1175 results about "Feature mapping" patented technology

The present invention relates to a convolutional-neural-network-based classifier, a classifying method by using a convolutional-neural-network-based classifier and a method for training the convolutional-neural-network-based classifier. The convolutional-neural-network-based classifier comprises: a plurality of feature map layers, at least one feature map in at least one of the plurality of feature map layers being divided into a plurality of regions; and a plurality of convolutional templates corresponding to the plurality of regions respectively, each of the convolutional templates being used for obtaining a response value of a neuron in the corresponding region.

Methods of detecting an object in an image using a convolutional neural network based architecture that processes multiple feature maps of differing scales from differing convolution layers within a convolutional network to create a regional-proposal bounding box. The bounding box is projected back to the feature maps of the individual convolution layers to obtain a set of regions of interest. These regions of interest are then processed to ultimately create a confidence score representing the confidence that the object detected in the bounding box is the desired object. These processes allow the method to utilize deep features encoded in both the global and the local representation for object regions, allowing the method to robustly deal with challenges in the problem of robust object detection. Software for executing the disclosed methods within an object-detection system is also disclosed.

A local-neighborhood Laplacian Eigenmap (LNLE) algorithm is provided for methods and systems for semi-supervised learning on manifolds of data points in a high-dimensional space. In one embodiment, an LNLE based method includes building an adjacency graph over a dataset of labelled and unlabelled points. The adjacency graph is then used for finding a set of local neighbors with respect to an unlabelled data point to be classified. An eigen decomposition of the local subgraph provides a smooth function over the subgraph. The smooth function can be evaluated and based on the function evaluation the unclassified data point can be labelled. In one embodiment, a transductive inference (TI) algorithmic approach is provided. In another embodiment, a semi-supervised inductive inference (SSII) algorithmic approach is provided for classification of subsequent data points. A confidence determination can be provided based on a number of labeled data points within the local neighborhood. Experimental results comparing LNLE and simple LE approaches are presented.

A method for unsupervised anomaly detection, which are algorithms that are designed to process unlabeled data. Data elements are mapped to a feature space which is typically a vector space . Anomalies are detected by determining which points lies in sparse regions of the feature space. Two feature maps are used for mapping data elements to a feature apace. A first map is a data-dependent normalization feature map which we apply to network connections. A second feature map is a spectrum kernel which we apply to system call traces.

The invention discloses an Android malicious application detection method and system based on multi-feature fusion. The method comprises the following steps that: carrying out decompilation on an Android application sample to obtain a decompilation file; extracting static features from the decompilation file; operating the Android application sample in an Android simulator to extract dynamic features; carrying out feature mapping on the static features and the dynamic features by the text Hash mapping part of a locality sensitive Hash algorithm, mapping to a low-dimensional feature space to obtain a fused feature vector; and on the basis of the fused feature vector, utilizing a machine learning classification algorithm to train to obtain a classifier, and utilizing the classifier to carry out classification detection. By use of the method, the high-dimensional feature analysis problem of a malicious code rare sample family can be solved, and detection accuracy is improved.

The invention discloses the hardware architecture of a binary weight convolution neural network accelerator and a calculation process thereof. The hardware architecture comprises three double-ended on-chip static random access memories which are used for buffering the binary weight of input neurons and a convolution layer, four convolution processing units capable of controlling calculation parts to complete major convolution calculation operation according to the calculation process, a feature map accumulation unit and a convolutional accumulation array. The feature map accumulation unit and the convolutional accumulation array are used for further processing the operation result of the convolution processing units to acquire a final correct output neuron value. The entire design exchanges data with an off-chip memory via a dynamic random access memory interface. In addition to the hardware architecture, the invention further provides the detailed calculation process which optimizes the hardware architecture and uses four lines of input feature map as a complete calculation unit. According to the invention, input data are reused to the greatest extent; the access of the off-chip memory is eliminated as much as possible; the power consumption of the deep binary convolution neural network calculation can be effectively reduced; a deep network is supported; and the scheme provided by the invention is a reasonable scheme which can be applied to an embedded system of visual application.

The invention, which belongs to the field of intelligent traffic sign recognition technology, relates to a traffic sign recognition method based on an asymmetric convolution neural network. With the method, problems of slow recognition speed and poor robustness during traffic sign recognition can be solved. According to the method, two convolution neural networks with different structures are used for carrying out feature mapping and extraction concurrently; the features are combined; and a full connection layer and a classifier are used for completing the whole classification process. The two convolution neural networks with different structures employ a random pooling operation and a maxout unit respectively, thereby guaranteeing diversity of the image features, improving the recognition precision, and accelerating the network operation speed. According to the invention, the structure of the traditional convolution neural network is modified and the two convolution neural networks with different structures are used for replacing the traditional convolution neural network structure. Therefore, the image feature diversity is guaranteed; the recognition precision is improved; and the network operation speed is accelerated.

