23 results about "Neural network identification" patented technology

The invention discloses an intelligent regulation and control method for a power battery positive electrode binder production process, and the method comprises the steps: synchronously collecting multi-dimensional process parameters such as temperature, viscosity and the like and performance indexes such as particle size distribution, bonding strength and the like, carrying out the normalization, denoising and time sequence alignment processing, and fusing hydrodynamic simulation and historical data to construct an initial parameter coupling model; identifying a parameter influence weight through an attention mechanism neural network, generating a decoupling factor matrix to reconstruct a parameter space, and establishing a virtual control channel; executing constrained gradient descent multi-objective optimization in the channel, generating a regulation and control instruction, and reversely mapping the regulation and control instruction into an equipment executable parameter; according to the method, performance indexes and optimization targets after regulation and control are continuously compared, model parameters are updated, retraining is triggered when deviation exceeds a threshold value, regulation and control precision and process stability are guaranteed, and the real-time performance of parameter regulation and control, the collaborative optimization capability and the quality consistency of the production process are remarkably improved.

The embodiment of the invention discloses an insulation aging early warning method and device for a medium-high frequency transformer and a medium, and relates to the technical field of transformers, and the method comprises the steps: collecting a multiband light intensity distribution signal of a target resin region in the transformer according to a preset collection frequency through a preset multispectral distributed optical fiber sensing array; the insulation state of the target resin area is predicted through a pre-constructed neural network identification model and a multi-spectral distributed optical fiber sensing array, and the predicted insulation state health degree of the target resin area is determined; acquiring real-time environment parameters of the target resin area, dynamically correcting a preset early-warning discharge threshold according to the real-time environment parameters, and determining a current early-warning dynamic discharge threshold; and extracting a plurality of current light intensity parameters in the multiband light intensity distribution signal, predicting an insulation state health degree and a current early warning dynamic discharge threshold through the plurality of current light intensity parameters, and performing insulation defect early warning on the insulation state of the target resin area.

The invention provides a geological disaster prediction and early warning method and system combining monitoring data and numerical simulation, and relates to the technical field of geological disaster early warning, and the method comprises the steps: arranging a plurality of sensors in a geological disaster prone region to continuously collect data, carrying out the preprocessing, forming a standardized data matrix, and constructing a catastrophe structure map; each variable is regarded as a map node to calculate an edge weight, threshold screening is carried out, a sparse weighted adjacent matrix is generated, node features are updated through a neural network, probability distribution of a catastrophe stage is output through a full connection layer to identify a current catastrophe state, a coupling strength index is calculated, and similarity matching is carried out; coupling strength and structural similarity are comprehensively evaluated, map migration weight is obtained, then critical variable threshold and time are predicted, critical parameters and neural network recognition results are fused to construct a unified risk index model, early warning information and corresponding measures are issued according to risk grade division, and risk assessment is performed. Structured fusion and dynamic intelligent identification of multi-source data are realized.

The invention discloses an unmanned aerial vehicle route automatic planning system based on AI identification, and relates to the technical field of unmanned aerial vehicle route planning and obstacle avoidance. The unmanned aerial vehicle route automatic planning system based on neural network identification processes camera and laser radar data in real time through a lightweight convolutional neural network of an environment sensing module; accurate identification and classification of dynamic obstacles are realized, the perception ability in a dense city environment is effectively improved, and the risk of obstacle avoidance failure caused by sensor data updating delay is reduced. The fusion and tracking module adopts a space-time alignment and multi-source data fusion technology to generate uniform occupation representation and motion trail, so that the system can adapt to sudden obstacle change, the dependence on a preloaded map is reduced, and the navigation reliability in an unknown or dynamic scene is enhanced; the path planning module integrates a reinforcement learning algorithm, takes a dynamic obstacle state as input, and optimizes path generation through a multi-target reward function.

The invention discloses a damage identification method and device for an in-service steel wire rope type horizontal lifeline, and belongs to the technical field of high-altitude operation safety facilities. The method comprises the following steps: synchronously acquiring a magnetic flux leakage signal and a surface image of an in-service steel wire rope through magnetic flux leakage detection equipment and a high-definition camera which are carried on a steel wire rope inspection robot; preprocessing the acquired magnetic flux leakage signal, wherein the preprocessing comprises singular value elimination and trend term removal processing; carrying out de-noising processing on the pre-processed signal by adopting an improved wavelet threshold de-noising algorithm fused with a Sigmoid function; extracting a characteristic value for representing the damage of the steel wire rope, and performing normalization processing to form a characteristic vector; and inputting into a BP neural network identification model optimized by a genetic algorithm for identification, and outputting an assessment result of the damage type and positioning of the steel wire rope. The method can realize automatic and quantitative detection and accurate identification of internal and external damages of the steel wire rope.

