is used for detecting and screening the coal gangue.

The invention provides a camera image processing method and device and electronic equipment, and relates to the field of image processing, and the method comprises the steps: determining a distortion parameter corresponding to a camera parameter from a preset database according to the camera parameter corresponding to a current image under the condition of determining that the computing power of a current CPU module is insufficient; sending the current image and the distortion parameter to a GPU module, indicating the GPU module to create a VPI flow according to the current image and the distortion parameter, and processing the current image in the VPI flow according to the distortion parameter to obtain a processed image; and receiving the processed image until all the images are processed. According to the method, the opportunity of executing the image processing task by the GPU is dynamically determined through various detection strategies, the computing power of the CPU is saved, the stability of the system under various load conditions is ensured, and when the computing power of the CPU is insufficient, the advantages of the GPU in the aspect of image processing are fully exerted by repeatedly executing the step of processing the image on the GPU module, so that the image processing efficiency and speed are improved.

The invention belongs to the technical field of large models of deep learning, and provides a large-model low-rank adaptive layer weight matrix fine tuning method in order to reduce calculation complexity in a low-rank weight matrix multiplication process in a low-rank adaptive method and optimize calculation power and memory requirements required by a training model. The two low-rank weight matrixes are copied, spliced and then added, or the two low-rank weight matrixes are divided into the sub-matrixes and then the sub-matrixes are added one by one, so that compared with a mode of multiplying the two low-rank weight matrixes, the calculation complexity of matrix low-rank decomposition can be reduced, meanwhile, the rank of the low-rank weight matrixes can be improved, and the calculation efficiency of matrix low-rank decomposition is improved. When different downstream tasks are finely adjusted, the corresponding weight parameters are independently trained and switched, so that the training efficiency of different fine adjustment weights is improved, and the weight memory requirement is reduced.

The invention discloses a charging pile operation state evaluation method and system, and the method comprises the following steps: S1, building a cross-source data integration library based on the operation state sample data of a charging pile, S2, carrying out the layering processing of the cross-source data integration library, S3, carrying out the segmentation processing of a charging pile cross-source data set, and S4, carrying out the evaluation of the operation state of the charging pile. Calculating the fatigue degree of the cross-source data set of the charging pile according to the cross-source data integration library, and marking when the fatigue degree exceeds a fatigue degree threshold value; s5, for the marked charging pile cross-source data set, the sensitivity of the charging pile cross-source data set is calculated according to the cross-source data integration library, and when the sensitivity exceeds a sensitivity threshold value, marking is carried out; s6, the fatigue degrees and the acuity degrees of all the charging piles in the charging pile cross-source data set are calculated independently, when the specific fatigue degree and the specific acuity degree exceed threshold values, marking is carried out, and the evaluation grade is improved; the normal operation state of the charging pile can be ensured to the maximum extent, the monitoring efficiency can be improved, and the computing power consumption is reduced.

The invention discloses an unloading control method, device and system, and relates to the field of control. Based on material weight information of mineral powder loaded by a vehicle loaded with materials, the unloading time is calculated in advance, after it is determined that the vehicle is parked in a parking space of a corresponding unloading hole groove and unloading is started, the pre-calculation time is waited, and the unloading time is shortened; the vehicle residual material state is determined through image detection, and the vehicle is controlled so that the vehicle can conduct sudden entering and sudden braking of the target number of times matched with the vehicle residual material state, so that refined control over residual material discharging can be achieved, the situation that sudden entering and sudden braking control is conducted repeatedly is avoided to a certain degree, and the vehicle residual material discharging efficiency is improved. The residual material state detection is started after the pre-calculation duration, and compared with the scheme of real-time detection in the discharging process, the platform calculation power is saved; in addition, in the unloading process through the unloading device, participation of a driver and factory ground crew is not needed, and manpower resources can be saved.

