Patents
Literature
Hiro is an intelligent assistant for R&D personnel, combined with Patent DNA, to facilitate innovative research.
Hiro

1025 results about "Test phase" patented technology

Test Phase. Description. The primary purpose of the Test Phase is to determine whether the automated system/application software or other IT solution developed or acquired and preliminarily tested during the Development Phase is ready for implementation.

Method of managing fails in a non-volatile memory device and relative memory device

A method of managing fails in a non-volatile memory device including an array of cells grouped in blocks of data storage cells includes defining in the array a first subset of user addressable blocks of cells, and a second subset of redundancy blocks of cells. Each block including at least one failed cell in the first subset is located during a test on wafer of the non-volatile memory device. Each block is marked as bad, and a bad block address table of respective codes is stored in a non-volatile memory buffer. At power-on, the bad block address table is copied from the non-volatile memory buffer to the random access memory. A block of memory cells of the first subset is verified as bad by looking up the bad block address table, and if a block is bad, then remapping access to a corresponding block of redundancy cells. A third subset of non-user addressable blocks of cells is defined in the array for storing the bad block address table of respective codes in an addressable page of cells of a block of the third subset. Each page of the third subset is associated to a corresponding redundancy block. If during the working life of the memory device a block of cells previously judged good in a test phase becomes failed, each block is marked as bad and the stored table in the random access memory is updated.
Owner:MICRON TECH INC

Biopotential Waveform Data Fusion Analysis and Classification Method

Biopotential waveforms such as ERPs, EEGs, ECGs, or EMGs are classified accurately by dynamically fusing classification information from multiple electrodes, tests, or other data sources. These different data sources or “channels” are ranked at different time instants according to their respective univariate classification accuracies. Channel rankings are determined during training phase in which classification accuracy of each channel at each time-instant is determined. Classifiers are simple univariate classifiers which only require univariate parameter estimation. Using classification information, a rule is formulated to dynamically select different channels at different time-instants during the testing phase. Independent decisions of selected channels at different time instants are fused into a decision fusion vector. The resulting decision fusion vector is optimally classified using a discrete Bayes classifier. Finally, the dynamic decision fusion system provides high classification accuracies, is quite flexible in operation, and overcomes major limitations of classifiers applied currently in biopotential waveform studies and clinical applications.
Owner:NEURONETRIX SOLUTIONS

Apparatus and method for developing, testing and monitoring secure software

A method of analyzing program instructions for security vulnerabilities includes applying a static analysis to program instructions during a development phase of the program instructions to identify security vulnerabilities. The security vulnerabilities are used to apply a security test to the program instructions during a testing phase of the program instructions. The security vulnerabilities are analyzed to develop security monitoring criteria to apply to the program instructions during a deployment phase of the program instructions.
Owner:FORTIFY SOFTWARE

Human face super-resolution algorithm based on regional depth convolution neural network

The invention discloses a human face super-resolution algorithm based on regional depth convolution neural network. The algorithm comprises the following steps: a training stage: S1) dividing the mutually overlapping image blocks in the pixel domain of an inputted human face image with low resolution to obtain a plurality of local regions; S2) extracting the local regions for local characteristics; S3) performing non-linear change to the local characteristics to obtain non-linear characteristics; S4) processing the non-linear characteristics to obtain reconstructed image blocks with high resolution; S5) splicing the image blocks with high resolution; adjusting the multi-layer convolution layers and correcting the parameters of the linear unit layer; and a testing stage: S6) inputting the tested human face image with low resolution; processing through the super-resolution network to obtain the human face image with high resolution. The regional convolution neural network proposed by the invention improves the quality of subjective and objective reconstruction of reconstructing high resolution images.
Owner:WUHAN INSTITUTE OF TECHNOLOGY

Test and burn-in apparatus, in-line system using the test and burn-in apparatus, and test method using the in-line system