This invention discloses a lightweight neural network identification method for high-speed dynamic impact signals. In the process of penetrating multi-layered hard targets, the presence of signal adhesion in the overload signals from multiple dynamic impacts makes identification exceptionally difficult. To address this problem, a lightweight network identification method based on an attention mechanism is proposed. First, time-frequency feature analysis is performed on the overload signal equivalent to that from multiple impact test benches, and continuous wavelet transform is used to extract the time-frequency features as input to the neural network. A lightweight network architecture based on an attention mechanism is designed, eliminating redundant layers and adding residual connection structures and a lightweight attention mechanism, thus ensuring recognition accuracy while significantly reducing parameters.

The application discloses a mobile terminal nasopharyngeal carcinoma identification method and belongs to the technical field of intelligent identification systems. The system can identify the nature of nasopharyngeal position lesions after training image information of confirmed nasopharyngeal carcinoma patient lesion sites, and has the characteristics of fast identification speed and high accuracy. The system comprises an image preprocessing module, an identification module, and a training identification module. The image preprocessing module is used for preprocessing images of nasopharyngeal sites. The identification module is used for storing a trained identification model. The training identification module is used for training an identification model through lesion images of known nasopharyngeal carcinoma patients. The image preprocessing module is connected with the identification module and the training identification module. The training identification module comprises a case library, an identification model unit, and a training unit. The case library is used for storing identification preprocessing data. The identification model unit is used for storing a trained identification model. The training unit is used for training a neural network identification model in the identification model unit through the identification preprocessing data in the case library.

The invention relates to the technical field of medical electronic equipment, in particular to an intelligent electrocardio lead system based on adaptive impedance matching and neural network recognition, which is characterized in that an impedance feature extraction module is used for collecting multi-lead multi-band impedance data of a human body, constructing a differential feature vector and generating parameterized impedance representation; the neural network identification module carries out hierarchical lead position judgment and outputs a position result and confidence; the self-adaptive optimization module is combined with operator feedback update parameters to execute multi-level learning so as to optimize the performance; and the interaction feedback module generates a voice prompt to guide an operator in real time, so that intelligent recognition and self-adaptive optimization of a lead position are realized, the electrocardiogram detection accuracy and the operation convenience are improved, the electrocardiogram examination accuracy is remarkably improved, and misdiagnosis and missed diagnosis caused by lead errors are effectively avoided.

The invention relates to a robust adaptive dynamic planning method for spacecraft attitude control, and belongs to the technical field of spacecraft control. According to the method, the neural network identifier is designed, online identification and reconstruction are carried out on unknown dynamic items caused by inertia parameter uncertainty in spacecraft attitude dynamics, the attitude tracking optimal control law based on adaptive dynamics is constructed on the basis, and under the condition that inertia parameter uncertainty exists, the attitude tracking optimal control law is optimized. The balance between tracking precision and control energy consumption is realized, and the applicability of the attitude control method in complex space tasks is improved.

The invention discloses a large-aperture space telescope fixed time sliding mode anti-interference control method and related equipment. The method comprises the following steps: establishing a double-inertia system dynamic model containing motor end and load end dynamic parameters and total disturbance; constructing a fixed time expansion state observer based on the model, and estimating the total disturbance of the motor end; converting the model into a generalized all-wheel-drive system model through an all-wheel-drive system theory, designing an expected closed-loop system model, and identifying the expected closed-loop system model by using a neural network to obtain a preset performance function; a motor end angle position tracking error is obtained, and a speed ring reference speed instruction is generated after preset performance function constraint transformation; defining an angular velocity tracking error based on the instruction, constructing a fixed time terminal integral sliding mode control surface and designing a sliding mode reaching law; and finally, synthesizing a motor control law and applying the motor control law to a driving motor to complete anti-interference control. The problems that a traditional preset performance function is single in form, an observer cannot converge at fixed time, and a traditional control mode is difficult to suppress vibration, high in cost and limited are solved.

The application discloses a low signal-to-noise ratio condition modulation identification method based on wavelet transform and channel attention, comprising: acquiring a received signal; inputting the received signal into a pre-trained neural network identification model to output a classification identification result and a classification accuracy rate. Based on the signal reconstruction and the identification model of the neural network, the wavelet threshold estimation module and the wavelet reconstruction module constitute a signal reconstruction path, the multi-scale feature extraction module and the prediction classification module constitute an identification path, the two paths are mutually enhanced, the advantages of the digital signal processing technology and the neural network are combined, the wavelet threshold estimation module is introduced to predict a denoising threshold through the neural network, and the network parameters of the wavelet threshold estimation module are updated according to the back propagation; the signal reconstruction path and the identification path are combined, the network interpretability is enhanced, the multi-scale modulation features are utilized, and the modulation identification accuracy rate in the low signal-to-noise ratio scene is significantly improved.