An artificial neural network system implemented on a computer for cell segmentation and classification of biological images. It includes a deep convolutional neural network as a feature extraction network, a first branch network connected to the feature extraction network to perform cell segmentation, and a second branch network connected to the feature extraction network to perform cell classification using the cell segmentation map generated by the first branch network. The feature extraction network is a modified VGG network where each convolutional layer uses multiple kernels of different sizes. The second branch network takes feature maps from two levels of the feature extraction network, and has multiple fully connected layers to independently process multiple cropped patches of the feature maps, the cropped patches being located at a centered and multiple shifted positions relative to the cell being classified; a voting method is used to determine the final cell classification.

A method constructs a classifier from training data and detects moving objects in test data using the trained classifier. High-level features are generated from low-level features extracted from training data. The high level features are positive definite matrices on an analytical manifold. A subset of the high-level features is selected, and an intrinsic mean matrix is determined. Each high-level feature is mapped to a feature vector on a tangent space of the analytical manifold using the intrinsic mean matrix. An untrained classifier is trained with the feature vectors to obtain a trained classifier. Test high-level features are similarly generated from test low-level features. The test high-level features are classified using the trained classifier to detect moving objects in the test data.

Provided are an object recognition method and apparatus which determine an object of interest included in a recognition target image using a trained machine learning model and determine an area in which the object of interest is located in the recognition target image. The object recognition method based on weakly supervised learning, performed by an object recognition apparatus, includes extracting a plurality of feature maps from a training target image given classification results of objects of interest, generating an activation map for each of the objects of interest by accumulating the feature maps, calculating a representative value of each of the objects of interest by aggregating activation values included in a corresponding activation map, determining an error by comparing classification results determined using the representative value of each of the objects of interest with the given classification results and updating a CNN-based object recognition model by back-propagating the error.

A method for unsupervised anomaly detection, which are algorithms that are designed to process unlabeled data. Data elements are mapped to a feature space which is typically a vector space d. Anomalies are detected by determining which points lies in sparse regions of the feature space. Two feature maps are used for mapping data elements to a feature apace. A first map is a data-dependent normalization feature map which we apply to network connections. A second feature map is a spectrum kernel which we apply to system call traces.

The invention belongs to the technical field of computer applying and the subfield of the machine learning theories and application and focuses on the traffic sign recognizing problem in the intelligent traffic technology. A traffic sign recognizing method based on multi-resolution convolution neural networks is characterized by solving the problem that the speed is low when the convolution neural network is used for recognizing traffic signs, two-dimensional images with different resolutions are used as input, the two convolution neural networks with the same structure are operated in parallel to carry out feature mapping and extracting, and accurate classifying and recognizing are carried out based on a weight threshold value trained by the networks. The two CNNs with different resolution branches are used for replacing a basic CNN structure, the overall and outline features can be mapped through the high-resolution image input, the local and detailed features can be mapped through the low-resolution images, the recognizing resolution is guaranteed, and the model training speed is increased.

A trustworthy inverted index system processes records to identify features for indexing, generates posting lists corresponding to features in a dictionary, maintains in a storage cache a tail of at least one of the posting lists to minimize random I / Os to the index, determines a desired number of the posting lists based on a desired level of insertion performance, a query performance, or a size of the storage cache, and reads a posting list corresponding to a search feature in a query to identify records that comprise the search feature. The system maps the features in the dictionary to the desired number of posting lists. The system uses a jump pointer to point from one entry to the next in the posting lists based on increasing values of entries in the posting lists.

Systems and techniques are provided for content filtering with convolutional neural networks. A spectrogram generated from audio data may be received. A convolution may be applied to the spectrogram to generate a feature map. Values for a hidden layer of a neural network may be determined based on the feature map. A label for the audio data may be determined based on the determined values for the hidden layer of the neural network. The hidden layer may include a vector including the values for the hidden layer. The vector may be stored as a vector representation of the audio data.

The invention discloses a target detection method based on a dense connection convolutional neural network, and the method employs a network structure in which a plurality of dense connection blocks and conversion layers are alternately connected to replace a conventional overall structure in order to reduce the parameter quantity and improve the feature reuse effect, and achieves the feature extraction, and can achieve the discrimination feature mapping in an image. The global attention module fuses the feature maps of the four different receptive fields to solve the problem that the sizes ofsingle-layer receptive fields are the same in the past; And meanwhile, the last three convolutional layers of each branch enable the feature map of the bottom layer to have enough excellent feature expression on the premise of ensuring the resolution. The image target detection model provided by the invention can effectively extract the features of the image and extract the feature map which hasdifferent size receptive fields and is fused with multi-level information; Meanwhile, the detection effect of the small object is improved through combination of the semantic information and the spatial information; Meanwhile, the whole network can achieve end-to-end training, the real-time detection speed is kept, and meanwhile the target detection effect is improved.

