20results about How to "Save memory" patented technology

This application relates to a method, apparatus, device, and medium for adaptive parameter pruning of a deep learning model. The method includes: acquiring a pre-set training set of a deep learning model; pre-training the deep learning model based on the training set to determine the parameters of each layer in the deep learning model; capturing the low-rank structure of the parameters of each layer in the deep learning model based on a unified tensor network model, and decomposing the parameters into multiple ring factor tensors and a core factor tensor; constructing an adaptive parameter pruning model based on a sparse Bayesian learning algorithm according to the prior distribution results of the Gaussian distributions corresponding to the ring factor tensors and the core factor tensor, and the prior distribution results of the Gamma distribution corresponding to the noise level; and pruning irrelevant components of the parameters based on the adaptive parameter pruning model that has reached convergence to determine the corresponding compressed parameters. This application enables the deep learning model to maintain a low number of parameters while still maintaining high accuracy.

The application discloses a laser vision real-time weld seam tracking method based on a lightweight target detection network, and the method comprises the following steps: before welding, an industrial camera collects an initial weld seam image and transmits the initial weld seam image to an embedded industrial computer, pixel coordinate values of initial weld seam feature points are acquired, and a starting position of the weld seam is obtained; a lightweight target neural network is constructed; during welding, the industrial camera continuously collects weld seam images and sends the weld seam images to the embedded industrial computer, pixel coordinate values of weld seam feature points in the weld seam images are extracted by using the lightweight target neural network; a welding position to be reached by a robot is obtained through the pixel coordinate values of the weld seam feature points; a difference between the welding position of the robot and a current position of the robot is calculated, and a deviation value is sent to a control cabinet, so that the welding gun of the robot is controlled to automatically track and move along the weld seam. The weld seam image is detected by using the lightweight target neural network, and the problem that the real-time rate of the existing weld seam tracking method is not high on a resource-limited device is solved.

This invention provides a method for UAV obstacle avoidance trajectory planning based on digital elevation maps, including digital elevation map conversion, initial trajectory loading, trajectory pre-planning, and trajectory re-planning. Using this invention, UAV obstacle avoidance trajectory planning can fully consider terrain variations during low-altitude flight missions to achieve ground-hugging flight. Simultaneously, this method first performs offline trajectory pre-planning and uploads the planned trajectory to the flight controller to provide flight guidance, effectively solving the problem of low real-time performance. When obstacles or threat areas appear on the pre-planned trajectory, the method provided by this invention can find the vertices of obstacles with complex shapes, handle obstacle overlap issues, and thus quickly plan a flight path.

The invention discloses a key value cache processing method and device, equipment and a storage medium, the method is applied to edge equipment in communication connection with remote storage, when a large language model service is used for reasoning, the first capacity of a local key value cache is detected, and when a preset capacity threshold value is reached, the first capacity of the local key value cache is determined. And creating and maintaining a sliding window covering a plurality of key value cache entries in a local memory of the edge device. New tokens are continuously generated along with large language model service reasoning, and local cache content is updated through the sliding window: on one hand, the oldest key value cache entry in the sliding window is removed to release memory space; on the other hand, a first cache entry corresponding to the newly generated token is synchronously obtained from the remote storage and stored in the local memory. According to the invention, communication traffic and delay with remote storage are effectively reduced, and realization of low-memory, low-delay and high-performance LLM reasoning service is facilitated. The technical scheme of the invention can be widely applied to the technical field of artificial intelligence.

The application discloses a modbus address optimization method based on a small stack chip, and specifically comprises the following steps: receiving a modbus reading command, determining a starting address and a reading length; traversing a structure body, searching for a module parameter according to a reading data address; saving the reading data to a reply frame buffer, adding one to the reading data address and adding one to the reading data length, returning to the previous step until the reading data is completed, and entering the next step; generating a reply data frame and sending the reply data frame. The application changes the original several global array methods into a plurality of global arrays defined according to parameter modules, and further, when reading a reserved address space, only 0 can be read, and when writing the reserved address space, any data can be written but no actual physical address is written, that is, about 40% of the original stack space is saved, and the saved stack space can be used in the development of logical codes while the modbus protocol can be normally used.

