38 results about "Pulse stream" patented technology

The invention relates to the technical field of artificial intelligence, and discloses a spiking neural network-based robot control method, which comprises the following steps of: acquiring dynamic visual information of an external environment, and encoding the dynamic visual information to generate a dynamic visual pulse stream; collecting motion state information of the robot, and encoding the motion state information to generate a motion state pulse code; through a pulse attention gate control mechanism, computing resources of the pulse neural network are dynamically allocated according to task priorities; inputting the dynamic visual pulse streams and the motion state pulse codes into a pulse neural network, processing the dynamic visual pulse streams and the motion state pulse codes by the pulse neural network according to computing resources, and outputting dynamic visual features and motion state features; fusing the dynamic visual features and the motion state features to generate a visual-motion pulse signal; and converting the vision-motion pulse signal into a motion control instruction, and controlling the robot to execute a corresponding action according to the motion control instruction.

The invention provides a three-dimensional depth estimation method based on pulse streams, and belongs to the technical field of three-dimensional computer vision and neuromorphic imaging. The method comprises the following steps: (1) constructing a training and testing data set of a synthetic pulse stream; (2) constructing a pulse stream data set of a real scene; (3) constructing a three-dimensional depth estimation network model based on the pulse flow; (4) training a three-dimensional depth estimation network model based on the pulse flow; and (5) performing depth prediction by adopting the trained pulse stream-based stereo depth estimation network model. According to the invention, depth estimation can be directly carried out from the original pulse stream data, and a high-definition depth image is obtained without sequence framing or other post-processing; compared with an existing method, the method has better interpretability and higher prediction precision under a pulse flow data test set through iterative updating of the multi-layer and multi-scale cyclic pulse neural network and an algorithm of the pulse neural network.

An optical transmitter for quantum key distribution. The transmitter comprises a first and a second. Each laser is configured to emit phase-randomised pulses. The transmitter further comprises an optical combiner configured to combine the pulses emitted from the first and second laser into a combined stream of pulses based on an adjustable combining ratio. The transmitter further comprises a Mach-Zehnder interferometer with a short and a long arm. The interferometer has an input port configured to receive the combined stream of pulses and an output port configured to provide an output of the interferometer. At least one of the arms comprises a phase controlling element to control a phase between the short and long arm.

The invention provides a radar signal sorting method based on an enhanced Transform cascade architecture (ETCA), and solves the limitation of a traditional method on novel radar and overlapped pulse flow classification. The core of the method is developed around model structure design, a model training method and a model reasoning method. The model structure design is integrated with a bidirectional LSTM time sequence encoder, a nuclear attention enhanced Transform feature refiner and a feature adapter; the model training method adopts a full-participation pre-training combined online increment adaptation strategy; according to the model reasoning method, a preprocessed pulse description word sequence serves as input, after pre-training model feature extraction is called, the probability is output through an adapter, and the maximum value serves as a sorting result. The method rapidly adapts to the dynamic complex electromagnetic environment, and high-precision sorting is achieved.

The application discloses a neuromorphic auditory event-based anti-reverberation processing method and system for security scene abnormal voiceprint recognition, which comprises the following steps: acquiring an asynchronous auditory event sequence in a high-reverberation space; anchoring an initial pulse cluster triggered by a direct sound wave and extracting the same as a reference feature; constructing a channel-level dynamic refractory period mask window based on the reference feature and an estimated reverberation duration; determining a newly-arrived pulse cluster similar to the reference feature as a delayed ghost pulse generated by an echo and intercepting and filtering the same within the mask window action period, so as to generate a pure pulse stream; and inputting the pure pulse stream into a pulse neural network for recognition. The application effectively filters out redundant pulses generated by physical echoes from a data source in a native asynchronous pulse domain, avoids high-cost waveform reconstruction, can effectively prevent the subsequent pulse neural network from making a misjudgment and power saturation due to redundant data, improves the accuracy, robustness and real-time performance of the system in the abnormal voiceprint recognition under a high-reverberation environment, and meanwhile, retains the low-power consumption advantage of event-driven sensing.