This invention discloses a neural network-based, disturbance-resistant, high-maneuverability attitude control method for variable-configuration unmanned aerial vehicles (UAVs). The method includes the following steps: Step 1: Establishing a multi-rigid-body kinematic model and a dynamic model of the variable-configuration UAV; Step 2: Constructing the input vector and sensitive region of the neural network identifier, and building an adaptive learning law for the weights of the neural network identifier; Step 3: Designing a baseline attitude controller for the variable-configuration UAV based on an offline-modeled nonlinear dynamic model; Step 4: Solving the control signal within a single control cycle and driving the servo motor to complete the actuation command; Step 5: In the next control cycle, using online measurement data from sensors to complete the adaptive learning and updating of the neural network identifier's weights. This method can comprehensively identify nonlinear unknown dynamics and external disturbances based on the current state of the variable-configuration UAV, and perform high-maneuverability attitude control based on the identification results, thereby improving the flight quality of the variable-configuration UAV.

The invention discloses a tactile data processing method and system based on an event-driven pulse neural network, and belongs to the technical field of robot perception. The method aims at solving the problems that in the prior art, topological representation of a tactile graph is insufficient, the generalization ability of a model on sparse data is weak, and training is unstable. According to the technical scheme, the method comprises the steps that firstly, according to three-dimensional space coordinates of a sensor unit, a model is built through a multi-stage graph combining a KD tree and Z-sequence coding, and an optimized tactile graph capable of representing local and non-local space relations at the same time is generated; secondly, preprocessing a real-time event driving pulse signal through a self-adaptive normalization model so as to enhance fine features; and finally, on a pulse neural network identification model which introduces Gaussian prior regularization for training, processing the preprocessed signal in combination with the optimized touch image. According to the method, the precision, the robustness and the generalization ability of the tactile recognition system on sparse data can be remarkably improved.

The application relates to the technical field of machine vision, in particular to a glass product real-time vision detection system based on a CNN neural network, which comprises an image acquisition subsystem and a neural network identification subsystem, and the specific functions are as follows: the image acquisition subsystem acquires a glass plate original image and an enhanced image based on a dynamic background, the dynamic background comprises a two-color grid image and a two-color stripe image; the neural network identification subsystem performs double-mode enhancement on the image of the glass product based on linear defect and mottle defect identification, uses an improved CNN to extract defect features, and outputs defect types and defect positions. The application enhances the features of the image of the glass product through the dynamic background, strengthens the contrast difference between the defect position and the normal position, effectively reduces the identification difficulty, meanwhile, the CNN parallel acquisition of the original image and the enhanced image avoids missing defect features, and strengthens the identification precision and speed of the neural network.

The present application relates to the technical field of medical electronic equipment, in particular to an intelligent electrocardio lead system based on adaptive impedance matching and neural network identification, which collects multi-lead multi-frequency band impedance data of human body through an impedance feature extraction module, constructs a differential feature vector and generates a parameterized impedance representation; a neural network identification module makes hierarchical lead position judgment accordingly, outputs a position result and a confidence level; an adaptive optimization module updates parameters in combination with operator feedback, performs multi-level learning to optimize performance; an interactive feedback module generates a voice prompt to guide the operator in real time, realizes intelligent recognition and adaptive optimization of lead position, improves electrocardio detection accuracy and operation convenience, significantly improves electrocardiogram examination accuracy, and effectively avoids misdiagnosis and missed diagnosis caused by lead errors.

The invention discloses a plateau yak oestrus sound recognition method based on a multi-branch fusion model. The method comprises the following steps: step 1, collecting audio data of a plateau yak; step 2, converting the audio data of the yak into a logarithm Mel spectrogram, performing time mask and frequency mask operation on the logarithm Mel spectrogram, and constructing a feature vector; step 3, performing enhancement processing on the audio data and the logarithm Mel spectrogram of the plateau yak; step 4, training a multi-branch fusion neural network identification model by using the enhanced yak sound data and the logarithm Mel spectrogram; a MobileNetV3-CBAM module, a bidirectional LSTM module and a visual Transform module of the multi-branch fusion neural network recognition model extract local time-frequency features, time sequence dynamic features and global dependency features respectively and input the local time-frequency features, the time sequence dynamic features and the global dependency features into a gating fusion unit to obtain fused features, the fused features are input into a full connection layer, and the full connection layer outputs the prediction probability of the yak sound category. And identifying the yak oestrus by using the trained multi-branch fusion neural network identification model.