The application discloses a simulation method and device of a multi-physical field coupling resonator, a storage medium and a terminal, wherein the method comprises the following steps: acquiring a resonator model to be solved; identifying a plurality of physical fields in the resonator model to be solved; performing cubic grid division on the resonator model to be solved, and dividing the resonator model to be solved after the grid division into a plurality of grid areas; and acquiring system matrices of all the grid areas by using a finite element method; wherein a degree of freedom vector in the system matrices comprises physical field components of the plurality of physical fields in the resonator model to be solved; the system matrices of all the grid areas are sequentially spliced based on a coupling mode of the same physical field components in the matrices, so as to acquire the system matrices of the resonator model to be solved; and the system matrices of the resonator model to be solved are solved, so as to acquire a target solving physical quantity. The application realizes the simulation combination of the plurality of physical fields in the resonator, reduces the solving time, and also reduces the memory required for calculating the matrices.

The application provides a log processing method of a database, a storage medium and equipment. The log processing method comprises: obtaining log configuration information, and determining whether log configuration has been completed according to the log configuration information; if not, assigning a preset level to a log entry to be written; generating an entry number according to the log level; and writing the log into a pre-set shared memory according to the entry number. Therefore, even if the log configuration has not been completely parsed, the log can be written into the pre-set shared memory by assigning a preset level to the log entry, thereby avoiding the problem of log loss before the log configuration is completely parsed. Moreover, because the log exists in the shared memory, even if the program is interrupted, the log in the shared memory will not be lost, and the log that has been written in advance can still be obtained.

This invention proposes a lightweight remote sensing image change detection method based on a binary neural network and large-kernel strip convolution. The implementation steps are as follows: obtaining training and testing sample sets; constructing a lightweight remote sensing image change detection network model based on a binary neural network and large-kernel strip convolution and iteratively training it; and obtaining the lightweight remote sensing image change detection results. In the process of iteratively training the lightweight remote sensing image change detection network model and obtaining the change detection results, this invention uses ResNet18 networks in two branches to perform deep feature extraction on the feature maps of both time periods. This allows the weights and activations to be quantized to +1 and -1, achieving a 32-fold memory saving and a 58-fold CPU acceleration, avoiding the drawback of existing technologies with a large number of parameters, and effectively improving detection efficiency. Simultaneously, the large-kernel strip convolution network compensates for the accuracy deficiency caused by binary quantization.

This invention relates to a method for estimating the radiant energy emitted from the top of the atmosphere based on high-orbit satellite sensors. The method calculates the radiance of the radiant energy emitted from the top of the atmosphere using the following formula: where F is the radiance of the radiant energy emitted from the top of the atmosphere, and E... m Ω represents the energy emitted from the top of the atmosphere received by high-orbit satellite sensors. e To obtain the solid angle of observation of the top of the atmosphere by a high-orbit satellite sensor, the solid angle can be solved using a complete elliptic integral of the third kind. A rectangular coordinate system based on the solid angle is then constructed to calculate the boundary curve of the observation field of view. The corresponding geocentric and geofixed coordinates are obtained through a coordinate transformation matrix. This invention can accurately estimate the emitted energy from the top of the atmosphere, providing support for related research.

Embodiments of the present disclosure disclose an adaptive memory conversion method and device for a distributed file system. A specific implementation of the method comprises: dividing a metadata set in a target directory to obtain a metadata subset group; setting a corresponding memory data structure form for each metadata subset; determining a target metadata subset; executing a corresponding access processing flow on the target metadata subset to obtain and output a final access result; obtaining access feature sets and system resource states of each metadata subset in the metadata subset group in real time to obtain a directory index set; determining memory data structure form conversion instructions corresponding to metadata subsets to be converted in the metadata subset group to obtain a conversion instruction set and a to-be-converted metadata subset group; and performing data reconstruction on the to-be-converted metadata subset group to obtain a converted metadata subset group. The implementation can improve memory utilization, reduce redundant memory occupation, and reduce unnecessary data access overhead.

