32 results about "Selective attention" patented technology

Disclosed is a deep long-term and short-term memory recurrent neural network acoustic model establishing method based on selective attention principles. According to the deep long-term and short-term memory recurrent neural network acoustic model establishing method based on the selective attention principles, attention gate units are added inside a deep long-term and short-term memory recurrent neural network acoustic model to represent instantaneous function change of auditory cortex neurons; the gate units are different in other gate units in that the other gate units are in one-to-one correspondence with time series, while the attention gate units represent short-term plasticity effects and accordingly have intervals in the time series; through the neural network acoustic model obtained by training mass voice data containing Cross-talk noise, robustness feature extraction of the Cross-talk noise and establishment of robust acoustic models can be achieved; the aim of improving the robustness of the acoustic models can be achieve by inhibiting influence of non-target flow on feature extraction. The deep long-term and short-term memory recurrent neural network acoustic model establishing method based on the selective attention principles can be widely applied to multiple voice recognition-related machine learning fields of speaker recognition, keyword recognition, man-machine interaction and the like.

The present invention discloses a system and method for the enhancement of attention and further the treatment of attention deficit disorder. The present invention also presents a virtual reality environment for enhancing attention shift, selective attention, and sustained attention by correlating a biofeedback device that estimates the current attention level of the patient.

This invention enables targeting, personalized measurement, and management of cognitive skills development by users, clinicians, teachers, and parents. The invention features a game based virtual learning curriculum for targeting and developing the underlying cognitive skills of executive functions. The methods and systems of the invention provide an effective and rapid video game-based training curriculum to improve the cognitive skills such as focused attention, sustained attention, cognitive inhibition, behavioral inhibition, selective attention, alternating attention, divided attention, interference control, novelty inhibition, delay of gratification, inner voice, motivational inhibition, and self-regulation. This curriculum utilizes: (i) each of the cognitive processes that underlie attention control and impulse inhibition; (ii) the identification of measurable and trainable cognitive skills; and (iii) game design and game mechanics that effectively train and enable retention of those skills. The game-based system provides a medical professional, clinician, parent, teacher and user with the ability to measure and manage training of targeted cognitive skills to reach a desired performance goal.

The invention discloses a CNN and selective attention mechanism based SAR image target detection method. An SAR image is obtained; a training data set is expanded; a classification model composed of the CNN is constructed; the expanded training data set is used to train the classification model; significance test is carried out on a test image via a simple attention model (a spectral residual error method) of image visual significance to obtain a significant characteristic image; and morphological processing is carried out on the significant characteristic image, the processed characteristic image is marked with connected domains, target candidate areas corresponding to different mass centers are extracted by taking the mass centers of the connected domains as the centers, and the target candidate areas are translated within pixels in the surrounding to generate an target detection result. According to the invention, the CNN and the selective attention mechanism are applied to SAR image target detection in a combined way, the efficiency and accuracy of SAR image target detection are improved, the method can be applied to target classification and identification, and the problem that detection in the prior art is low in detection efficiency and accuracy is solved mainly.

An agent learning apparatus comprises a sensor (301) for acquiring a sense input, an action controller (307) for creating an action output in response to the sense input and giving the action output to a controlled object, an action state evaluator (303) for evaluating the behavior of the controlled object, a selective attention mechanism (304) for storing the action output and the sense input corresponding to the action output in one of the columns according to the evaluation, calculating a probability model from the action outputs stored in the columns, and outputting, as a learning result, the action output related to a newly given sense input in the column where the highest confidence obtained by applying the newly given sense input to the probability model is stored. By thus learning, the selective attention mechanism (304) obtains a probability relationship between the sense input and the column. An action output is calculated on the basis of the column evaluated as a stable column. As a result, the dispersion of the action output is quickly minimized, and thereby the controlled object can be stabilized.