The invention discloses a PRI sorting method and system in a complex electromagnetic environment, and the method specifically comprises the steps: carrying out the sorting and extraction of a pulse sequence through employing a dynamic two-dimensional feature vector matching algorithm, and obtaining a pulse sequence of a fixed and staggered PRI modulation signal; then carrying out sorting extraction on the residual pulses by using a square sine interpolation algorithm to obtain a pulse sequence of the jitter PRI modulation signal; then, through a feature recognition method, obtaining modulation features of the pulse sequences of the fixed and staggered PRI modulation signals and the pulse sequences of the jitter PRI modulation signals; and finally, carrying out intra-frame blending on the sorting result of the pulse stream, and carrying out inter-frame blending on the sorting result of the previous pulse stream to finish sorting. According to the method, the PRI sorting accuracy is improved, the batch increasing phenomenon is reduced, multiple PRI modulation patterns can be sorted and recognized, the pulse sequences belonging to different radar radiation sources are separated from a large number of mixed radar signals, and a reliable data basis is provided for subsequent recognition, positioning and confrontation.

The application discloses a time-domain-based analog Softmax and Top-K calculation circuit, and belongs to the technical field of integrated circuits and artificial intelligence accelerator. N The circuit comprises one parallel analog channel and a global module. The analog channel comprises a voltage-controlled oscillation unit, an edge pulse shaper, a state latch and an output control switch, and is used for converting an input analog voltage into an extremely narrow pulse stream with a corresponding frequency; the global module comprises normalization logic, Top-K logic and global control logic, and is used for counting the total number of global pulses and the winning state of the channel, and jointly issuing a channel enable bus; through the analog channel and the global module, exponential normalization and feature screening are originally completed in the analog time domain, and high-precision analog-to-digital converters and digital divider trees required by each channel are completely saved; moreover, through a pulse competition screening Top-K mechanism, the processing of redundant data and dynamic power consumption overhead are completely eliminated, and the energy efficiency ratio of an AI chip is improved.

The application belongs to the technical field of image processing, and discloses a pulse video signal visualization FPGA hardware and a reconstruction method, which comprises a Capture module, an SNN_IP algorithm module and a MIPI_OUT module, the Capture module is used for storing a pulse signal captured by a camera and pre-processing the pulse signal, the SNN_IP algorithm module is used for reconstructing gray pixel data from the pre-processed pulse data by using a TFP algorithm, and the MIPI_OUT module is used for parallel output of the reconstructed pixel data.The application solves the problem in two parts of algorithm optimization and system level optimization respectively. In the aspect of algorithm optimization, the TFP algorithm is used to process the video pulse stream on the reconstruction algorithm, and the fast reconstruction of the pulse pixel is realized. In the aspect of hardware level optimization, the pipeline structure is adopted, and the running time is greatly reduced under the premise of limiting resources.

The application discloses a LOS state detection circuit for an SFP optical module, which comprises an input inversion module, a signal conditioning and pulse expansion module and a threshold comparison and decision module connected in sequence. The input inversion module comprises a receiving original port link state indication signal, which is used for logical inversion of the original signal and enhancement of driving capacity. The signal conditioning module converts the data pulse stream output by a PHY chip into a smooth direct current voltage by using peak holding and RC expansion technology. The threshold comparison module intelligently judges whether the voltage represents valid signal activity through a hysteresis comparator, and finally outputs a hardware LOS signal which strictly conforms to the SFP MSA standard and is highly stable. The application realizes reliable optical signal loss detection in a full hardware mode, solves the problems of unstable output and easy misjudgment of a traditional scheme, and has the advantages of low cost, small area, fast response and high reliability, and is particularly suitable for integrated SFP optical module design.

The invention provides a digital orthogonal phase-locked amplifier based on a Sigma-Delta modulator. Comprising a Sigma-Delta modulator, a digital low-pass filter, an orthogonal phase-locked amplifier, a dynamic threshold module and a scaling multiple calculation module, the output end of the Sigma-Delta modulator is electrically connected with the input end of the digital low-pass filter, and the Sigma-Delta modulator is used for performing noise shaping on an input digital signal, converting the input digital signal into a high-speed low-resolution digital pulse stream and pushing quantized noise energy to a high-frequency area outside a target frequency band; the output end of the digital low-pass filter is electrically connected with the input end of the orthogonal lock-in amplifier, and the digital low-pass filter is used for filtering high-frequency quantization noise and downsampling a high-speed pulse flow into a high-precision multi-bit digital signal in a target frequency band, and the high-dynamic-range and low-noise weak signal detection can be realized on the premise of low cost.