The invention discloses a driving and braking redundancy cooperative control method for a full-vector power chassis automobile, relates to the technical field of automobile driving and braking automatic control, and solves the problems that a driving and braking control strategy of an existing full-vector power chassis automobile is not comprehensively considered, and the computing power requirement cannot be well reduced. According to the method, the upper layer calculates the longitudinal moment and the yawing moment required by the vehicle, and the lower layer distributes the longitudinal moment and the yawing moment sent by the upper layer controller in a rule and table look-up combined mode. An intelligent evolution method is used for training an upper-layer actuator neural network and a lower-layer actuator neural network, reinforcement learning control from the vehicle state to instruction output is achieved, the links of signal collection, calculation processing, output execution and the like of a control system are simplified to the maximum degree, particularly the hysteresis effect of the calculation processing link is simplified, and the control efficiency and precision are improved; the economical efficiency, safety and stability of the vehicle in the whole life cycle operation process are guaranteed.

The invention relates to the technical field of image enhancement, in particular to an endoscope image real-time enhancement method for emergency treatment. The method comprises the following steps: acquiring an image by using an endoscope, obtaining an initial self-adaptive convolution kernel according to a gradient value of a feature point and a difference expression between the gradient value of the feature point and an average gradient value of four neighborhoods of the feature point, and carrying out negative feedback on the size of the convolution kernel through an evaluation result so as to obtain an optimal self-adaptive convolution kernel of each pixel point on the image acquired by the endoscope; the optimal adaptive convolution kernel of the non-feature points is obtained by using differences such as distances and gray values between the non-feature points and the feature points; and sharpening each pixel point on the image according to the obtained optimal adaptive convolution kernel on the endoscope image so as to obtain an optimal sharpened image. According to the method, the endoscope image enhancement efficiency can be improved, and the calculation amount of the endoscope image enhancement process can be reduced.

The invention provides a rear wheel steering vehicle track planning method and system in a road scene based on iterative optimization, and the method comprises the steps: S1, building an optimization problem, and deducing a kinematics formula of a rear wheel steering vehicle, an obstacle avoidance constraint formula, and a calculation formula of the distance and angle error between a rear axle center point and a road center line; s2, on the basis of a road scene structure, constructing a collision-free track which fits a road center line as much as possible to serve as an initial guess in an optimization problem; s3, solving an approximate feasible track, softening the kinematics constraint into a penalty function, constructing a space-time corridor constraint, and constructing and solving an optimization problem to obtain the approximate kinematics feasible track; and S4, solving an optimal track, and gradually increasing sampling points in the optimization problem until the number of the sampling points reaches a threshold value. The method is combined with a vehicle positioning algorithm, track planning of the rear wheel steering vehicle in a road scene can be realized, scene robustness is realized, and the quality of the planned track is high.

The invention provides a video image processing method and device of an unmanned aerial vehicle pan-tilt camera and a storage medium. The method comprises the steps of collecting multiple paths of first yuv images; rTSP protocol communication between the client and the pan-tilt camera is established; in response to the acquired lens switching instruction sent by the client, acquiring a corresponding first yuv image according to the lens switching instruction; in response to the condition that the original resolution of the obtained first yuv image is different from the target resolution, zooming the first yuv image to a preset resolution to obtain a second yuv image, creating a first memory area based on the preset resolution, and carrying and storing the second yuv image to the first memory area by using ive operation; creating a second memory area based on the target resolution, creating a blank yuv image of the target resolution and storing the blank yuv image in the second memory area, and setting the yuv value of the blank yuv image to be a black pixel value by using ive operation to obtain a filled yuv image; and carrying and covering the second yuv image to the filled yuv image by using ive operation to obtain a first target yuv image.

The invention provides a road anomaly detection and vehicle lamp interaction method and system based on a vehicle-mounted laser radar, and belongs to the technical field of vehicle lamp control. The method comprises the following steps: S1, transmitting and receiving a pulse signal through a vehicle-mounted laser radar; s2, calculating the pulse signals transmitted and received by the vehicle-mounted laser radar to obtain point cloud information of the surrounding environment of the vehicle-mounted laser radar, wherein the point cloud information comprises point cloud depth information; s3, based on the point cloud depth information, a point cloud clustering algorithm is adopted to detect road abnormal conditions, and a road abnormal detection result is obtained; s4, based on a road anomaly detection result, projecting a road anomaly region to a corresponding LED vehicle lamp pixel point by adopting a multi-point perspective imaging technology; and S5, carrying out local light intensity adjustment on the LED vehicle lamp pixel points corresponding to the road abnormal area by adopting a self-adaptive high beam technology so as to enable the road abnormal area to be normally displayed under the irradiation of the vehicle lamp. Through interaction between the vehicle-mounted laser radar and the vehicle lamp, the robustness and driving safety of the vehicle lamp are improved.