This invention enables targeting, personalized measurement, and management of cognitive skills development by users, clinicians, teachers, and parents. The invention features a game based virtual learning curriculum for targeting and developing the underlying cognitive skills of executive functions. The methods and systems of the invention provide an effective and rapid video game-based training curriculum to improve the cognitive skills such as focused attention, sustained attention, cognitive inhibition, behavioral inhibition, selective attention, alternating attention, divided attention, interference control, novelty inhibition, delay of gratification, inner voice, motivational inhibition, and self-regulation. This curriculum utilizes: (i) each of the cognitive processes that underlie attention control and impulse inhibition; (ii) the identification of measurable and trainable cognitive skills; and (iii) game design and game mechanics that effectively train and enable retention of those skills. The game-based system provides a medical professional, clinician, parent, teacher and user with the ability to measure and manage training of targeted cognitive skills to reach a desiredperformance goal.

The invention provides a multi-channel speech enhancement method based on semantic prior selective attention. The method comprises the following steps: picking speech signals from any directions in a reverberant environment by virtue of a multi-microphone array, collecting the multiple paths of speech signals and pre-processing the speech signals; detecting special activation words in the pre-processed speech signals by virtue of an activation word speech recognition model; processing signals which are not cut and include activation word segments, so as to obtain a complete activation word segment; analyzing the activation word segment by virtue of a multi-channel phase difference sound localization method based on reverberation robust, so as to obtain an acoustic wave reaching direction of a target sound source; and enhancing speech in the direction and inhibiting noise from other directions and room reverberation in a remote speak scene, so that enhanced speech in the target direction is obtained. The method provided by the invention is applicable to such occasions as intelligent household electrical appliance, smart home, vehicle-mounted and wearable devices and the like that remote speak type speech input and interaction are required, and the method is especially applicable to complex acoustic noise and interference environment occasions.

The invention relates to an objective advertisement design evaluation method based on a visual attention mechanism. The method overcomes the defect that the subjectivity of the traditional advertisement design evaluation method is too strong. The method comprises the steps of firstly, performing significance detection on a plane advertisement image to obtain significant areas which can reflect areas of interest of human eyes; generating an attention focus according to the significant areas in the plane advertisement image in combination with a WTA competitive mechanism and a no-return mechanism, wherein the obtained attention focus can reflect the degree of significance of each significant area of the plane advertisement; and judging whether pre-expression information areas of the advertisement attract the attention of advertisement audiences according to the significant areas of the plane advertisement, wherein the generated attention focus can further show the degree of attraction of each pre-expression information area in the advertisement image. The method sufficiently utilizes the simulation of a significance detection algorithm on the selective attention mechanism of a human eye visual system, provides an objective and scientific reference for advertisement design evaluation, and has extremely important application value in the advertisement marketing field.

The invention relates to a mixed digital image halftoning method based on a significance visual attention model, and belongs to the technical field of image processing. The method is characterized by adopting a bottom-up visual selective attention model, and carrying out calculation to obtain image intensity, color and direction feature patterns from an input image through Gaussian pyramids and a "center-surround" operator; carrying out normalization on the feature patterns to enable the feature patterns to be superposed into a total significance pattern, and extracting regions of interest (ROIs) of the image; carrying out image halftoning on the images in the ROIs through a model-based weighted least square halftone iteration method; carrying out halftone image conversion in non regions-of-interest (NROI) through a tone-based error diffusion method, and carrying out halftone parallel computing in the two regions of the image; and objectively evaluating digital image halftoning performance through a quality evaluation method based on the selective attention model, and analyzing algorithm complexity to obtain an optimum halftone image.

The invention discloses an image defogging and rain-removing algorithm based on a selective attention mechanism. The image defogging and rain-removing algorithm specifically comprises the following three processes of expanding a training sample, training a network on a training set, defogging and removing rain from a test image, and evaluating a restored image and a clear image. The method comprises s that: firstly, a defogging data set conducts cutting, intercepting, rotating and other operations on a used training set to expand an original data set; the pictures are input into a convolutional network based on a selective attention mechanism, and the network is trained by using a multi-loss function; and finally, the test pictures are input into the network to obtain defogging / rain-removing results. According to the invention, an end-to-end defogging / rain removing algorithm is realized, and rain-removing / defogging can be efficiently and accurately carried out on an image with rain / fog.