The invention discloses an action recognition method for a wearable finger ring, the method is applied to a system comprising at least two finger rings, and the method comprises the following steps: synchronously collecting posture, movement and pressure data of each finger through a sensor of each finger ring; the current operation mode is dynamically determined based on the action of the thumb ring, and different operation modes correspond to different instruction mapping rules; performing corresponding processing and function mapping on data of a non-thumb ring according to the current operation mode so as to identify and generate a corresponding operation instruction; and on the basis of the cooperative motion data of the at least two rings, through space coordinate transformation and distance calculation, recognizing a continuous gesture action. The method has the beneficial effects that high-dimensional and natural interaction is realized; the gesture ambiguity and the false touch rate are fundamentally reduced; high-precision recognition of complex continuous gestures is supported; an extensible identification framework is provided; and the immersion and efficiency of interaction are improved.

The present application relates to a kind of non-cooperative communication signal modulation identification method based on multiscale feature fusion extraction, belong to cognitive radio communication technical field, including generating the time domain signal of communication signal under different modulation style and being preprocessed to obtain multiple baseband IQ signal as sample data, with corresponding modulation style as sample label composition communication signal dataset;Initialization multiscale feature fusion extraction network model, utilize the modulation signal dataset to train the multiscale feature fusion extraction network model, obtain trained multiscale feature fusion extraction network model;Real-time acquisition non-cooperative communication signal time domain signal and pre-processing, input the trained multiscale feature fusion extraction network model, obtain the identification result of corresponding time domain signal modulation style.Accurate and efficient identification of non-cooperative communication signal modulation style is realized.

This invention discloses a distributed decoding method and system for web-based video surveillance, belonging to the field of network video playback technology. The method includes: the browser actively initiating a WebSocket connection via a local loopback address port (127.0.0.1:19022) to detect whether the user terminal has installed a distributed decoding program; the distributed decoding program calls the local FFmpeg library, utilizes the terminal CPU to perform real-time decoding of the video stream, and uses OpenCV to convert the decoded video keyframes frame by frame into an image format natively supported by the browser; the distributed decoding program encapsulates the binary data and address identifier of each frame into a WebSocket message and actively pushes it to the browser, which then refreshes and displays the image in real-time in a designated area of ​​the page based on the address identifier; this invention achieves zero-plugin, low-latency, cross-platform, and secure web-based video decoding, significantly reducing server load and network latency, and improving user experience and system compatibility.

The application discloses a kind of firmware differentiation hot upgrade method, comprising the following steps: new and old firmware file is imported into host computer;Extract the difference information of new and old firmware, construct upgrade information file;According to the function module of firmware, new and old firmware is segmented into new firmware segmented file and old firmware segmented file;In host computer, the difference information of new and old firmware segmented file is input and run upgrade patch construction algorithm, and the optimal upgrade patch file is obtained;Host computer transmits upgrade information file and upgrade file to lower computer;Lower computer restores new firmware by old firmware and upgrade file, and performs hot upgrade on firmware file according to upgrade information file in function module unit.The application segments firmware according to function module, and selects difference file in segmented file, so that the device can upgrade incremental patch during operation, and the difference algorithm is improved to reduce the size of difference file, with higher upgrade efficiency.

The application discloses a kind of life prediction method of power semiconductor module, terminal equipment and storage medium, determine the service life influencing factor of power semiconductor device under the operation of power cycle condition;From the influence factor, construct training data set, and the data in training data set is normalized, the training data set after normalization is used to train deep convolutional neural network, and the life prediction model is obtained.The present application considers the individual and mutual influence of characteristic parameters on service life, and produces more accurate prediction results.