The application provides a method for eliminating burrs and improving precision of a temperature detection device, and comprises the following steps: S1, acquiring current measured temperature data within a preset time, calculating a feature vector according to the temperature data, and presetting a standard deviation in a constant temperature environment; S2, performing physical constraint checking and statistical anomaly checking according to the feature vector; S3, performing neural network identification and correction; and S4, judging a confidence C and outputting a corresponding result. The method can effectively eliminate burrs generated during temperature measurement, improve measurement precision, and has high temperature measurement reliability.

The invention discloses a collaborative attack falling angle identification method based on a neural network. The method comprises the following steps: establishing an impact angle identification model; establishing a neural network identification model based on the impact angle identification model; and acquiring the impact angle of the incoming aircraft at the current moment by adopting a neural network identification model. The method disclosed by the invention is high in recognition precision and strong in robustness.

The invention relates to the technical field of medical data processing, in particular to an acute shock early recognition and typing system based on multi-parameter fusion. Firstly, various related physical sign parameter time sequence data are acquired from the data acquisition module; the anomaly analysis module analyzes the deviation between the data value at each moment and a preset range to obtain an anomaly expression coefficient, reflect the degree of deviation from a normal range, compare the difference of the anomaly expression coefficients between the moments, determine the treatment influence expression degree and evaluate the treatment measure effect; the multi-feature fusion module analyzes the time sequence and similar features of the treatment influence expression degree, fuses the time sequence and similar features with abnormal expression coefficient difference features, calculates significance indexes, and comprehensively evaluates the importance of physical sign parameters; and finally, in a weight adjustment and neural network analysis module, adjusting a preset importance weight based on the significance index, so that the system adjusts different sign parameter weights in real time, automatically learns the feature importance degree, and improves the neural network identification and accurate typing capabilities.

This invention discloses a control method and system for permanent magnet synchronous motors based on brain-based emotional learning, belonging to the field of permanent magnet synchronous motor control. This invention establishes a sensory input function and an emotional cue function based on the speed error e, generates a q-axis current setpoint through a brain-based emotional control method, and obtains a predicted value y through a neural network identifier. m (k), where the input variables of the neural network identifier are x1 = e(k) - e(k-1), x2 = e(k), and x3 = e(k) - 2*e(k-1) + e(k-2); the adjustable coefficients of the emotional cue function are optimized using the predicted values. This invention improves the control effect of the brain emotion controller on the PMSM speed regulation system, and further enhances the controller's anti-interference capability while simplifying the adjustment parameter process of brain emotion control.

The application provides a kind of intelligent pavement maintenance scheme automatic generation method and system, it is related to road maintenance technical field, including by obtaining pavement image and with convolution neural network identification disease;Adaptive extraction of geometric features is used by multi-scale grid, and dynamic evaluation vector classification is constructed;Based on evaluation result matching basic scheme;Through cross attention enhanced hierarchical decision network optimization maintenance scheme;Carry out space-time influence analysis and dynamic adjustment.The application improves the pavement maintenance efficiency and quality, reduces resource waste, reduces the influence on traffic.

This application relates to the field of data processing technology, and in particular to an artificial intelligence-based heart failure compensation identification system. The proposed scheme involves acquiring electrocardiogram (ECG) signals, bioimpedance signals, posture information, and chest displacement signals. Based on posture and respiratory information, the bioimpedance signals undergo positional correction and respiratory separation to calculate bioimpedance. Then, the temporal characteristics of the bioimpedance and ECG signals are correlated and modeled to construct an electrohydraulic coupling precursor loop reflecting the relationship between electrophysiology and fluid conduction hysteresis. Topological features such as loop area, direction, number of connections, and morphological stability are extracted. A temporal neural network identification model is used to perform temporal characterization modeling of the topological features, outputting heart failure compensation identification results and management prompts. This application can stably identify the trend of pleural fluid accumulation under postural and respiratory disturbances, enabling intelligent monitoring and proactive intervention for early heart failure compensation abnormalities.

The invention relates to the technical field of data processing, in particular to a heart failure compensation recognition system based on artificial intelligence, and provides the following scheme that electrocardiosignals, biological impedance signals, posture information and thoracic displacement signals are obtained, body position correction and respiration separation are conducted on the biological impedance signals according to the posture and respiration information, and biological impedance is calculated; correlation modeling is conducted on the biological impedance and the time sequence characteristics of the electrocardiosignals, an electro-hydraulic coupling threatening ring reflecting the electrophysiology and liquid conduction lagging relation is constructed, and topological characteristics such as the annular area, the annular direction, the communication number and the morphological stability are extracted; and performing time sequence representation modeling on the topological features through a time sequence neural network recognition model, and outputting a heart failure compensation recognition result and control prompt information. The chest fluid accumulation trend can be stably recognized under the body position and breathing interference, and intelligent monitoring and active intervention on heart failure early-stage compensation abnormity are achieved.