A test and burn-in apparatus for semiconductor chip package devices, an in-line system which includes the test and burn-in apparatus, and a test method which employs the in-line system are provided. A test and burn-in apparatus for testing semiconductor devices allows various testing processes, including a burn-in process, to be performed at the same testing stage. The apparatus employs test trays which contain the semiconductor devices. These test trays are used throughout the in-line system so that an entire back-end process can be performed without the need for loading / unloading the semiconductor devices into and from device trays between the various tests. The test and burn-in apparatus according to this invention can therefore occupy less space than the prior art testing apparatuses. The in-line system includes multiple test and burn-in apparatuses as well as a single sorting unit which performs a composite sorting operation after all the testing processes have been performed. Furthermore, the method for testing the semiconductor devices in the in-line system includes testing the semiconductor devices in the test trays using the test and burn-in apparatus, generating a test tray map corresponding to results of the test, transferring the test trays to a different testing apparatus for a second testing and test tray map generation process, and finally sorting the semiconductor devices in the sorting unit after all testing processes have been performed based on a final sorting map created by combining the test tray maps of each of the tests. The benefits of this invention are reduced time and space requirements because neither transferring the devices to device trays between tests nor performing multiple sorting steps are required.
Owner:SAMSUNG ELECTRONICS CO LTD

Method and system for detecting voice spoofing attack of speakers on basis of deep learning

The invention discloses a method and a system for detecting voice spoofing attack of speakers on the basis of deep learning. The method includes constructing audio-frequency training sets, initializing deep feed-forward neural networks and deep recurrent neural networks and respectively training the deep feed-forward neural networks and the deep recurrent neural networks by the aid of multi-frame feature vectors and single-frame vector sequences of the training sets; respectively leading frame level and sequence level feature vectors of to-be-tested audio frequencies into two trained linear differential analysis models in test phases, weighting two obtained result grades to obtain scores and comparing the scores to predefined threshold values so as to discriminate voice spoofing. The method and the system have the advantages that local features can be captured, and global information can be grasped; the linear differential analysis models are used as classifiers in identification and verification phases, the voice spoofing attack can be judged by means of grade fusion, and accordingly the voice spoofing detection accuracy can be greatly improved.
Owner:AISPEECH CO LTD

System and method for performing rapid application life cycle quality assurance

Performance of quality assurance for a computer application development proceeds in a series of phases. Each of the phases in the QA life cycle corresponds to one or more phases in a computer application development life cycle. For example, in one embodiment, the quality assurance life cycle includes four phases—a test strategy phase, a test planning phase, a test execution phase and a rollout phase. Each phase of the QA life cycle has entrance and exit criteria. The phase can be entered when the entrance criteria have been met. When the exit criteria are met, the phase causes a gate to be closed to allow the software development life cycle to proceed to the next phase. In one embodiment, the software life cycle includes six phases—a requirements phase, a design phase, a code / unit test phase, a system test phase, a pre-production test phase and a production test phase.
Owner:BELLSOUTH INTPROP COR

Human skeleton behavior recognition method and device based on deep reinforcement learning

The invention discloses a human skeleton behavior recognition method and device based on deep reinforcement learning. The method comprises: uniform sampling is carried out on each video segment in a training set to obtain a video with a fixed frame number, thereby training a graphic convolutional neural network; after parameter fixation of the graphic convolutional neural network, an extraction frame network is trained by using the graphic convolutional neural network to obtain a representative frame meeting a preset condition; the graphic convolutional neural network is updated by using the representative frame meeting the preset condition; a target video is obtained and uniform sampling is carried out on the target video, so that a frame obtained by sampling is sent to the extraction frame network to obtain a key frame; and the key frame is sent to the updated graphic convolutional neural network to obtain a final type of the behavior. Therefore, the discriminability of the selectedframe is enhanced; redundant information is removed; the recognition performance is improved; and the calculation amount at the test phase is reduced. Besides, with full utilization of the topologicalrelationship of the human skeletons, the performance of the behavior recognition is improved.
Owner:TSINGHUA UNIV