This invention relates to the field of image recognition and automated control technology, specifically disclosing an automatic sorting and defect detection system for toy parts based on image recognition. The system includes an event vision sensing module, an asynchronous pulse stream processing module, a multimodal feature fusion and defect determination module, and a high-speed sorting execution module. By capturing the asynchronous pulse stream of part movement through an event camera, a blurred spatiotemporal contour is reconstructed and adaptively fused with global color image features. This achieves high-precision identification of part categories and simultaneous detection of multiple defects such as size, scratches, and color unevenness, ultimately driving the execution mechanism to complete the sorting. This invention improves detection accuracy, speed, and system robustness.

The invention discloses a real-time simulation resolving system for a complex electromagnetic environment. The system adopts a heterogeneous architecture combining radio frequency scene real-time resolving software and a hardware acceleration unit. Comprising a high-performance computing processing platform, radio frequency scene real-time resolving software and a hardware acceleration unit. The core of the system is that task division is realized through software and hardware cooperation: the calculation processing platform runs radio frequency scene real-time resolving software and is responsible for executing beat-level calculation, including analysis of a guide instruction and an external message, generation of a radiation source signal template and calculation of scene characteristic parameters; the hardware acceleration unit is specially used for pulse-level calculation, generates and modulates a multi-path radiation source pulse sequence in real time according to a signal template and scene characteristic parameters, and finally outputs a high-density staggered pulse stream through a high-speed optical fiber interface after interweaving and sorting. By means of the beat-pulse secondary calculation model, the problems that a traditional scheme is poor in real-time performance and low in fidelity are effectively solved, environment pulse density real-time calculation not lower than one million pulses per second can be supported, and the high-precision and real-time performance test verification requirements of electronic information equipment in the complex and changeable electromagnetic environment are met.

The invention provides a labyrinth pulse flow type micro-channel liquid cooling heat dissipation module which is arranged by being matched with a plurality of heat sources arranged in an array, so that all the heat sources obtain relatively stable uniform-temperature heat dissipation performance, stable and reliable operation of a product is ensured, the structure is simple, and excessive product space is not occupied. The heat exchanger is adaptive to a product with N rows of heat sources in the width direction, and N is a natural number larger than or equal to 3. The lower cover plate comprises a reference plane area and a flow channel assembly formed in a downward concave mode, the flow channel assembly comprises an inlet flow dividing channel, an outlet flow collecting channel, (N-1) sets of flow dividing channels arranged in parallel and a set of flow collecting heat dissipation channels, and the flow collecting heat dissipation channels are arranged corresponding to the heat sources in the corresponding columns on one side of the width direction; the inlet shunting channel is respectively communicated with the inlet of each group of shunting channels; a liquid inlet assembly; and a liquid outlet assembly.

Method and system for determining a pulse repetition frequency (PRF) pattern from a stream of pulses are disclosed. Intervals between successive pulses are determined and stored in a buffer at sequential indices, designating a first interval as a reference value, and specifying a lower bound, Kmin, of a number of intervals within the PRF pattern. Initial Kmin interval values are entered in an array, and each subsequent interval that differs from the reference value is appended to the array. Upon reading a value matching the reference value, determining a count of entries, identifying a number of successive values matching successive entries from start of the array. If the number is less than the count, adding the number of successive values, in addition to each succeeding value from the buffer that precedes a value matching the reference value, and continue appending. If the number reaches the count, PRF pattern is determined.

The present invention provides, a device for a controlling a quantum computer is provided. The device comprises a Quantum Computer Control System, QCCS, channel. The QCCS channel comprises a transceiver configured generate a 1-bit pulse stream and modulate the 1-bit pulse stream by pulse density modulation to generate a modulated signal, a low pass filter configured to filter the modulated signal and output a corresponding qubit control signal for a quantum computer. The invention further comprises a corresponding system, method and a use of a transceiver for controlling a qubit of a quantum computer.

A high-efficiency facial expression recognition method based on a double-branch pulse Transformer, the method comprising: an image acquisition module acquires a facial original image; an image preprocessing module performs a preprocessing operation on the facial original image to obtain a facial standard image; a pulse space-time coding module in a facial recognition network converts the static space features of the facial standard image into a space-time pulse stream to construct a facial feature pulse sequence stream; a multi-scale feature extraction module performs hierarchical feature extraction and long-range dependence modeling on the facial feature pulse sequence stream to obtain global spatial distribution features; a cross-branch guided fusion module performs cross modulation and integration of multi-dimensional expression representation on the global spatial distribution features to obtain a feature representation tensor; and a pulse classification output module performs decision reasoning on the feature representation tensor and outputs a facial expression recognition result. Effect: high-precision facial expression recognition can be realized under ultra-low power consumption.

