9results about How to "Improve decoding accuracy" patented technology

The invention discloses a face key point detection network training method, a face key point detection method and a medium, and belongs to the technical field of face key point detection, and the method comprises the steps: obtaining a face image data set; in the face image data set, each sample is a face image marked with key point coordinates; for each sample in the face image data set, key points in the sample are divided into boundary endpoints and boundary internal points, and coordinates of the key points are converted to a heat map coordinate system; under the heat map coordinate system, generating a true value heat map for each boundary endpoint by adopting isotropic Gaussian distribution, generating a true value heat map for each boundary internal point by adopting anisotropic Gaussian distribution, and forming a training data set by the face image and the true value heat map of each key point in the face image; and training a face key point detection network used for predicting a heat map of each key point in the face image by using the training data set to obtain a trained face key point detection network. According to the invention, the detection precision of the face key points can be effectively improved.

Embodiments of the present application provide an audio data processing method, a server and a program product. The method is applied to a server, and the method comprises: performing redundant coding processing on to-be-processed audio data to obtain coded audio data, in a case where it is determined based on network state information that the redundant coding process is to be performed; and performing a first redundancy processing operation on the coded audio data to obtain processed audio data. The method significantly improves the quality of audio transmission in a weak network environment, reduces audio delay, freezing and distortion, and improves user experience.

The invention discloses a low-duration calibration electroencephalogram decoding method and system based on invariant feature extraction, and relates to the technical field of electroencephalogram signal decoding, and the method comprises the steps: obtaining electroencephalogram data with labels, and carrying out Fourier transform to obtain amplitude information and phase information; inputting the phase information and the tag into a teacher network for training to obtain feature output of the teacher network; performing random pairing and reconstruction on the labeled electroencephalogram data to obtain reconstructed electroencephalogram data, inputting the reconstructed electroencephalogram data into a student network, and learning phase information extracted from a teacher network through the student network to obtain feature output of the student network; and acquiring real-time data, and inputting the real-time data into a student network to obtain an electroencephalogram decoding recognition result. Similarity features among different individuals are extracted through an invariant feature extraction method, when electroencephalogram signals of a new user are analyzed, the similarity features can be directly used for electroencephalogram analysis of a new subject, and the use efficiency of a non-intrusive brain-computer interface and the electroencephalogram recognition accuracy are remarkably improved.

The invention discloses a single-line transmission method, device and equipment and a computer readable storage medium, which are applied to the technical field of computers, and comprise the following steps: packaging internal bus parallel data into a parallel data message according to a register read-write message protocol, and converting the parallel data message into a serial bit stream; mapping each logic bit of the serial bit stream into a level signal of a corresponding duty ratio according to a duty ratio coding rule; the duty ratio coding rule is that a logic bit 0 represents that a single high level exists in a clock period, and the duty ratio of the single high level does not reach half, and a logic bit 1 represents that a plurality of discontinuous high levels exist in the clock period, and the duty ratios of the plurality of discontinuous high levels reach half; and controlling the level conversion of the single signal line through a switching circuit according to the level signal, and transmitting the coded bit stream to the slave device bit by bit. According to the invention, the problems of high PCB design difficulty, high cost, signal interference and data error risk caused by a large number of signal lines are solved.

The invention relates to a track circuit signal decoding method and device based on nonlinear transformation enhancement and a storage medium. The method comprises the steps that S1, track circuit signals are collected; s2, the collected track circuit signals are preprocessed; s3, performing first processing on the pre-processed signal, the signal of the positive half cycle and / or the signal of the negative half cycle to obtain a target signal; s4, performing FFT calculation on the target signal to obtain a spectrogram; s5, calculating a rough value of the center frequency of the carrier frequency signal and a rough value of the low-frequency signal based on the frequency doubling in the spectrogram; s6, determining the rough value of the center frequency of the target frequency multiplication based on the rough value of the center frequency of the carrier frequency signal; and S7, on the basis of the obtained rough value of the center frequency of the target frequency multiplication, identifying and obtaining the frequencies of the two spectrum peaks of the target frequency multiplication in the frequency diagram, and calculating the accurate value of the center frequency of the target frequency multiplication and the accurate value of the low-frequency signal. Compared with the prior art, the method has the advantages that the decoding precision is improved under the condition that the chip calculation amount is basically unchanged, and the like.