Data-driven and task-driven image classification method

The invention discloses a data-driven and task-driven image classification method. The data-driven and task-driven classification method comprises the steps that a convolutional neural network structure is designed according to the scale of data sets and image content; a convolutional neural network model is trained through the given classified data sets; feature expression is extracted from training set images through a trained convolution neural network; images to be tested are input into the trained convolutional neural network and are classified. The data-driven and task-driven image classification method is based on nonlinear convolution feature learning, and the model can be adapted to the data sets through a date driving mode, so that the specific data set can be better described; errors of K-nearest neighbors can be directly optimized through a task-driving mode, and therefore a better performance can be obtained with respect to a K-nearest neighbor task; efficient training can be conducted through a GPU in the training stage, and efficient K-nearest neighbor image classification can be achieved just through a CPU in the testing stage; in this way, the data-driven and task-driven image classification method is quite suitable for a large-scale image classification task, a retrieval task and the like.
Owner:INST OF AUTOMATION CHINESE ACAD OF SCI

Method and system for age estimation based on relative ages of pairwise facial images of people

The present invention is a system and method for estimating the age of people based on their facial images. It addresses the difficulty of annotating the age of a person from facial image by utilizing relative age (such as older than, or younger than) and face-based class similarity (gender, ethnicity or appearance-based cluster) of sampled pair-wise facial images. It involves a unique method for the pair-wise face training and a learning machine (or multiple learning machines) which output the relative age along with the face-based class similarity, of the pairwise facial images. At the testing stage, the given input face image is paired with some number of reference images to be fed to the trained machines. The age of the input face is determined by comparing the estimated relative ages of the pairwise facial images to the ages of reference face images. Because age comparison is more meaningful when the pair belongs to the same demographics category (such as gender and ethnicity) or when the pair has similar appearance, the estimated relative ages are weighted according to the face-based class similarity score between the reference face and the input face.
Owner:VIDEOMINING CORP

Dynamic gesture recognition method based on hybrid deep learning model

ActiveCN106991372AAchieving an efficient space-time representationEasy to identifyCharacter and pattern recognitionFrame basedModel parameters
The invention discloses a dynamic gesture recognition method based on a hybrid deep learning model. The dynamic gesture recognition method includes a training phase and a test phase. The training phase includes first, training a CNN based on an image set constituting a gesture video and then extracting spatial features of each frame of the dynamic gesture video sequence frame by frame using the trained CNN; for each gesture video sequence to be recognized, organizing the frame-level features learned by the CNN into a matrix in chronological order; inputting the matrix to an MVRBM to learn gesture action spatiotemporal features that fuse spatiotemporal attributes; and introducing a discriminative NN; and taking the MVRBM as a pre-training process of NN model parameters and network weights and bias that are learned by the MVRBM as initial values of the weights and bias of the NN, and fine-tuning the weights and bias of the NN by a back propagation algorithm. The test phase includes extracting and splicing features of each frame of the dynamic gesture video sequence frame by frame based on CNN, and inputting the features into the trained NN for gesture recognition. The effective spatiotemporal representation of the 3D dynamic gesture video sequence is realized by adopting the technical scheme of the invention.
Owner:BEIJING UNIV OF TECH

Small object semantic segmentation method combined with object detection

The invention discloses a small target semantic segmentation method combined with target detection. Attention semantic segmentation network, training the network to get the whole semantic segmentationmodel; making small target detection dataset and small target semantic segmentation dataset; training the small target detection network based on YOLOv2 through the small target detection data set; asmall target semantic segmentation network is designed and trained by using the small target semantic segmentation data set to obtain the small target semantic segmentation model. In the testing phase, the test image is used as the input of the whole semantic segmentation model and the small target detection network, and the segmentation result and the small target boundary box of the whole imageare obtained, which is modified by the small target semantic segmentation model. The invention can greatly reduce the segmentation difficulty of the small target, thereby effectively improving the segmentation performance of the small target.
Owner:NANJING NORMAL UNIVERSITY

