20results about How to "Suitable for deployment" patented technology

This application provides one or more embodiments of a model compression method, apparatus, device, and storage medium. The method includes: inputting descriptive information corresponding to at least one private data sample into a first large language model; using the descriptive information as a generation condition to generate at least one generated data sample corresponding to the descriptive information; selecting generated data samples similar to the at least one private data sample from the at least one generated data sample; performing model compression on a pre-trained second large language model based on the selected generated data samples to obtain a compressed model corresponding to the second large language model; wherein the second large language model is pre-trained based on public data samples; and fine-tuning the compressed model based on the private data samples to complete the compression processing for the second large language model.

The utility model discloses an urban road lighting system which comprises a plurality of lamp subsystems, namely an Internet of Things lamp subsystem, an energy storage lamp subsystem and a photovoltaic lamp subsystem. Each communication unit is in signal connection with the corresponding lamp subsystem, the service management cloud platform is in signal connection with the communication units, and the service management cloud platform is configured to issue instruction signals to the corresponding lamp subsystems through the communication units. And receiving the internal information and / or environment information of the plurality of lamp subsystems through the communication unit so as to realize unified management of various and large-scale lamp subsystems. Through the synergistic effect of the communication unit and the business management cloud platform, unified management of various types of lamps is realized, and the business management cloud platform can receive internal information and environment information of each lamp subsystem through the communication unit, and issues instruction signals to each lamp subsystem, so that the lamp subsystems are controlled to be in a unified manner. Therefore, centralized control and monitoring of large-scale lamps are realized.

This invention discloses an adjustable activation neuron circuit and a multilayer inference acceleration device based on phase-change memory (PCM), relating to the field of micro-nano electronics technology. It includes: a neuron input terminal for receiving data to be activated; a neuron output terminal for outputting data after nonlinear activation processing; an activation parameter adjustment circuit connected to an activation function circuit, which includes: a first transmission gate, a second transmission gate, a third transmission gate, a fourth transmission gate, a PCM, and an operational amplifier. The inverting input terminal of the operational amplifier receives the data signal to be activated, the non-inverting input terminal is grounded, and the output terminal serves as the neuron output terminal. The activation parameter adjustment circuit is used to adjust the resistance state of the PCM through pulse signals, changing the nonlinear relationship between the output voltage and the input current. Based on the physical characteristics of PCM, this invention develops a neuron circuit with continuously adjustable activation functions, improving the data processing capability of neural networks.

The invention provides a floating type wave energy capturing magnetic induction power generation device and method, and relates to the technical field of ocean renewable energy power generation. The stress floating plate is arranged on the sea surface; the generator support is fixedly installed on the stress floating plate, and the coil assembly is installed on the generator support. The sliding rail rod is fixedly arranged on the generator support, the sliding base is arranged on the sliding rail rod in a sliding mode, and the magnet assembly is rotationally hinged to the sliding base through a pin shaft. And the sliding block rod is fixedly connected with the sliding seat and forms a motion transmission matching relationship with the generator bracket. When waves act on the stress floating plate, the motion of the waves is transmitted through the generator support and the sliding block rod, the magnet assembly generates stable reciprocating relative motion relative to the coil assembly, magnetic flux in the coil is periodically changed, and therefore electric energy is output according to the electromagnetic induction principle. The device is simple in structure and short in energy conversion path, reduces complex mechanical transmission links, and has the advantages of being high in sea condition adaptability, high in operation reliability, low in maintenance cost and the like.

The invention discloses an infant paper diaper defect detection method. The method comprises the following steps: firstly, collecting baby diaper pictures and constructing a training data set; thirdly, constructing an infant paper diaper surface defect detection model, and training the infant paper diaper surface defect detection model by utilizing the training data set to obtain a trained infant paper diaper surface defect detection model; the infant paper diaper surface defect detection model is obtained by improving a yo11n model. And finally, inputting a to-be-detected baby paper diaper picture into the trained baby paper diaper surface defect detection model, and outputting a defect detection result by the model. According to the infant paper diaper surface defect detection model provided by the invention, on the premise that the model parameter quantity and the calculation quantity are remarkably reduced, the average precision can still be kept at the same level as that of an existing reference model, and good balance of precision and efficiency is realized.