This invention discloses a method and system for processing abnormal data in virtual games, relating to the field of big data processing technology. The method involves receiving a sequence of interactive data frames uploaded by a terminal device and parsing it to obtain a set of input device sampling points and displacement coordinate vectors. Feature extraction is performed on the input device sampling point set to obtain a high-dimensional feature vector. This high-dimensional feature vector is mapped to a preset behavioral feature space, and the multidimensional statistical distance of the high-dimensional feature vector in the behavioral feature space is calculated to generate a first anomaly confidence coefficient. Spatial mapping data corresponding to the virtual scene is invoked. Discretized ray stepping detection is performed in the spatial mapping data based on the displacement coordinate vector to generate a second anomaly confidence coefficient. The first and second anomaly confidence coefficients are fused using a dynamic weighted algorithm to obtain a comprehensive anomaly score. Abnormal data is determined based on the comprehensive anomaly score, and a state rollback operation is performed on the virtual object.

The embodiment of the invention discloses a vector data learning type indexing method and device, equipment and a storage medium, and relates to the technical field of artificial intelligence and databases, the method comprises construction operation, retrieval operation and updating operation of an index structure, and the index structure is composed of a mapping relation table from cluster numbers to vector ID sets and a classification model. According to the construction operation, reference vectors in a vector data management system are divided, a mapping relation table from cluster numbers to vector ID sets is formed, and meanwhile, a classification model is trained to be used for predicting L-nearest neighbor clusters of query vectors. In the retrieval operation, candidate vectors are collected through one-time prediction of the classification model, and k designated candidate vectors with the closest distance are returned by calculating the preset distance between the query vector and the candidate vectors. The updating operation updates the index structure in real time according to addition, deletion, modification and change of the reference vector. The invention provides a construction method and a dynamic updating mechanism of a vector data learning type index structure, and more efficient approximate nearest neighbor search is realized.

The present application relates to the technical field of target detection, and discloses a road traffic sign detection method, device, equipment and storage medium, the method comprising: acquiring a road traffic sign picture, and obtaining a traffic sign data set; preprocessing the traffic sign data set; labeling the preprocessed traffic sign data set, and dividing the labeled traffic sign data set to obtain a training set and a test set; constructing an initial road traffic sign detection model based on a YOLOv5 algorithm framework, replacing the main network with a lightweight network and adding a small target detection layer; inputting the training set into the model for training to obtain a trained road traffic sign detection model; inputting the test set into the trained road traffic sign detection model for detection and identification to obtain a road traffic sign detection result, which can improve the structure of the initial prediction detection model, reduce the memory, and improve the detection accuracy of small road traffic signs.

The embodiment of the invention provides an image processing method and device, electronic equipment and a storage medium, and relates to the technical field of computer processing. The method comprises the following steps: in response to an image processing request, obtaining a to-be-processed image, the to-be-processed image comprising a target object; determining a target distance value from a plurality of image pixel points in the to-be-processed image to an object boundary of the target object, and determining a target pixel point corresponding to the target object in the plurality of image pixel points according to the target distance value; and determining a shadow pixel value of the target object acting on the target pixel point according to a preset color, determining a target image according to the shadow pixel value corresponding to the target pixel point, and displaying the target image. According to the technical scheme provided by the embodiment of the invention, the shadow effect can be quickly and effectively added to the target object in the to-be-processed image under the conditions of relatively small memory occupation and relatively low operation complexity, and the image display effect is enriched.

The application discloses a kind of high-resolution fan blade image small defect detection method and related equipment, wherein method includes: obtaining high-resolution blade image, and the image is preprocessed, and initial dataset is constructed;Defect classification network is constructed in combination with field prior knowledge, and defect classification network is trained;Defect classification network after training is used, whether the image region exists defect is judged and defect category probability is calculated, and defect saliency picture is generated;Defect detection network is constructed, and initial dataset is used to train defect detection network;Defect saliency picture and the down-sampling image of original drawing are input into defect detection network, respectively utilize detection head to obtain the detection frame coordinates of defect corresponding category, then according to original drawing coordinate, detection result fusion is carried out.The present application is based on field prior knowledge, and saliency image can be screened in high-resolution blade image the area with highest defect probability in priority, facilitate the initial positioning of small defect, improve detection efficiency and precision.