The invention relates to a method for detecting weak edges of images on the basis of discharge information of multilayer neuron groups. The method includes constructing the multilayer neuron groups with interconnected inhibitory synapses, inputting digital images into input-layer neuron groups and representing image pixels by the time-space information of first discharge of various neurons; describing space details of the images by time variances by the aid of visual receptive fields and discharge time sequences of the various neurons, selecting attentive mechanisms in the consideration of lateral inhibition so as to acquire visual attention data of information of the images; implementing space variable-resolution mechanisms on the basis of a combination of selective attention procedures by the aid of Log-Gabor multi-direction filter results, acquiring reconstructed information of the edges of the images and reinforcing the information of the edges of the images by the aid of output-layer neuron groups. The method has the advantages that a synapse interconnection characteristic of the neuron groups is taken into consideration; simple visual information procedures of cortices are reflected by the aid of multi-direction filter mechanisms; the weak edges of the images can be effectively detected by the aid of the multilayer neuron groups.

The invention relates to a vehicle logo positioning method based on active vision. The method comprises the steps of: (1) using the active vision to search rearview mirrors of a vehicle and carrying out inclination correction; (2) searching the positions of a license plate and vehicle lamps of the vehicle; (3) coarsely positioning a vehicle logo; and (4) adopting different methods to realize fine positioning of the vehicle logo according to different positions of the vehicle logo and different radiator grid textures. According to the invention, the concept of active vision is introduced, an active selectivity attention mechanism in the active vision is utilized to actively search the rearview mirrors and gradually search the regions of the vehicle lamps and the license plate, the inclination correction is carried out by the rearview mirrors, and a horizontal range and a vertical range are determined by the two regions; in the positioning process, a gradual positioning method from coarse positioning to fine positioning is used, so that the accuracy is improved; in addition, three different kinds of radiator grid textures are distinguished in vehicle logo positioning, and the applicability of the algorithm is improved.

The invention discloses a method for predicting visual attention area transfer in gray images, which comprises four steps of: determining leaders, searching followers, calculating significant values and ordering the significant values. The step of determining the leaders is to calculate side potentials of all pixels and determine the leaders in the pixels according to the acquired side potentials and a threshold value. The step of searching the followers is to determine the followers of each leader in all the pixels according to the connectivity and the similarity to form different areas, wherein each area may comprise more than one leader, but in the actual computer implementation process, the leader following other leaders is regarded as a follower. The step of calculating the significant values is to calculate the obtained areas respectively, and one area corresponds to one significant value. The step of ordering the significant values is to order all the areas according to the magnitude of the significant values and take the front three areas. The method successfully introduces the selective attention function of a human vision system into a computer vision system, and can simulate and predict attention transfer of a human eye view point among different areas.

The invention provides a method for analyzing spatial selective attention on the basis of grey theories, relates to the field of cognition research, and particularly relates to the field of selective attention. The method includes 1, establishing vision stimulation systems; 2, carrying out attention testing experiments on the systems and reading brain electric signals; 3, carrying out Fourier transformation on the read brain electric signals; 4, generating a group of reference signals (sinusoidal signals with the frequencies equal to stimulation flicker frequencies) and solving Fourier transformation on the reference signals; 5, carrying out grey correlation on the transformed brain electric signals and the transformed reference signals and solving grey correlation coefficients of the brain electric signals and the reference signals; 6, classifying correlation vectors by the aid of support vector machine classifiers to obtain spatial concentration locations of the attention. The correlation vectors comprise the grey correlation coefficients. The method has the advantages that as shown by repeated measurement, the detection speed can be increased by algorithms, the detection precision can be improved by the algorithms, and accordingly tested spatial selective attention modes and states can be possibly monitored in real time.