A data object can be received at a first time at a first inbound interface of a first software component from a second software component. The first inbound interface can be configured to map a first feature of the data object to at least one first internal data structure of the first software component but not configured to map a second data feature of the data object to any corresponding internal data structure of the first software component. Characteristic information of the second data feature can be stored in a generic data container added to the internal data structure. The characteristic information can include a value of the second data feature and metadata describing an identifier of the second data feature. A reconstructed data object can be formed at a second time subsequent to the first time. Related methods, systems, and articles of manufacture are disclosed.

The invention provides a magnetic resonance image feature extraction and classification method based on deep learning, comprising: S1, taking a magnetic resonance image, and performing pretreatment operation and feature mapping operation on the magnetic resonance image; S2, constructing a multilayer convolutional neural network including an input layer, a plurality of convolutional layers, at least one pooling layer / lower sampling layer and a fully connected layer, wherein the convolutional layers and the pooling layer / lower sampling layer are successively alternatively arranged between the input layer and the fully connected layer, and the convolutional layers are one more than the pooling layer / lower sampling layer; S3, employing the multilayer convolutional neural network constructed in Step 2 to extract features of the magnetic resonance image; and S4, inputting feature vectors outputted in Step 3 into a Softmax classifier, and determining the disease attribute of the magnetic resonance image. The magnetic resonance image feature extraction and classification method can automatically obtain highly distinguishable features / feature combinations based on the nonlinear mapping of the multilayer convolutional neural network, and continuously optimize a network structure to obtain better classification effects.

The invention discloses a small target detecting method based on R-FCN, wherein the method relates to the field of image processing. The method comprises the steps of introducing a to-be-detected image into a convolutional network, successively performing characteristic extraction on a to-be-detected image through M network layers according to a sequence from a topmost layer of M network layers to a downmost layer and according to a sequence from the downmost layer of the M network layers to the topmost layer, generating characteristic mapping graphs with different scales, selecting an N characteristic mapping graphs into an RPN for performing foreground-and-background classification, determining the coordinate of a foreground area, processing a characteristic mapping block which corresponds with the coordinate of the foreground area for obtaining a characteristic vector; inputting each characteristic vector into a classifier for performing secondary classification, detecting whether the kind to which the characteristic vector is affiliated corresponds with a to-be-detected small target and outputting a detecting result. According to the small target detecting method, a manner of combining a top-down characteristic pyramid and a down-top characteristic pyramid is utilized for performing small target detection on the characteristic mapping graphs with different scales, thereby reducing report omission for the small target and improving detecting precision.

An ultrasound machine is disclosed that displays a color representation of moving structure, such as a cardiac wall tissue, within a region of interest on a monitor. The color representation is generated by displaying at least one color characteristic corresponding to a movement parameter of the structure, such as velocity or strain rate. The movement parameter is mapped to the color characteristic by apparatus comprising a front-end that generates received signals in response to ultrasound waves. A Doppler processor generates a set of parameter signals representing values of the movement parameter within the structure. A control processor adaptively generates a mapping function based on the distribution of the parameter signals to map the parameter signals to a set of color characteristic signals. A display processor applies the mapped values of the color characteristic legend to the values of the movement parameter representing the moving structure, to display a color representation on the monitor in response to the mapping function.

A method of augmenting training data includes converting a feature sequence of a source speaker determined from a plurality of utterances within a transcript to a feature sequence of a target speaker under the same transcript, training a speaker-dependent acoustic model for the target speaker for corresponding speaker-specific acoustic characteristics, estimating a mapping function between the feature sequence of the source speaker and the speaker-dependent acoustic model of the target speaker, and mapping each utterance from each speaker in a training set using the mapping function to multiple selected target speakers in the training set.

An iterative method and associated algorithm for performing image registration to map features in a first image to corresponding features in a second image in accordance with a transformation model. Upon convergence of a parameter vector associated with the model, a current bootstrap region includes and exceeds an initial bootstrap region that is initially established. During the iterations, the parameter vector is estimated by minimizing an objective function with respect to the parameter vector. Each iteration generates a next bootstrap region that minimally includes the current bootstrap region and may exceed the current bootstrap region. The model may change in successive iterations.