The invention discloses a small sample plant disease identification method and system based on CLIP model lightweight adaptation, and the method comprises the steps: collecting plant disease and insect pest images, carrying out the disease category labeling, and constructing a plant disease and insect pest image data set; generating multi-source text descriptions in one-to-one correspondence with the plant disease and insect pest images in the plant disease and insect pest image data set; a small sample plant disease recognition model is constructed based on the CLIP model, and a lightweight adapter module is embedded into the output end of an image encoder of the CLIP model to output adaptive image features; taking the plant disease and insect pest image in the training set and the matched multi-source text description as input, and extracting image features and text features in the plant disease and insect pest image; training the small sample plant disease recognition model by adopting a mixed loss function; verifying the model performance through the verification set to determine an optimal model parameter; and inputting a to-be-detected plant image into the trained small sample plant disease identification model to obtain a category identification result of the plant disease.

The invention discloses a visual language model zero sample distribution external detection method, a medium and computer equipment. The method mainly comprises the steps of optimal transmission OT probability alignment, adaptive pseudo-label generation, lightweight detector training and self-enhancement feedback circulation. According to the detection method based on self-enhancement optimal transmission, on the premise of not depending on any label data and not finely adjusting a model backbone network, the posterior probability is aligned through optimal transmission, and the alignment process is corrected through feedback circulation of a lightweight detector; the distributed out-of-distribution detection performance of the vision-language model on the target task can be remarkably improved.

The application discloses a multi-agent cooperative navigation optimization method, and the optimization method takes a multi-agent deep deterministic policy gradient algorithm as a basic framework; the following mechanism is introduced in the training stage: in the multi-agent cooperative navigation training stage, a joint priority experience extraction mechanism is introduced, joint experience samples are extracted and spliced according to the time difference error priority through cross-agent synchronization indexing; an information discarding and correction compensation mechanism is introduced at the input end of the centralized Critic network, information from other agents is divided into independent information blocks, the information blocks are randomly discarded with a preset probability, and a correction factor is applied to the retained information blocks to maintain the mathematical expectation of the input data unbiased, and the input of the reduced Critic network is obtained. The application provides a multi-agent cooperative navigation optimization method which can improve the key experience utilization rate, reduce the dependence of the model on high-dimensional global input, and enhance the robustness in a communication limited environment.

The application discloses an infrared small target detection method based on space-time context perception and compact geometric representation, comprising the following steps: acquiring three adjacent images in a sequence of infrared images to be detected; inputting a pre-trained small target detection model to output a center heat map, a center offset and an effective radius prediction result, and completing positioning and scale estimation of the infrared small target; the small target detection model is used to extract multi-time space features through a backbone network and a feature pyramid sharing weights, obtain space-time representation features suitable for infrared small target detection by introducing space-time context perception information and constructing a time domain difference enhancement and global gating adjustment mechanism, and realize direct prediction of the center position and the effective radius of the small target in combination with a decoupled geometric parameter prediction network. The infrared small target detection task is modeled as target center position and effective radius prediction, so that effective detection is realized while reducing model calculation complexity and labeling cost.

The invention discloses a video compression method based on motion estimation and affine transformation substitution. The method comprises the following steps of: 1, performing feature analysis and initial coding on an image to realize initial compression of an original video frame, and establishing spatial geometric association between adjacent frames; 2, mapping high-resolution homography to a low-resolution coding space by using a down-sampling matrix based on spatial geometric correlation to obtain a down-sampling homography matrix; step 3, according to the mask region of the moment coding data obtained by calculation in the step 2, separating to obtain a code of a cutting background, namely a code of an increment background, and realizing differential compression; and 4, according to the codes of the incremental background obtained in the step 3, the decoding end combines the stored codes at the T-1 moment to splice and fuse reconstructed coding data at the T moment, and then video frames are decoded. According to the invention, the adaptive compression coding capability for different regions in the video compression process is effectively improved.

This invention discloses an online identification method for key parameters of a magnetically driven rotor for twin control. The method includes: constructing a parameterized digital twin model of the magnetically driven rotor, defining the time-varying key internal parameters in the model as parameter vectors to be identified; defining an objective function within a set time window, transforming the parameter identification problem into a nonlinear optimization problem that minimizes the objective function; adaptively identifying system parameters based on the LM optimization method to obtain parameter update vectors; passing the parameter update vectors to the system matrix of the parameterized digital twin model for state prediction at the next time step, and outputting the parameter update vectors to complete the online identification of key parameters. This invention effectively improves the accuracy, real-time performance, and embedded platform adaptability of parameter identification, providing a reliable model foundation for high-precision control, performance degradation monitoring, and predictive maintenance of magnetically driven rotors.