The invention discloses a two-stage sparse method based on a small sample rapid sparse large-scale language model, and the method comprises the following steps: S1, mask initialization: based on a plurality of importance indexes for measuring parameters, employing a plurality of mask initialization methods to initialize a mask of a language model; s2, reordering the masks: reordering the masks based on the small sample calibration set and the straight-through estimator; s3, mask adjustment: based on the small sample calibration set and the reordered mask matrix in the step S2, taking an original model as a teacher model, taking a sparse model as a student model, realizing a block-level self-distillation process of the large language model in a time dimension before and after sparsification, and obtaining the sparse model; the method consumes less computing power, uses a smaller data set, occupies less GPU memory while realizing considerable performance, does not influence the reasoning speed of the sparse model, and has remarkable practical value and application prospect.

The invention provides a quality evaluation method and system for image correction and a rail transit detection device, and the method comprises the steps: carrying out the grading segmentation of a standard whole vehicle image with a single vehicle as a unit, and obtaining a standard key part region image; performing hierarchical segmentation on a target whole vehicle image to be evaluated by taking a single vehicle as a unit to obtain a target key component region image; performing feature extraction and feature segmentation on the standard key component region image to obtain a standard feature matrix; performing feature extraction and feature segmentation on the target key component region image to obtain a target feature matrix; and aligning the standard feature matrix with the target feature matrix, determining a feature change rate matrix, and evaluating the image correction quality based on the feature change rate matrix. According to the method, the whole vehicle image is segmented by taking a single vehicle as a unit, so that the computing power is saved, and the used memory is optimized. And the problem of quality evaluation after line scanning image correction is solved in a targeted manner.

The invention provides an industrial production safety intelligent management platform and management method. The platform comprises a real-time monitoring module, a risk assessment module, an operation permission and process management module, a data analysis and decision module, an emergency plan module, a fault maintenance module, an early warning module and a platform display module. And the risk assessment module comprises a first-level risk module, a second-level risk module and a third-level risk module of which the risk degrees are sequentially reduced. The real-time monitoring module transmits collected data to the risk assessment module, the fault maintenance module and the data analysis and decision module. And the operation permission and process management module is used for storing the monitoring data corresponding to the operation equipment of the real-time monitoring module, comparing the monitoring data with preset normal operation judgment conditions, and performing online application and approval on the operation of the real-time monitoring module. According to the method, closing operation is rapidly adopted for operation accidents, the step of analyzing emergency plans is saved, and the processing efficiency of emergently solving abnormities or accidents is improved.

The invention relates to a vehicle driving area detection method, electronic equipment and a storage medium. The method comprises the following steps: acquiring image data and point cloud data in a driving area acquired by a target vehicle; lane line detection is carried out on the driving area based on the image data, and obstacle detection is carried out on the driving area based on the point cloud data; and when it is detected that a lane line exists in the driving area, obtaining a detection result of whether an obstacle exists in a target lane corresponding to the lane line. According to the embodiment of the invention, the image data and the point cloud data are combined, lane line detection is carried out by utilizing the image data, and obstacle detection is carried out by utilizing the point cloud data, so that whether an obstacle exists in the target lane or not can be quickly and accurately judged, the computing power can be saved to a great extent, and a better recognition result can be obtained.

The invention discloses a flood situation detection method and device based on big data hyper-fusion and a medium. The method comprises the steps of generating a corresponding ponding area list and a monitoring area list; performing spatial association on the ponding area list, the camera list and the water level gauge sensor list so as to determine the camera and the water level gauge sensor which have a spatial association relationship with the ponding area; for each ponding area, calculating the water depth of the ponding area corresponding to the ponding area according to the water depth data collected by the camera and the water level gauge sensor which have the spatial association relationship with the ponding area, and predicting the predicted ponding depth of the ponding area in the future according to the water depth of the ponding area; dividing camera monitoring grading lists of different grades according to the first camera list, the second camera list and the fourth camera list; and feeding back the gathered flood situation information and the map capturing frequency to a control center, so that the control center carries out rescue scheduling according to the flood situation information, and carries out dynamic scheduling on cameras in the ponding area and the monitoring area.