The application discloses a pinyin decoding method and related equipment, and relates to the technical field of natural language processing. The application first acquires a first text sequence containing a historical input text sequence and a current to-be-decoded pinyin string, and extracts semantic features corresponding to the text sequence by using a neural network model; then, corresponding statistical information is acquired by relying on a statistical language model. The statistical language model can store various types of text static statistical information, including candidate word entries and corresponding probability information, and the statistical information is mapped to a feature space consistent with the semantic features to obtain statistical features. The semantic features and the statistical features are fused to obtain fused semantic features, and a decoding result corresponding to the pinyin string is generated according to the fused semantic features. The application combines the semantic understanding ability of the neural network and the prior statistical advantage of the statistical language model, and takes into account the decoding accuracy and the use stability, is efficient in reasoning, and is suitable for various input method use scenarios.

The invention discloses an optical sharing crystal module, a detector and a height decoding method thereof, and emission imaging equipment. The light sharing crystal module comprises a scintillation crystal array, wherein the size of the scintillation crystal array is 2N * 2N; wherein N is a natural number; in the scintillation crystal array, except the scintillation crystals in the first row and the first column, the first row and the 2N column, the 2N row and the first column, and the 2N row and the 2N column, one side surface of each of the other scintillation crystals is provided with a light-transmitting window, and the light-transmitting windows of two adjacent scintillation crystals are arranged in a face-to-face manner; wherein the light-transmitting windows are regularly arranged, the light-transmitting windows in the first column and the 2Nth column are located in the 2nth row, and n is larger than or equal to 1 and smaller than or equal to N-1; the light-transmitting windows in the first row and the 2Nth row are located in the 2nth column, and n is larger than or equal to 1 and smaller than or equal to N-1.

The invention discloses an auditory space attention decoding method based on a low-density electroencephalogram cross-hemisphere attention mechanism, electronic equipment and a medium, and belongs to the field of brain-computer interface and neural signal processing. The method comprises the following steps: preprocessing low-density electroencephalogram signals, and respectively mapping the low-density electroencephalogram signals into three-dimensional tensors with a time dimension and a two-dimensional space dimension according to a left hemisphere and a right hemisphere; a space-channel attention module is utilized to extract key spatio-temporal characteristics of each hemisphere; dynamic interaction between left and right hemisphere features is modeled through a cross-hemisphere attention module, and information fusion is realized; and finally, classifying the fused features, and outputting auditory attention directions. According to the method, an electroencephalogram space-time structure is reserved through three-dimensional tensor mapping, feature discrimination is enhanced in combination with a double attention mechanism, high-precision decoding can still be achieved under the low-density condition that only 4-8 channels are used, meanwhile, the method has the advantages of being low in calculation complexity and high in physiological interpretability, and the method is suitable for wearable brain-computer interfaces and intelligent hearing-aid equipment.

The application relates to a local-global time sequence interaction network for unilateral directional motor imagery electroencephalogram decoding, and belongs to the technical field of brain-computer interfaces. The method comprises the following steps: a shallow CNN module is used to extract a localized time-frequency representation from an original EEG signal; a multi-scale time sequence feature extraction and compression module is used to extract complementary features through four parallel time sequence processing paths of a short time sequence convolution kernel, a long time sequence convolution kernel, an average pooling layer and a step convolution layer; a multi-scale fusion feature is reduced in dimension based on a deep separable convolution; a local-global time attention module is used to replace fixed position coding with learnable position coding in an encoder, model global time sequence dependence, perform adaptive average pooling on a segmentation length based on a time pyramid pooling module, and output multi-scale local features and splice. The organic fusion of multi-scale local features and global time sequence dependence features is realized, and the decoding performance of unilateral directional motor imagery electroencephalogram signals is effectively improved.