Development of parallel/distributed applications

One embodiment of the present invention provides a method for supporting the development of a parallel / distributed application, wherein the development process comprises a design phase, an implementation phase and a test phase. A script language can be provided in the design phase for representing elements of a connectivity graph and the connectivity between them. In the implementation phase, modules can be provided for implementing functionality of the application, executors can be provided for defining a type of execution for the modules, and process-instances can be provided for distributing the application over several computing devices. In the test phase, abstraction levels can be provided for monitoring and testing the application.
Owner:HONDA RES INST EUROPE

Depth study based method for detecting salient regions in natural image

The invention relates to a depth study method for detecting salient regions in a natural image. During a training phase, a certain number of pictures are selected from a natural image database, basic features of the images are extracted to form a training sample, subsequently, the extracted features are studied by using a depth study model so as to obtain enhanced advanced features which are more abstractive and more distinguishable, and finally, a classifier is trained by using studied features. During a testing phase, as to any test image, firstly, the base features are extracted, secondly, the enhanced advanced features are extracted by using the trained depth model, finally, salience is predicted by using the classifier, and a predicted value of each pixel point serves as a salient value of the point. In such a way, a salient image of the whole image can be obtained, and the higher the salient value is, the more salient the image is.
Owner:NORTHWESTERN POLYTECHNICAL UNIV

Method and apparatus for performing dynamic mechanical analyses

In a dynamic mechanical analysis, a test specimen (2) is coupled to an excitation device (4) by means of a holder device (3). The excitation device (4) applies an excitation force comprised of a static pre-tensioning force component and a time-variable force component to the test specimen, and a deformation of the test specimen (2) is measured by means of one or more displacement sensors (5). The method includes a test phase with the steps: applying the excitation force to the test specimen; while said excitation force is in effect, determining at least one decision parameter, said decision parameter being indicative of a degree of slack in said coupling of the test specimen; comparing said decision parameter to at least one reference value; based on said comparison, determining whether or not the test specimen is coupled to the excitation device in a completely slack-free state, so that physical values derived from said measurements of the deformation will not be subject to errors caused by an insufficient amount of said pre-tensioning force component.
Owner:METTLER TOLEDO GMBH

Method For Regulating the Transmission Parameters of Broadband Transmission Channels Assembled to Form a Group

The invention relates to a novel algebraic method wherein sub-groups of subscribers that spectrally influence each other, in a group of broadband subscriber connections or transmission channels connected to an intelligent network node in a star-shaped manner are classified by observing status variations in operation or in test phases of the individual transmissions, and the spectral influence relation or interference relation is identified with or without using specific test signals during the operation of the network node, without interrupting the data transmissions. The fundamental parameters of the data transmission methods of all of the connected subscribers are optimized from the information obtained in this way, for maximum operator use.
Owner:NOKIA SIEMENS NETWORKS GMBH & CO KG

Face detection method based on convolutional neural network

The invention discloses a face detection method based on a convolutional neural network. The face detection method comprises a training stage and a testing stage, wherein the training stage comprises the following steps: inputting a training sample into the convolutional neural network, and obtaining a connection weight and an offset value of the convolutional neural network. The testing stage comprises the following steps: reading a video image; when a moving object is detected, extracting an interested area, moving for one pixel each time by utilizing a block with a n*n pixel size, and carrying out partitioning processing to the interested area to obtain a plurality of pictures; zooming the partitioned pictures to a size which is the same with the size of the training sample; and inputting the pictures into the convolutional neural network which finishes training for classification, wherein n is greater than or equal to 50 and less than or equal to 70. The convolutional neural network is used as a classifier, so that detection precision and speed can be improved.
Owner:NANJING AIKELESI NETWORK TECH CO LTD