The invention relates to the field of computer vision, in particular to a traffic target recognition method based on 4D millimeter wave radar and vision fusion, which comprises the following steps: clustering radar point cloud data by adopting a density-based clustering algorithm, and expanding the center coordinate of each cluster output by the clustering algorithm to obtain an expansion frame; converting the extension value into a two-dimensional visual coordinate system based on a conversion matrix between a radar coordinate and a camera coordinate to obtain an interested target frame under the visual coordinate system; fusing the interested target frame in the visual coordinate system with image data collected by a camera, and taking a fused feature map as the input of a YOLOV8 target detection model; the YOLOV8 target detection model establishes detection branches according to the number of interested target frames, and each branch detects an image of one interested frame part in the fused feature map; according to the method, all-weather sensing work can be realized, the traffic target identification speed is high, and the precision is high.

The invention discloses a gesture recognition interaction method and device, equipment, a storage medium and a product, and relates to the technical field of man-machine interaction, and the method comprises the steps: obtaining single-frame image data including a gesture action in real time, inputting the single-frame image data into a hand shape feature extraction model, obtaining single-frame hand shape features, and obtaining a hand shape feature extraction model; and determining multi-frame sequence features based on the single-frame hand shape features corresponding to the continuously acquired single-frame image data. And inputting the multi-frame sequence features into a gesture action judgment model to obtain a target gesture action, and based on the target gesture action, triggering an interaction response corresponding to the target gesture action. According to the method, the target gesture action is obtained through the gesture action judgment model, direct image-level feature extraction can be achieved, effective information is utilized to the maximum extent, the calculation power of gesture interaction in an AR scene is reduced, and the model can run on a low-calculation-power platform in real time.

The invention relates to the field of distributed archive management, in particular to a distributed budget archive collaborative management system based on big data, which is provided with an archive acquisition module, a feature analysis module, a label setting module and an archive optimization module. A label setting module triggers and determines a data representation difference representation value from a data standard form feature and sets a label for an archive file, then an archive optimization module is adopted to carry out targeted classification processing on the archive file according to the label of the archive file subsequently, and for the archive file with the difference representation label, it is considered that the difference of the archive file is high, and the classification efficiency is high. According to the method, a complete information extraction reconstruction mode is adopted, archive files are optimized, non-differential expression labels can be suggested to be processed due to the fact that the standard types of the archive files are high, and therefore when massive archive files are faced, targeted processing can be carried out, the computing power is saved, and the consistency of the archive files is improved.

The invention relates to a general appearance defect detection method based on a visual large model in the field of industrial computer vision, and the method comprises the following steps: collecting and preprocessing a picture with defects, obtaining a defect picture feature T1, obtaining a target region feature # imgabs0 # and a target embedding feature T2 through coding calculation, storing # imgabs1 # and T2 in a memory bank, and obtaining a target embedded feature T1; a defect detection end-to-end neural network model is constructed and deployed, a to-be-detected picture is preprocessed and input into the defect detection end-to-end neural network model to obtain a detection result, and through combination of defect detection and a visual large model and dependence on a complex data set pre-trained by the large model, the training cost is reduced, and the detection precision is improved. Specific defect pictures are used as neural network prompts, the generalization of the method and the flexibility of model deployment are enhanced, the method adapts to different defect detection items, and customer requirements are met.

The present invention discloses a workflow-based method for automated analysis and model optimization of hyperspectral data, and the steps are as follows: S1, processing the acquired hyperspectral image of the target sample, extracting and saving the spectral data of the characteristic area, and constructing a one-dimensional hyperspectral data set of all target samples; S2, constructing a DeepSpNet model based on a one-dimensional convolutional neural network; S3, setting the training parameters of the DeepSpNet model, and establishing a hyperparameter combination; S4, dividing the one-dimensional hyperspectral data set into a training set and a test set in proportion, training the DeepSpNet model with the training set, evaluating the accuracy, precision, and recall of the DeepSpNet model through the test set, and automatically screening the best hyperparameter combination; S5, saving the DeepSpNet model with the best hyperparameter combination for target prediction. The present invention can accurately capture the defect characteristics inside the sample and effectively improve the depth detection capability.