The invention relates to the technical field of radiation source identification, and provides a pulse stream segmentation and signal pattern identification method and device. The method comprises the following steps: performing outlier detection on a radar pulse stream sequence to determine an anchor point; segmenting the radar pulse stream sequence into a plurality of pulse subsequences according to the anchor points; performing feature extraction on the plurality of pulse subsequences to obtain original features; re-weighting the original features to obtain features to be classified; and predicting a signal pattern of the to-be-classified features to calculate classification loss, and iteratively updating to obtain a target recognition model so as to use the target recognition model to perform signal pattern recognition on the plurality of segmented pulse subsequences. According to the method, the problem that the performance of a radar signal pattern recognition algorithm is reduced due to the fact that outliers exist in a radar pulse flow sequence is solved.

The application discloses a wave front detection system and method based on a pulse camera, and belongs to the technical field of adaptive optics. The system comprises a micro optical unit for dividing an incident wave front into a sub-spot array; a pulse camera, whose photosensitive surface is located at the focal plane of the micro optical unit, for receiving the sub-spot array and outputting spot change information in the form of an asynchronous pulse stream; and a data processing unit for processing the pulse stream in real time, outputting a transient wave front phase through pulse classification, centroid calculation and wave front reconstruction. The application utilizes the high time resolution and asynchronous output characteristics of the pulse camera, can track the high-speed movement of a spot with a precision of microseconds, significantly improves the time resolution and dynamic range of wave front detection, and is suitable for fields such as astronomical observation, high-speed flow imaging, laser communication and the like which require high-frequency and large-aberration measurement.

The utility model discloses a jet pulse fluidization rocking screen which comprises a machine frame, a screen box and a screen device, the screen device is provided with at least one screen set, and each screen set is composed of a plurality of layers of screens arranged from top to bottom. A pulse blowing device is arranged below the screen group in the screen box, the pulse blowing device is provided with a plurality of blowing pipes, an airflow pipe and a shunt pipe, the blowing pipes are horizontally arranged at intervals, and a plurality of air jet hole rows which are arranged at intervals are formed in the surfaces, facing the screens, of the blowing pipes; each air injection row is composed of a plurality of air injection holes arranged at intervals in the circumferential direction of the air blowing pipes, the airflow pipes are communicated with the air blowing pipes in a one-to-one correspondence mode through a plurality of flow dividing pipes, and a pulse air generating device for providing pulse airflow for the airflow pipes is installed outside the screen box. A dust removal pipe communicated with the hollow cavity of the screen box is installed on the outer side wall of the screen box and located below the screen mesh device, and a dust removal mechanism communicated with the dust removal pipe is arranged on the screen box. Compared with the prior art, the jet effect of airflow is good.

The invention relates to the technical field of power electronic measurement, and discloses a current transformer zero flux control method and system based on magnetic potential balance constraint, and the method comprises the steps: converting a current signal of a current transformer into a pulse sequence, and generating an event drive data format; processing the pulse sequence by using an event-driven wavelet decomposition algorithm, and outputting a multiband pulse stream; analyzing the multiband pulse stream, and calculating a coupling relationship between frequency components; calculating a magnetic potential balance compensation parameter by using a pulse frequency coding algorithm based on the coupling relationship between the frequency components; the magnetic potential balance compensation parameters are converted into an asynchronous control pulse sequence, and compensation current is generated and output to a magnetic potential balance execution mechanism; by introducing a neuromorphic calculation and event-driven processing mechanism, the problem of inherent delay of a traditional synchronous clock sampling method is solved, and microsecond-level response control over the current transformer in the high-frequency switching power supply system is achieved.

The invention discloses a microwave photon enhancement device for transient weak pulse spectrum sensing and a sensing method. The microwave photon enhancement device comprises an electro-optical modulation module, a first transient pulse replication branch, a second transient pulse replication branch and an autocorrelation detection module. A to-be-measured signal forms a first carrier signal at the electro-optical modulation module modulating the carrier light source; the first pulse copy loop copies the first carrier signal to form a first pulse stream signal; the second pulse copy loop copies the first carrier signal, and forms a second pulse stream signal after passing through the optical delayer; the self-correlation detection module carries out self-correlation operation on the first pulse flow signal and the second pulse flow signal. According to the microwave photon enhancement device for transient weak pulse spectrum sensing and the sensing method, the pulse replication loop is adopted to perform periodic replication on transient extremely narrow pulses, instantaneous self-correlation operation of a radio frequency domain of an analog signal to be detected is realized through the microwave photon link, the signal-to-noise ratio of the signal to be detected is improved, and the sensitivity of an instantaneous frequency measurement system is improved.