The invention relates to the technical field of video communication, particularly relating to a method for controlling a video code rate based on a vision significance model, wherein the method comprises the following steps: building a significance model for a video image according to the character of the video image and the vision sensing character of human eyes and acquiring a significance areaof a video scene according to the model; improving the existing H.264 frame code rate distribution mechanism according to the size of the significance area of the video scene, and calculating a target bit rate for distributing the current frame; in the macro block-level code rate distribution, building a macro block-level code rate distribution mechanism based on the significance model according to the texture information abundant degree, motion severe degree and significance of a currently coded macro block; and repeating the steps till the current frame is coded completely. According to themethod, the code rate distribution is performed according to a visual selective attention mechanism of the human eyes and the significance of the video scene, the visual coding character of the humaneyes is met, and the high quality of a subjective video image can be acquired under the condition of limited code rate.

The invention provides a visual cognition theory-based product form optimization design method, which comprises the following steps of: selecting a target sample picture and extracting design elements, acquiring eye movement indexes more remarkably related to the design elements by utilizing an eye movement tracking experiment, and establishing a design element and eye movement index relation model and relation model verification; According to the invention, the visual cognition theory is utilized to establish a product morphological design element and eye movement index relation model, so that direct data of visual perception selective attention can be obtained; the influence degree of different design elements on the visual response of the user is analyzed through the model; A product form optimization design method is formed, a product form optimization design is driven based on objective data of user cognition from the perspective of taking a user as the center, a designer can be helped to seek a reliable basis for optimizing the product form from the objective data of the user cognition, and therefore the scientificity and reasonability of the design are improved.

A method and apparatus for aiding the vision and motor function of an individual with autism and other complex developmental delays due to the inability to sustain convergence in addition to ocular motor deficiencies. The method allows improved direct midline gaze increasing selective attention with less averted gaze. The method is practiced by the steps of mounting one or more prisms with bases most often nasally of one or both lenses of a pair of glasses. An individual wearing these glasses even while sleeping will awaken to a more normalized and less confusing visual spatial world.

The embodiment of the invention provides a defective anchor detection method and device capable of selectively paying attention to clue guidance, and electronic equipment. The method comprises the steps of obtaining a to-be-detected live broadcast audio and video; generating a second coupling feature based on the to-be-tested live broadcast audio and video; calculating a Hamming distance between the second coupling feature and a pre-stored first coupling feature corresponding to each type of audio and video data, the second coupling feature being generated in the same manner as the first coupling feature; and judging the type of the audio-video data corresponding to the first coupling feature with the Hamming distance to the second coupling feature smaller than a preset threshold value asthe type of the live audio-video to be tested, and outputting the type of the live audio-video to be tested. Therefore, the type of the live audio and video can be reliably detected.

The invention relates to the field of motion vision, and particularly relates to a multi-selectivity attention evaluation and training method and system, and aims to solve the problems that a trainingresult cannot be quantified and precise training cannot be realized for different users. The evaluation method comprises the steps of displaying a plurality of target objects with different appearances on a display screen; recording an ID number of a selected target object in the plurality of target objects; shielding the appearance distinguishing features of the plurality of target objects to enable the appearances of the plurality of target objects to be the same; controlling the plurality of target objects to move on the screen according to the movement parameters, simultaneously acquiringan eye movement track and an electroencephalogram map when the testee observes the selected target object, and acquiring a judgment value of the testee for the position of the selected target objectat the end of movement; and generating an evaluation result according to the actual motion trail, the eye movement trail and the electroencephalogram map of the selected target object and the judgmentvalue of the testee for the position of the selected target object at the movement end moment. According to the invention, targeted attention training and improvement of the user are realized.