The present invention discloses an underwater video fish recognition method, system and storage medium, wherein the recognition method comprises: S01, constructing a fish image database; S02, acquiring underwater video images; S03, dividing the underwater video images into image frame files, then importing the files into an image recognition system for recognition processing, and outputting detection image frames with box selection marks; S04, extracting areas with box selection marks in the detection image frames, importing the areas into a fish image database for matching processing, and outputting image matching results; S05, when the matching results meet preset conditions, outputting fish information; the scheme is reliable to implement, has high processing efficiency and good user experience.

The utility model discloses a circuit of an alarm help-seeking terminal with a voice recognition function, which relates to the technical field of alarm help seeking and comprises a master control circuit module, a communication circuit module, a power supply circuit module, a server processing terminal and a voice recognition circuit module. The master control circuit module is electrically connected with the communication circuit module, the communication circuit module is connected with a server processing terminal through a wireless network, environment sound is picked up at any time, and corresponding alarm help-seeking processing such as direct call making and alarm help-seeking triggering is carried out by replacing traditional physical keys through keyword screening and screening. And whether alarm help seeking is successful or not can be judged through voice broadcast, screen display, light flickering and the like.

The invention discloses a hyperspectral data automatic analysis and model optimization method based on workflow, and the method comprises the following steps: S1, carrying out the processing of an obtained target sample hyperspectral image, extracting and storing the spectral data of a feature region, and constructing a one-dimensional hyperspectral data set of all target samples; s2, constructing a DeepSpNet model based on a one-dimensional convolutional neural network; s3, training parameters of the DeepSpNet model are set, and a hyper-parameter combination is established; s4, dividing the one-dimensional hyperspectral data set into a training set and a test set in proportion, training the DeepSpNet model by adopting the training set, evaluating the precision, precision ratio and recall ratio of the DeepSpNet model through the test set, and automatically screening an optimal hyper-parameter combination; and S5, storing the DeepSpNet model of the optimal hyper-parameter combination for target prediction. According to the invention, defect features in the sample can be accurately captured, and the depth detection capability is effectively improved.

Embodiments of the present application relate to the technical field of human-computer interaction, and disclose a method and device for determining an interactive gesture and an electronic device. The method comprises the following steps: first, target component detection is performed; then, gesture detection and recognition are performed in a local area (around a boundary box) near the target component. In addition, for N gesture detection boxes (multiple users) in a current video frame, M target gesture detection boxes are selected therefrom to perform hand component positioning and gesture recognition processing, which can reduce detection time consumption and improve real-time performance. On this basis, the interactive gesture is determined according to the gesture detection result of the current video frame and the gesture detection results of a plurality of historical video frames, that is, gesture recognition is selectively performed on the user in each frame in a plurality of continuous video frames to determine the interactive gesture, which can reduce the computing power, reduce the time consumption, improve the real-time performance, and improve the accuracy and stability of the interactive gesture. That is, in the multi-user human-computer interaction scene, the interactive gesture can be accurately and timely determined.

The invention relates to an artificial intelligence technology, and provides a to-be-financed invoice verification method, apparatus and device, and a medium, which perform table restoration on contract information based on a large language model, and perform subsequent processing after the contract information is restored into a table, so that the reasoning effect of the large language model can be improved. The target cue word is generated according to each piece of commodity information in the target invoice to be financed and the contract data table, and the target cue word is input to the large language model for reasoning, so that the large language model can be utilized to perform automatic matching on the invoice to be financed and the contract information, manual intervention is not needed, the matching efficiency is improved, and the error rate is low; according to the method, the reasoning result is post-processed to remove invalid information, so that the valid information is utilized to verify the to-be-financing invoice, and the calculation power is saved, so that the verification efficiency is improved.