Simulation testing method for partition application software of embedded system

A simulation testing method for partition application software of an embedded system comprises five steps: step one, building a testing model; step two, generating testing configuration information and a virtual partition code according to the testing model; step three, writing a testing script, and writing a response action for an allocated event occurring during the testing process; step four, performing testing on a partition application to be tested; and step five, collecting and analyzing the testing data. The simulation testing method for the partition application software of the embedded system can be implemented in a development and integration test phase of a partition embedded software applicationand can test or verify the execution sequence, the scheduling performance and the interface communication function of the partition embedded software application. The technical scheme of the simulation testing method for the partition application software of the embedded system comprises the steps of utilizing a calculation and storage unit of a core operating system, simulating the dynamic behaviors of other partition applications on line through a virtual partition mode, and exerting testing actuation to the application to be tested and receiving response. The simulation testing method for the partition application software of the embedded system has a broad application prospect in the technical field of software testing.
Owner:BEIHANG UNIV

Image block deep learning characteristic based infrared pedestrian detection method

ActiveCN106096561AAddresses poor selection algorithm performanceSufficient dataCharacter and pattern recognitionVisual technologyData set
The invention relates to an image block deep learning characteristic based infrared pedestrian detection method, and belongs to the technical fields of image processing a computer vision. According to the method, a data set is divided into a training set and a test set. In a training stage, firstly, small image blocks are extracted in a sliding manner on positive and negative samples of the infrared pedestrian data set, clustering is carried out, and one convolutional neural network is trained for each type of image blocks; and then feature extraction is carried out on the positive and negative samples by using the trained convolutional neural network group, and an SVM classifier is trained. In a test stage, firstly, a region-of interest is extracted for a test image, then feature extraction is carried out on the region-of-interest by using the trained convolutional neural network group, and finally prediction is performed by using the SVM classifier. The infrared pedestrian detection method achieves a purpose of pedestrian detection via a mode of detecting whether each region-of-interest belongs to a pedestrian region or not, so that pedestrians in an infrared image can be detected accurately under the conditions such that the detection scene is complicated, the environment temperature is high, and the pedestrians vary greatly in scale attitude, and the method provides support for research in follow-up related fields such as intelligent video.
Owner:CHONGQING UNIV OF POSTS & TELECOMM

Probabilistic boosting tree framework for learning discriminative models

InactiveUS20070053563A1Image enhancementImage analysisProbability propagationKnowledge combination
A probabilistic boosting tree framework for computing two-class and multi-class discriminative models is disclosed. In the learning stage, the probabilistic boosting tree (PBT) automatically constructs a tree in which each node combines a number of weak classifiers (e.g., evidence, knowledge) into a strong classifier or conditional posterior probability. The PBT approaches the target posterior distribution by data augmentation (e.g., tree expansion) through a divide-and-conquer strategy. In the testing stage, the conditional probability is computed at each tree node based on the learned classifier which guides the probability propagation in its sub-trees. The top node of the tree therefore outputs the overall posterior probability by integrating the probabilities gathered from its sub-trees. In the training stage, a tree is recursively constructed in which each tree node is a strong classifier. The input training set is divided into two new sets, left and right ones, according to the learned classifier. Each set is then used to train the left and right sub-trees recursively.
Owner:SIEMENS MEDICAL SOLUTIONS USA INC

Positioning method and system in WiFi environment

The invention relates to a positioning method and a system in a WiFi environment. The method comprises the following steps of: 1, carrying out position partition on the area of the WiFi environment, collecting the signal strength of all positions under the condition that all information sources are not missing and constructing a mapping model between the signal strength and a position label; 2, collecting the signal strength of part of positions under a condition that part of information sources are missing, and solving a regression function by using a current regularization method according to the collected signal strength of all positions under the condition that all information sources are not mising and the collected signal strength of part of positions under the condition that part of information sources are missing; and 3, detecting the signal strength of the information source received by a position to be positioned and detected in the test phase, substituting the signal strength to the regression function, calculating the supplement signal strength corresponding to the information source with the signal strength which is not received on the position to be positioned, and calculating a position label by the mapping model according to the received signal strength and the supplement signal strength. The invention can improve the precision of positioning in the WIFi environment.
Owner:INST OF COMPUTING TECH CHINESE ACAD OF SCI