The invention discloses a multiple attention and multiple scale-based image describing method. The method comprises the steps of selecting an image detecting model for extracting image features, dividing into a network training set, a verification set and a test set, extracting the image features, constructing an attention recurrent neural network model, training the attention recurrent neural network model and carrying out image description. According to the method disclosed by the invention, an image description generating network model formed by original image feature extracting, multi-attention multi-scale feature mapping, recurrent neural network residual connecting and recurrent neural network language decoding is constructed, so that the quality of the image description is improvedand the detail of the image description is enriched. By means of the method disclosed by the invention, a high quality image can be generated by adopting the neural network model to carry out description under the circumstance of only having an image.

A content server uses a form of artificial intelligence such as machine learning to identify audio content with musicological characteristics. The content server obtains an indication of a music item presented by a client device and obtains reference music features describing musicological characteristics of the music item. The content server identifies candidate audio content associated with candidate music features. The candidate music features are determined by analyzing acoustic features of the candidate audio content and mapping the acoustic features to music features according to a music feature model. Acoustic features quantify low-level properties of the candidate audio content. One of the candidate audio content items is selected according to comparisons between the candidate music features of the candidate audio advertisements and the reference music features of the music item. The selected audio content is provided the client device for presentation.

The invention discloses a deep learning model-based image Chinese description method and belongs to the field of computer vision and natural language processing. The method comprises the steps of preparing an ImageNet image data set and an AI Challenger image Chinese description data set; pre-training the ImageNet image data set by utilizing a DCNN to obtain a pre-trained DCNN model; performing image feature extraction and image feature mapping on the AI Challenger image Chinese description data set, and transmitting image features to a GRU threshold recursive network recurrent neural network;performing word coding matrix construction on an AI Challenger image mark set in the AI Challenger image Chinese description data set; extracting word embedding features by utilizing an NNLM, and finishing text feature mapping; taking the GRU threshold recursive network recurrent neural network as a language generation model, and finishing image description model building; and generating a Chinese description statement. According to the method, the blank of image Chinese description is filled up; a function of automatically generating the image Chinese description is realized; the accuracy ofdescription contents is well improved; and a foundation is laid for development of Chinese NLP and computer vision.

Some embodiments efficiently locate and map item locations in a distribution site using robots and a set of markers that the robots can scan without cessation of movement. Some embodiments optimally position a robot for item retrieval or placement. The optimal position is determined by affine transform computation or feature mapping. The robot first aligns itself according to the expected item position as indicated by one or more of the markers. The alignment is determined based on size and orientation of the markers in images obtained using the robot's camera. The robot then aligns itself according to the actual item position. Here, the repositioning is determined based on size and orientation of the actual item in images obtained using the robot's camera. Using the robot cameras and feature mapping, robots can traverse the shelves to identify and map item location therein.

The invention is a method and an apparatus for generating a displacement map of a first input dataset and a second input dataset of an input dataset pair. The apparatus comprisesa neural network based feature extractor (25) for processing the first input dataset and the second input dataset so as to generate a feature map hierarchy (50) comprising a base pair of feature maps (20e, 30e) and a refinement pair of feature maps (20b, 30b, 20c, 30c, 20d, 30d),a displacement unit comprising a first comparator unit for obtaining an initial displacement map using the base pair of feature maps (20e, 30e); anda displacement refinement unit for obtain an updated displacement map for the refinement pair of feature maps (20d, 30d).

The present invention relates to an image fusion method, especially to an image fusion method based on depth learning. The method comprises: employing a convolution layer to construct basic units based on an automatic encoder; stacking up a plurality of basic units for training to obtain a depth stack neural network, and employing an end-to-end mode to regulate the stack network; employing the stack network to decompose input images, obtaining high-frequency and low-frequency feature mapping pictures of each input image, and employing local variance maximum and region matching degree to merge the high-frequency and low-frequency feature mapping pictures; and putting a high-frequency fusion feature mapping picture and a low-frequency fusion feature mapping picture back to the last layer of the network, and obtaining a final fusion image. The image fusion method based on depth learning can perform adaptive decomposition and reconstruction of images, one high-frequency feature mapping picture and one low-frequency mapping picture are only needed when fusion, the number of the types of filters do not need artificial definition, the number of the layers of decomposition and the number of filtering directions of the images do not need selection, and the dependence of the fusion algorithm on the prior knowledge can be greatly improved.

A system and method for detecting fraudulent identification tags, such as Universal Product Codes (UPC) applied to goods to be purchased is disclosed. Images of the goods to be purchased and corresponding feature models are stored in a database. When a customer desires to purchase an item containing a UPC, a scanned image of the item about to be purchased may be acquired and a feature model of the scanned image may be created. The system may retrieve from the database the image and feature model previously stored for the item associated with the just-scanned UPC. A variety of image processing techniques may be used to compare the scanned and database images and / or the scanned and database feature models. In one embodiment, these image processing techniques may include determining a geometric transformation that maps the features of the scanned image onto the features of a database model.