The invention, which relates to the field of computer vision technology, in particular provides a visual attention detection method and device based on lateral suppression, thereby improving the effectiveness and practicability of the computer visual attention mechanism. The visual attention detection method comprises: an image type of a target image is obtained; feedback information received by each activation function neuron is obtained; according to the feedback information, whether the corresponding activation function neuron needs to be turned off is determined; according to the determination result, the corresponding activation function neuron is turned off and then a response value of the activation function neuron is normalized to obtain a selective attention result graph; and theselective attention result graph is normalized to obtain a saliency target detection result graph. According to the invention, on the basis of the convolutional neural network, the information of thetarget object is detected by using feedforward and feedback information of the convolutional neural network. Besides, the visual attention detection device can be implemented to realize the method.

The invention provides a crowdsourcing task personalized recommendation method and system based on deep learning, and relates to the technical field of task recommendation. According to the method, key information in data can be effectively obtained by utilizing the advantages of selective attention of an Attention mechanism on a hidden layer state, accuracy of time sequence prediction of an LSTM network and the like, so that interest changes and core concerns of crowdsourcing workers on task selection are obtained, dynamic factor capture of whole behavior sequence changes of a user is realized, and the user experience is improved. And by introducing a Word2Vec word vector model, the similarity correlation degree between the crowdsourcing worker and the task is calculated, and the crowdsourcing task is accurately recommended to the crowdsourcing worker. Meanwhile, crowdsourcing data are segmented into structured data and unstructured data, and explicit features and implicit features of historical behavior information are obtained from the structured data and the unstructured data, so that more valuable potential feature information is mined, and recommendation accuracy is further improved.

The invention discloses an alertness testing method based on an attention system theory. The method comprises the steps that the alertness level of a testee is collected; continuous alertness, staged alertness and selective attention indexes are obtained, and the corresponding degree of the testee is judged; average reaction time data is obtained; a standard deviation during reaction is calculated; a variable coefficient and a correct rate are calculated; the alertness level of the testee is determined; comprehensively the alertness overall state of the testee, and the like are reflected. According to the method, the attention system theory is creatively applied to the alertness test, the alertness test can be carried out from the dimensions of continuous alertness, staged alertness and selective attention, and the alertness level of the testee can be reflected more visually and comprehensively. The alertness level of the testee is determined by adopting two specific indexes, namely the variable coefficient and the accuracy rate when the testee reacts, and the result is more scientific and accurate. And in combination with a Karolinska Sleepiness Scale, subjective test and objective test of the alertness are organically combined together, so that the test of the alertness is more comprehensive.

The invention discloses a method for predicting visual attention area transfer in gray images, which comprises four steps of: determining leaders, searching followers, calculating significant values and ordering the significant values. The step of determining the leaders is to calculate side potentials of all pixels and determine the leaders in the pixels according to the acquired side potentialsand a threshold value. The step of searching the followers is to determine the followers of each leader in all the pixels according to the connectivity and the similarity to form different areas, wherein each area may comprise more than one leader, but in the actual computer implementation process, the leader following other leaders is regarded as a follower. The step of calculating the significant values is to calculate the obtained areas respectively, and one area corresponds to one significant value. The step of ordering the significant values is to order all the areas according to the magnitude of the significant values and take the front three areas. The method successfully introduces the selective attention function of a human vision system into a computer vision system, and can simulate and predict attention transfer of a human eye view point among different areas.

The invention provides a multi-channel speech enhancement method based on semantic prior selective attention. The method comprises the following steps: picking speech signals from any directions in a reverberant environment by virtue of a multi-microphone array, collecting the multiple paths of speech signals and pre-processing the speech signals; detecting special activation words in the pre-processed speech signals by virtue of an activation word speech recognition model; processing signals which are not cut and include activation word segments, so as to obtain a complete activation word segment; analyzing the activation word segment by virtue of a multi-channel phase difference sound localization method based on reverberation robust, so as to obtain an acoustic wave reaching direction of a target sound source; and enhancing speech in the direction and inhibiting noise from other directions and room reverberation in a remote speak scene, so that enhanced speech in the target direction is obtained. The method provided by the invention is applicable to such occasions as intelligent household electrical appliance, smart home, vehicle-mounted and wearable devices and the like that remote speak type speech input and interaction are required, and the method is especially applicable to complex acoustic noise and interference environment occasions.