The application relates to the technical field of artificial intelligence and computing resource optimization, and discloses a neural network dynamic computing resource allocation method based on information entropy. The method embeds an information entropy gating module at the end of each processing layer of a deep pulse neural network, calculates the binary information entropy of the current layer output pulse stream in real time, dynamically determines whether the input data has been sufficiently processed in the inference stage, and when the information entropy is lower than a preset threshold, determines that the model has reached sufficient certainty for the current input, skips the subsequent unexecuted processing layer, and directly transmits the current output to the output layer. The method makes simple input consume fewer computing layers and complex input consume more layers, and realizes adaptive on-demand allocation of computing resources. Behavior level simulation verification shows that in a 12-layer pulse neural network, about 40% to 60% of the inputs can exit in the 6th to 9th layer in advance, and the inference delay is reduced by an average of 30% to 50%.

The invention provides a dual-channel weighted SDIF radar signal sorting method, and relates to the technical field of electronic reconnaissance signal processing. The method comprises the following steps: firstly, receiving an interlaced pulse stream and recording PDW data through a radar signal reconnaissance system, then carrying out local selection by utilizing an improved sequence time difference histogram (SDIF) method, and extracting candidate PRI by adopting a double-weight peak detection strategy (low order is dominated and high order is dominated); and after harmonic elimination, pulse grouping, PRI refinement and deletion completion, and PRI approximate same sequence combination, the signal type is determined according to a stability index. The method solves the problem that the traditional SDIF method is difficult to effectively sort the stagger repetition frequency, and improves the sorting accuracy of the fixed repetition frequency radar signals in the complex electromagnetic environment.

The invention discloses a PLC control method and system for stable control of ore dressing pulp flow, and relates to the technical field of gravity ore dressing. The control system mainly comprises an upper computer DCS (Distributed Control System), a PLC (Programmable Logic Controller), a touch screen, a flow meter, a concentration meter and an execution assembly. The control method mainly comprises the following steps: acquiring ore pulp flow data of each pipeline through a detection element; a dynamic median-arithmetic average filtering fusion method is adopted to filter the ore pulp flow data, and a dynamic weight coefficient alpha is calculated according to a formula alpha = sigma / (sigma + delta x + epsilon) so as to eliminate the influence of pulse flow noise on collected data during ore feeding; and the filtered flow data serve as feedback signals to be input into a pulse suppression PID controller, and the pulse suppression PID controller adopts a control strategy combining incomplete differential, differential advance and integral separation to automatically adjust the opening degree of an execution assembly. According to the invention, the anti-interference capability of the PLC control system to noise is obviously improved, and the method has the advantages of high precision, fast response, strong stability and the like in flow control of ore pulp such as feeding of gravity beneficiation such as spiral beneficiation.

The application belongs to the technical field of computer vision and multi-modal data processing. The application provides an ANN-SNN hybrid architecture sound source positioning method based on event driving. The ANN branch is used to process spatial semantic information, realize high-fidelity semantic understanding of acoustic scene spatial distribution, and capture sound event timing dynamic characteristics by the SNN branch, so as to play the advantages of event driving, low power consumption and noise resistance. The DBSI mechanism is designed to extract semantic and time clues from the same input pulse stream in parallel, realize cross-modal information alignment and integration through a multi-scale feature fusion strategy, and improve multi-modal perception collaboration efficiency. The SGAM module introduces a hard attention mechanism of acoustic pulse gating management, accurately identifies the significant starting time of the sound event, activates visual positioning update only at the key time point, suppresses the positioning tailing phenomenon, and improves the timing resolution and response accuracy.