The application discloses a residual attribute prompt driven detection and incremental learning method for unknown scattering objects, aiming to solve the problems of missing detection, false detection and lack of self-learning ability of existing inspection on training set outside scattering objects. When the inspection vehicle or unmanned aerial vehicle enters the working area, start the visible light camera to collect real-time video stream; use the context residual learning module to detect the abnormal area of the video image, generate the residual energy graph to locate the potential unknown scattering object; based on the shape, texture and reflectivity statistics of the candidate frame area, construct the attribute vector and automatically generate the descriptive semantic prompt word, the prompt and the preset word library are fused through the gate weight to drive the open visual-linguistic segmentation, output the boundary information and semantic label of the scattering object; through the adaptive incremental learning module, the new detection sample is updated and knowledge playback with few samples, realizing the dynamic expansion and continuous evolution of the knowledge base. The application realizes the discovery of unknown targets through residual detection, completes semantic recognition and boundary extraction through open segmentation, and continuously expands the detection category through adaptive incremental learning, constructs a 'discovery-recognition-learning' closed loop mechanism, thereby significantly improving the processing ability of road inspection on unknown scattering objects in complex environment.

The invention discloses a lithium iron phosphate battery remaining capacity calibration method and device, and the method mainly comprises the steps: obtaining the spectrum data of a battery at different SOC and temperatures through an offline electrochemical impedance spectroscopy test, carrying out the inversion to obtain a relaxation time distribution DRT function, and building a detailed SOC-T-DRT feature database; during online estimation, the system collects a relaxation voltage time sequence after the vehicle stands; meanwhile, the current temperature, the SOC estimation value and a priori DRT feature vector obtained through interpolation in an offline database are input into a BiLSTM bidirectional long short-term memory network together, and the network inverts a high-precision SOC estimation value from the input through a mapping relation obtained through training learning. According to the method and the device, the relaxation time distribution characteristics and the deep learning model are introduced, so that the problem of insufficient SOC estimation precision in a lithium iron phosphate platform area in a traditional method is solved, and high-precision and robust SOC estimation is realized.

The invention provides a continuous blood glucose monitoring signal denoising method based on a two-factor layered adaptive wavelet threshold. The method comprises the following steps: firstly, performing discrete wavelet decomposition on an original noisy blood glucose signal; then, on the basis of noise characteristics of each scale, independently designing a threshold value for each layer of decomposition by adopting a hierarchical adaptive strategy; further, a novel two-factor nonlinear threshold function is introduced to perform contraction processing on the wavelet coefficient, a smooth transition factor controls function continuity near a threshold boundary to suppress a reconstruction ringing effect, and an amplitude retention factor adjusts the attenuation degree of a large-scale wavelet coefficient to retain key physiological features such as rapid change of blood glucose; and finally, reconstructing the de-noised signal through wavelet inverse transformation. According to the method, measurement noise and baseline drift are effectively suppressed, meanwhile, the fidelity capability of dynamic characteristics such as a blood glucose trend turning point is remarkably improved, and the inherent contradiction between excessive smoothness and noise residue in a traditional wavelet threshold method is relieved.

This application provides a method and apparatus for target motion state recognition based on optical flow field. The method acquires a monitoring image of a target object, inputs the monitoring image into a target motion detection model to obtain the target object's detection information, extracts sparse feature points from the monitoring image, estimates the optical flow information of the sparse feature points, removes the feature points corresponding to the target object from the sparse feature points to obtain background region feature points, constructs a background optical flow field estimation model based on the optical flow information of the background region feature points, inputs the center position coordinates of the target object into the background optical flow field estimation model to obtain a background displacement vector, determines the target object's displacement vector based on the center position coordinates of the target object, and determines the target object's motion state in the monitoring image based on the difference between the background displacement vector and the target object's displacement vector. This application improves the accuracy of target object motion state recognition and meets the real-time recognition and monitoring needs of UAV inspection scenarios.

