31results about How to "Reduce the impact of recognition accuracy" patented technology

The present invention discloses an action sequence detection-based user identity identification method applied to an indoor WiFi environment. The invention aims to solve the technical problem of low accuracy of an existing user identity identification method based on WiFi signals. According to the technical schemes of the method, sensing is realize through using devices such as commercial WiFi and notebooks; sensed data are preprocessed, so that the quality of the data can be improved; features are extracted from the data so as to depict user identities; a classification model is constructed; and the classification model is invoked to calculate the probability distribution of user identities of single actions; the probability distribution results of all identifiable actions in action sequences are put into statistics, so that identity identification is realized; and in the process of the identity identification, the data are used to accurately depict waveforms, and the user identities are judged many times through the action sequences, and therefore, the influence of complex environments on identification accuracy can be decreased, and high-accuracy identity identification is realized based on the results of a plurality of times of judgment.

The invention discloses an adversarial sample defense method based on feature remapping and application. The method comprises the steps of: constructing a feature remapping model, wherein the featureremapping model comprises a significant feature generation model used for generating significant features, a non-significant feature generation model used for generating non-significant features and ashared discrimination model used for discriminating the authenticity of the significant features and the non-significant features; constructing a detector according to the significant feature generation model and the non-significant feature generation model, wherein the detector is used for detecting an adversarial sample and a benign sample; constructing a re-identifier according to the significant feature generation model, wherein the re-identifier is used for identifying the category of the adversarial sample; when adversarial sample detection is carried out, connecting a detector to the output of the target model, and carrying out adversarial sample detection by utilizing the detector; and during adversarial sample identification, connecting the re-identifier to the output of the target model, and performing adversarial sample identification by using the re-identifier. The dual defense effect of detection and re-identification of the adversarial sample can be realized.

The invention discloses a programmable deep neural network processor, which includes a program control unit, a filter buffer area, and a feature map buffer area. The feature map buffer area is used to cache multiple feature maps. The filter buffer area is used for It also includes a layer processing engine, and the convolution unit part of the layer processing engine includes a multiplication and accumulation unit, a convolution accumulation unit and a feature map accumulation unit arranged in sequence, and the feature map cache The area and the filter buffer area are connected to the input end of the layer processing engine, and a data shaping and multiplexing unit is also arranged between the feature map buffer area and the input end of the layer processing engine. The present invention realizes a low-power, low-cost programmable deep neural network processor through the multiplexing control of the multiply-accumulate unit, the feature map data reading control, the feature map accumulation control, and the redundant data elimination control.

The invention discloses a method for optimizing an audio-visual cognitive event-related potential experimental paradigm. The method comprises the following steps of: on the basis of two first optimal Oddball sequences, performing combined stimulation of different matching modes of target stimulation and background stimulation modes on each subject respectively to induce a cognitive electroencephalographic experiment, establishing a plurality of second Oddball sequences, synchronously recording a plurality of second electroencephalographic signals, performing off-line analysis, and acquiring a plurality of second event-related potential data; acquiring the second Oddball sequence which corresponds to the second event-related potential data with the highest identification correct rate, wherein the second Oddball sequence is taken as a second optimal Oddball sequence of each subject; and gradually changing the appearance proportion of target stimulation, the duration and the stimulation interval of each subject in the second optimal Oddball sequence to induce cognitive electroencephalographic signals, acquiring a plurality of event-related potential data respectively, performing off-line analysis, and constructing an effective Oddball sequence scheme. The method is used for the auxiliary identification of the target stimulation, and can accurately position the target stimulation.

The invention relates to a face picture compensation recognition method based on ship security data recognition, which includes: respectively performing concentric ring partitions on the recognized face picture and the standard face picture, and dividing the features of each area in the recognized face picture Points are matched with the feature points of the corresponding area in the standard picture; the distance between the recognized face picture and the corresponding two feature points in the standard picture is respectively obtained, and the picture deformation ratio is obtained according to the distance between the corresponding two feature points ; The deformation intensity data is obtained through the deformation ratio of the image; the deformation amount of the facial feature points is obtained through the deformation intensity data; and the feature points are compensated according to the deformation amount. Through the technical method proposed by the invention, the influence of the focal length on the fat and thin deformation of the human face is reduced, thereby reducing the influence on the recognition accuracy of the human face and effectively improving the detection accuracy.