The invention discloses a manned unmanned aerial vehicle autonomous obstacle avoidance control method and system based on flight environment perception, and relates to the technical field of intelligent control. The method comprises the following steps: acquiring a multi-physical field tensor of a navigation environment through a complementary detector, and converting the multi-physical field tensor into a sparse asynchronous pulse stream through event-driven coding; inputting the sparse asynchronous pulse flow into a first decision maker, executing multi-layer recursive analysis based on spatial voxel-physical feature combination, and determining a reference control instruction; constructing a manifold keeping controller, executing quadratic programming based on a first constraint condition and a second constraint condition on the reference control instruction, and solving an optimal control quantity as a target control sequence; and according to the target control sequence, unmanned aerial vehicle obstacle avoidance navigation control is executed. The technical problem of poor obstacle avoidance reliability caused by insufficient environment perception, lagging obstacle avoidance decision and insufficient control constraint coupling of the manned unmanned aerial vehicle in a complex flight environment in the prior art is solved, and the technical effect of improving the autonomous obstacle avoidance reliability of the manned unmanned aerial vehicle is achieved.

The utility model relates to a kind of buffer based on peristaltic pump pulse inhibition, comprising: encapsulation shell, hard pipe is equipped in encapsulation shell, and adapter is installed in encapsulation shell, and the opening of hard pipe is sealedly connected with adapter, and fluid inlet is threadedly connected with adapter, and adapter is communicated with flow outlet;After the hard pipe of pulse fluid is introduced, buffer area is provided for fluid, channel volume is increased to weaken pulse energy, and the hard pipe of U shape prolongs fluid flow path and residence time, further enhances buffering, so that pulse energy is gradually dissipated, to ensure that fluid is smoothly transported to subsequent equipment or system.

The present application is a kind of deep coal seam pulse hydraulic fracturing device and method, the device comprises: ground pulse modulation system, including pulse jet generation device for optical casing, for modulating continuous pumping fracturing fluid into pulse fluid and injecting into wellbore on the ground;Pulse fluid can pulse hydraulic fracturing of coal seam;Downhole pulse modulation system, at least including coiled tubing pulse jet generation device, coiled tubing pulse jet generation device is used to generate pulse jet in the set position in well, and pulse jet can jet fracturing to coal seam.The present application proposes an optimization mechanism based on geology-engineering dessert drilling conditions, through flexible selection of operation mode, realizes the depth coupling of pulse generation device and existing mature fracturing process of deep coal seam, applies cyclic alternating load type shock wave to deep coal seam, makes it stress oscillation, fatigue failure, thereby reduces the crack initiation pressure, promotes the formation of crack network, and is expected to become a new technology for deep coal seam reservoir reconstruction.

The present application relates to the technical field of power electronic measurement, and discloses a current transformer zero magnetic flux control method and system based on magnetic potential balance constraint, wherein the current transformer zero magnetic flux control method based on magnetic potential balance constraint comprises: converting a current signal of a current transformer into a pulse sequence to generate an event-driven data format; processing the pulse sequence by using an event-driven wavelet decomposition algorithm to output a multi-band pulse stream; analyzing the multi-band pulse stream to calculate a coupling relationship between frequency components; based on the coupling relationship between the frequency components, using a pulse frequency coding algorithm to calculate magnetic potential balance compensation parameters; converting the magnetic potential balance compensation parameters into asynchronous control pulse sequences to generate compensation currents and output to a magnetic potential balance execution mechanism; the present application overcomes the inherent delay problem of the traditional synchronous clock sampling method by introducing a neuromorphic computing and event-driven processing mechanism, and realizes microsecond-level response control of the current transformer in a high-frequency switching power supply system.

The invention discloses a radar signal increment sorting recognition method and system based on pulse stream synthesis playback, mainly solving the problem of poor increment sorting recognition effect in the prior art, and the implementation scheme comprises the following steps: obtaining a radar signal pulse description word sequence collected offline and a corresponding pulse-by-pulse radar tag; dividing the training data into a plurality of pure subsequences, adding the pure subsequences into a memory bank taking a category as a unit, extracting the pure subsequences from the memory bank according to a category probability, and performing time translation and interleaving synthesis to generate an aliasing playback sample; iteratively training the neural network by using the aliasing playback sample, and obtaining a trained incremental neural network through verification; a radar signal pulse description word sequence to be sorted and recognized is obtained in real time and input into the trained incremental neural network to obtain pulse-by-pulse radar tags, and deep feature scoring and stratified sampling results are obtained through various types of pure subsequences input into the network to maintain a memory bank. The method can improve the effect of incremental sorting recognition, is high in generalization ability, controllable in training cost and high in overall efficiency, and can be used for electronic reconnaissance in a dynamic signal environment.