The invention discloses a power distribution network equipment fault classification method based on adaptive period selection, and the method comprises the steps: carrying out the preprocessing of an original signal of a power distribution network, and generating standardized time series data; dynamically extracting periodic characteristics of the signals from the standardized time sequence data by using FFT and EMA; integrating the periodic features and the standardized time sequence data based on a lightweight periodic module to generate an optimized feature vector; processing the optimized feature vector by adopting bidirectional position coding to obtain an optimized feature vector after time dependence enhancement; and carrying out classification processing on the optimized feature vectors through a classifier, and finally outputting a fault category with the highest probability. According to the invention, the efficiency, accuracy and robustness of intelligent power grid power distribution equipment monitoring can be greatly improved, the problems of fault classification instantaneity and reliability in complex power grid environments (such as cities, villages and industrial parks) are solved, the maintenance cost of a power grid is reduced, and the power supply stability is improved.

The application discloses a weak and small target detection method based on a space-time adaptive resonance Mamba network and belongs to the technical field of computer vision and image processing. The method comprises the following steps: acquiring a to-be-detected image sequence, determining an input image sequence from the to-be-detected image sequence by using a sliding time window and constructing an input image sequence tensor; inputting the input image sequence tensor into a visual state space encoder to extract spatial features of each frame of input image; inputting the extracted spatial features into a space-time adaptive resonance fusion module, wherein the module performs implicit motion alignment, motion resonance perception, bidirectional time scanning and local contrast adaptive spatial scanning, finally obtains space-time enhanced features and inputs the space-time enhanced features into a decoder; and finally reconstructing the features by the decoder and outputting a target detection result. The application effectively solves the problems of easy loss of weak and small targets and high false alarm rate in a complex dynamic scene, and significantly improves the detection precision and robustness of weak and small targets.

This invention belongs to the field of road defect detection technology, specifically relating to a method for detecting road defects in UAV remote sensing images based on SCDLW-YOLO. The method includes the following steps: acquiring UAV road remote sensing images, performing data augmentation processing, constructing an expanded road defect dataset and dividing it proportionally into training, validation, and test sets; constructing an SCDLW-YOLO road defect detection model based on the YOLOv8s detection framework; inputting the training set into the SCDLW-YOLO road defect detection model for training; using the validation set to supervise and iteratively optimize the model training process; using the test set to further test and confirm the performance of the converged model; finally, obtaining the optimal SCDLW-YOLO road defect detection model and deploying it on a UAV or embedded device. This invention can effectively improve the detection accuracy and real-time performance of road defects such as cracks, potholes, and repair areas in complex scenarios, while also maintaining a lightweight model, making it suitable for road inspection and intelligent defect recognition using UAVs and edge devices.

The invention discloses an infrared small target detection device and method based on a Mama state space model, and relates to the field of infrared image processing. Comprising a data preprocessing module used for preprocessing an original infrared image to obtain an infrared image set, and dividing the infrared image set into a training set, a verification set and a test set; the model construction module is used for constructing an infrared small target detection model based on a Mama state space model; the model training module is used for inputting the infrared images of the training set and the verification set into the infrared small target detection model for target detection training; and the result output module is used for inputting the infrared image of the test set into the trained infrared small target detection model to obtain a detection result of the infrared small target. The method and the device are used for realizing efficient and accurate infrared small target detection under limited computing resources.

The present application belongs to the cross field of biomedical signal processing and artificial intelligence, and particularly relates to a sleep staging system and method based on a double-flow parallel neural network. The system comprises a brain electrical device, a data acquisition and preprocessing module, a HypnoMamba-Dual neural network model and a downstream application module connected in sequence; the brain electrical device is a single-channel electroencephalogram signal acquisition device for acquiring original EEG signals; the data acquisition and preprocessing module is used for filtering and windowing the original EEG signals for standardized operation; the HypnoMamba-Dual neural network model is used for analyzing and calculating the preprocessed EEG signal sequence and outputting a sleep staging sequence. The double-flow architecture of the present application, especially the introduction of the "local reserved flow", provides a mechanism for the model to resist the "context smoothing effect". When the context flow tends to ignore the short N1 period, the strong instantaneous features provided by the local reserved flow can effectively correct the decision, thereby significantly improving the F1 score of the N1 period.