The invention discloses a user identity recognition method based on action sequence detection in an indoor WiFi environment, which is used to solve the technical problem of poor accuracy of the existing WiFi signal-based user identity recognition method. The technical solution is to use commercial WiFi and notebook devices to achieve perception, preprocess the perceived data, improve data quality, extract features from the data, characterize user identities, construct a classification model, and call the classification model to perform a single action on the user. The probability distribution calculation of the identity is carried out, and the identity recognition is realized by making statistics on the probability distribution results of all identifiable actions in the action sequence; Judgment for one time, reduce the impact of complex environment on the recognition accuracy, and integrate the results of multiple judgments to achieve high-accuracy identification.

The present invention relates to an information recognition method, electronic device and readable storage medium of a form image, the method comprising: identifying the form category of the form image to be processed, and determining a first standard template corresponding to the form category of the form image and the second standard template; using a predetermined correction rule to perform angle correction on the form image, using a pre-trained text line detection model to detect all text lines on the angle-corrected form image, and combining the detected text lines with The position of the text line in the first standard template is compared to determine the position correspondence between the first standard template and the angle-corrected form image; according to the position correspondence, the key information position label box on the second standard template is mapped to On the form image after angle correction, the key information on the form image is recognized by using a predetermined character recognition model. The invention effectively improves the information recognition accuracy and efficiency of form images.

The present invention is applicable to the medical field, and provides a method and device for identifying red blood cells infected by Plasmodium, the method comprising: acquiring forward scattered light signals and side scattered light signals of cells in a sample, and optional fluorescent signals; Obtain a first two-dimensional scatter diagram according to the forward scattered light signal and side scattered light signal, or obtain a three-dimensional scatter diagram based on the forward scattered light signal, side scattered light signal and fluorescence signal; The first two-dimensional scattergram or the three-dimensional scattergram shows cells identified as Plasmodium-infected red blood cells within a predetermined region. The technical solution provided by the invention has the advantage of high recognition accuracy.

The invention relates to an OCR recognition method based on a deep learning model and a terminal, belonging to the field of data processing. The present invention obtains a set of single-character images by dividing the preset character segment image into multiple single-character images; the preset first OCR deep learning model sequentially recognizes the elements in the single-character image set to obtain the first feature vector A set; a single-character image corresponding to a first feature vector; according to a preset feature database, converting each first feature vector in the first feature vector set into a corresponding single character to obtain a single-character set; A record in the feature database saves a single character and a feature vector corresponding to the single character; arranging the elements in the single character set to obtain a character string corresponding to the character fragment image. Realize improving the anti-interference ability of OCR recognition characters.

The invention discloses a feature-level fusion identification method of radar jamming signals based on a deep convolutional neural network, which belongs to the field of radar jamming signal identification. The present invention aims at the problem that the characteristic parameters of the current radar interference signal rely on manual extraction, which is easily affected by noise and features redundancy. Including establishing a radar jamming time domain data set, extracting feature vectors from the radar jamming time domain data in the radar jamming time domain dataset in two different forms, and then merging the two extracted feature vectors in series; using the fused feature vector training Support vector machine to obtain the trained radar interference signal feature-level fusion recognition model; input the collected test samples into the recognition model to obtain the radar interference signal recognition result. The invention uses CNN to extract the deep features of the radar interference signal, and designs different radar interference signal data fusion models at the feature level, so that the signal identification is free from the influence of noise, and the feature redundancy phenomenon is eliminated at the same time.

The embodiment of the present application discloses an eye tracking method and related products. The method is applied to electronic equipment, and the electronic equipment includes: a front-facing camera, and the front-facing camera supports HDR mode. It is characterized in that the method includes the following Step: the electronic device collects the first image through the front camera, and judges whether the first picture contains leukoplakia; when the electronic device determines that the first picture contains leukoplakia, the electronic device triggers the HDR mode to collect n images, and among the n images The EV value of each image is different; the value of n is an integer greater than or equal to 3; the electronic device synthesizes the n images into a second image through the HDR algorithm, and transmits the second image to the eye tracking algorithm for identification to realize eye tracking, which improves the user's vision. experience.

An embodiment of the present invention provides a three-dimensional face recognition method, device, terminal equipment, and computer-readable medium, the method comprising: acquiring a near-infrared image and a depth image of an actual scene; normalizing the near-infrared image and the depth image process, and extract the first multi-dimensional feature vector of the face in the actual scene; calculate the deflection angle of the face according to the key points of the face in the near-infrared image, and deflect the 3D model of the face entered in advance according to the deflection angle The angle is rotated, and the rotated 3D face model is projected on a plane to obtain the input near-infrared image and depth image of the face; normalize and extract the second multi-dimensional feature vector; calculate the first multi-dimensional feature The similarity between the vector and the second multi-dimensional feature vector, and face recognition based on the similarity, can effectively recognize faces with different deflection angles, improve the efficiency of three-dimensional recognition, and reduce the impact of background on face recognition accuracy.