Display means for vital parameters

A display is connected to at least one sensor for detecting and transmitting physiological measurement values. In addition, one or more displays are used to present the measurement values and / or other associated data visually, acoustically, mechanically, or in some other way perceptible to the senses. A sequence control approach is implemented, according to which the measurement value, before reaching the display, passes through a testing stage, which, on the basis of at least one measured or stored additional value, verifies, checks, or changes the physiological measurement value detected at the sensor or supplements it with additional information. In a process for the noninvasive determination of the oxygen content CaO2 of human and / or animal tissue, the blood oxygen saturation and data on the blood hemoglobin concentration are obtained after the measurements have been acquired by the use of radiation with wavelengths in the range of 400-1,800 nanometers.
Owner:SCHOLLER BERND +5

Display panel and drive method thereof, and display device

The present invention discloses a display panel and a drive method thereof, and a display device. A plurality of crack detection points are arranged on a crack detection line, equivalently, a non-display area is divided into a plurality of areas, in an electricity test phase, each detection switch is controlled to be conducted, signals loaded on each data line are detected to determine whether there are cracks between the detection points or not; if no, uniform bright images or dark images are displayed, if yes, an open circuit is generated between the cracks and the detection points, the signals transmitted from the crack detection line cannot be transmitted to the data lines corresponding to the detection points to cause that the data lines corresponding to other detection points in thedirection of transmission of signals of the detection point are in a suspension state, there is brightness difference between the images of pixel rows corresponding to the data lines and the images displayed when there is no crack, in an IC binding phase, it is determined whether there is a crack or not through a resistance value on the crack detection line is detected, if the resistance value onthe crack detection line is infinite, it shows that there are cracks.
Owner:BOE TECH GRP CO LTD +1

Face and face key point joint detection method based on transfer learning

The invention discloses a rapid face and face key point joint detection method, comprising the following steps: 1, constructing a teacher network and a student network; 2, inputting a batch of training images, and carrying out data enhancement; 3, dividing positive and negative anchor frame samples according to a self-adaptive scale matching strategy; 4, mining positive and negative samples, calculating a multi-task loss function, and updating network parameters; 5, turning to the step 2 until the training converges to obtain a teacher network model; 6, repeating the steps 2 to 5, adding a transfer learning loss function by using the teacher network model, and training to obtain a student network model; and 7, in a test stage, inputting a test image to the student network model to obtain adetection result. The rapid face and face key point joint detection method can obtain face and key point detection results at the same time, and can increase the speed of the face recognition preprocessing process. The lightweight network provided by the invention is high in reasoning speed and can be deployed in embedded equipment with limited computing power.
Owner:SOUTHEAST UNIV

Road traffic sign identification method with multiple-camera integration based on DS evidence theory

The present invention relates to a road traffic sign identification method with multiple-camera integration based on the DS evidence theory, belonging to the technical field of image processing. According to the method, five types of road traffic indication signs which are going straight, turning left, turning right, going straight and turning left, and going straight and turning right are mainly identified, and the method is divided into two parts which are training and testing. In a training stage, the direction gradient histogram feature of a training sample is extracted, thus a sample characteristic and a category label are introduced into a support vector machine to carry out classification training. In a testing stage, an interested region is obtained through image pre-processing, the direction gradient histogram feature of the interested region is extracted and is sent into a classifier to carry out classification, according to the credibility of the sign to be identified obtained by the classifier belonging to each category, and combined with a DS evidence theory data integration method and a maximum credibility decision rule, a final sign identification result is determined. According to the invention, a multiple-camera data integration method based on the DS evidence theory is employed, the information of multiple cameras are integrated to obtain a final identification result, and the road traffic signs can be stably and efficiently identified.
Owner:CHONGQING UNIV OF POSTS & TELECOMM