The invention relates to a hybrid digital image halftone method of saliency visual attention model, and belongs to the technical field of image processing. The invention applies a bottom-up visual selective attention model to calculate feature maps of image intensity, color and direction from the input image through Gaussian pyramid and "center-periphery" operator. Each feature map is normalized and superimposed into a total saliency map, and the region of interest ROIs of the image are extracted. A model-based weighted least squares halftoning iteration method is used to halftonize the image in ROIs. The hue-based error diffusion method is used to convert halftone images in non-interest areas, and halftones are calculated in parallel in two areas of the image. The quality evaluation method based on the selective attention model is applied to objectively evaluate the performance of digital image halftones, and the complexity of the algorithm is analyzed to obtain the optimal halftone image.

A construction method for a deep long short-term memory recurrent neural network acoustic model based on a selective attention principle. Change of an instant function of neurons of an auditory cortex is represented by adding an attention gate (103) unit in the deep long short-term memory recurrent neural network acoustic model, and the attention gate (103) unit is different from other gate units in that: the other gate units correspond to a time sequence on a one-to-one basis, but the attention gate (103) unit shows a short-term plasticity effect, thereby having intervals on the time sequence; extraction of robust features about Cross-talk noise and construction of a robust acoustic model are realized via the recurrent neural network acoustic model obtained by training a large amount of voice data containing the Cross-talk noise, and the purpose of increasing the robustness about the acoustic model can be achieved by restraining the influence of a non-target stream against the extraction of the features; the method can be extensively applied to the field of a plurality of machine learning related to speaker recognition and keyword recognition in voice recognition, human-machine interaction and the like.

The Randomized Intermittent Dichotomous Stimulation (RIDS) paradigm disclosed provides an alternate stimulus for the patient to attend to, besides the typical SCS signal, thus providing a somatotopically-matched non-noxious stimulus to replace the cognitive attention to the nociceptive stimulus. The randomly intermittent stimulus is paired with an implicit virtual task to reinforce selective attention to the non-noxious stimulus.

A monitoring device is provided, which includes an inputter configured to receive an input of a plurality of images captured at separate positions and a plurality of sound sources heard at separate positions, a saliency map generator configured to generate a plurality of mono saliency maps for the plurality of images and to generate a dynamic saliency map using the plurality of mono saliency maps generated, a position determinator configured to determine the positions of the sound sources through analysis of the plurality of sound sources, a scan path recognizer configured to generate scan paths of the plurality of images based on the generated dynamic saliency map and the determined positions of the sound sources, and an outputter configured to output the generated scan paths.

The invention discloses an attention training system and method based on target detection and SSVEP. The attention training system comprises a camera, an electroencephalogram cap, a computer and a mechanical arm. The camera is used for collecting a video stream sent to the target detection module for target detection; the electroencephalogram cap is used for collecting and transmitting SSVEP electroencephalogram signals; the computer comprises an attention training interface, a target detection unit and an SSVEP decoding unit, the attention training interface draws a target frame and generates a flicker block of a corresponding article in real time, the SSVEP decoding unit decodes the electroencephalogram signal through preprocessing and an FBCCA algorithm to obtain the flicker block watched by a user, converts a decoding result into a control instruction and sends the control instruction to the mechanical arm, and the mechanical arm is used for controlling the target frame to be detected. And the mechanical arm grabs corresponding articles according to the instruction. By means of natural man-machine interaction, attention training can be carried out, the selective attention level and the continuous attention level are effectively improved, meanwhile, reference is provided for attention training methods and means, and good application prospects of the brain-computer interface technology in the fields of child development and the like are shown.