Image object recognition method based on SURF

The invention provides an image object recognition method based on SURF (Speed Up Robust Feature), comprising the following steps: first, preprocessing images; second, extracting SURF corners and SURF descriptors of the images to describe the features of the images; third, processing the features through PCA data whitening and dimension reduction; establishing a bag-of-visual-words model through Kmeans clustering based on the features after processing, and using the bag-of-visual-words model to construct a visual vocabulary histogram of the images; and finally, carrying out training by a nonlinear support vector machine (SVM) classification method, and classifying the images to different categories. After classification model building of different images is completed in the training phase, the images tested in a concentrated way are detected in the testing phase, and therefore, different image objects can be recognized. The method has excellent performance in the aspects of recognition rate and speed, and can reflect the content of images more objectively and accurately. In addition, the classification result of an SVM classifier is optimized, and the error rate of judgment of the classifier and the limitation of the categories of training samples are reduced.
Owner:SHANGHAI JIAO TONG UNIV +1

A method for segmentation of paint cracks on an ICGA image based on a conditional generative adversarial network

A method for segmentation of paint cracks on an ICGA image based on a conditional generative adversarial network includes: (1), collecting an original ICGA image, extracting a complete fundus oculography image, labeling it with gold standard, normalizing fundus oculography image and gold standard, splicing it into a group of images as sample data, distributing the sample into a training set and atest set according to proportion; (2) based on the principle of conditional generative countermeasure network, constructing the network of generators and discriminators; (3) inputting that data of thetraining set into the network for adversarial train, defining a loss function, and generating a paint crack image correspond to the original picture by the training generator; (4) in the testing phase, inputting the data of the test set, and getting the corresponding paint crack segmentation result diagram through the trained generator G. The segmentation method provided by the invention can be used for solving the problems that the sample size of the ICGA image is small and the acquisition of the contrast image is difficult, and has the characteristics of high accuracy of the segmentation result.
Owner:SUZHOU BIGVISION MEDICAL TECH CO LTD

Real-time image semantic segmentation method based on lightweight full convolutional neural network

The invention discloses a real-time image semantic segmentation method based on a lightweight full convolutional neural network. The method comprises the following steps of 1) constructing a full convolutional neural network by using the design elements of a lightweight neural network, wherein the network totally comprises three stages of a feature extension stage, a feature processing stage and acomprehensive prediction stage, and the feature processing stage uses a multi-receptive field feature fusion structure, a multi-size convolutional fusion structure and a receptive field amplificationstructure; 2) at a training stage, training the network by using a semantic segmentation data set, using a cross entropy function as a loss function, using an Adam algorithm as a parameter optimization algorithm, and using an online difficult sample retraining strategy in the process; and 3) at a test stage, inputting the test image into the network to obtain a semantic segmentation result. According to the present invention, the high-precision real-time semantic segmentation method suitable for running on a mobile terminal platform is obtained by adjusting a network structure and adapting asemantic segmentation task while controlling the scale of the model.
Owner:NANJING UNIV

A heart left ventricle segmentation method based on a deep full convolutional neural network

The invention discloses a heart left ventricle segmentation method based on a deep full convolutional neural network. According to the method, a deep learning idea is introduced into heart magnetic resonance short-axis image left ventricle segmentation; The process is mainly divided into a training stage and a prediction stage, in the training stage, a preprocessed 128 * 128 heart magnetic resonance image serves as input, a manually processed label serves as a label of a network to be used for calculating errors, and along with increase of training iteration times, the error of a training setand the error of a verification set are gradually reduced; And in the test stage, inputting data in the test set into the trained model, and finally outputting prediction of each pixel by the networkto generate a segmentation result. According to the method, segmentation of the heart magnetic resonance short-axis image is achieved from the perspective of data driving, the problem that manual outline drawing is time-consuming and labor-consuming is effectively solved, the defects of a traditional image segmentation algorithm can be overcome, and high-precision and high-robustness left ventricle segmentation is achieved.
Owner:ZHEJIANG UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products