599 results about "Data expansion" patented technology

A compressed light field imaging system is described. The light field 3D data is analyzed to determine optimal subset of light field samples to be (acquired) rendered, while the remaining samples are generated using multi-reference depth-image based rendering. The light field is encoded and transmitted to the display. The 3D display directly reconstructs the light field and avoids data expansion that usually occurs in conventional imaging systems. The present invention enables the realization of full parallax 3D compressed imaging system that achieves high compression performance while minimizing memory and computational requirements.

The deterioration (darkness) of image quality due to a reduction in the brightness of a single color as a result of the conversion from RGB pixels to RGBW pixels is prevented and a reduction in the power is achieved. A processing portion for conversion from RGB to RGBW 106 is formed of a W generating circuit 201, which is the same as in the prior art, a sub-pixel rendering circuit 202, a W intensity calculating portion 203 which transmits a W intensity setting value 205 to a W generating circuit 201, and a low power backlight control circuit 204 which expands data on the basis of the RGBW pixels generated by the sub-pixel rendering portion 202 and lowers the backlight in accordance with the amount by which the data is expanded. The inputted RGB data is used as the RGBW data with the W intensity calculated by the W intensity calculating portion 203. A backlight control signal is generated in accordance with the amount of data expansion in the sub-pixel rendering portion 202.

A liquid crystal driving device is provided, which can reduce the problem of contrast lowering of the liquid crystal display screen due to the decrease of the driving current of LED, by the control which is performed in order to cope with the decrease of the maximum rated current of LED as a light source of the backlight at a high temperature. The liquid crystal driving device includes a liquid crystal driving circuit, a backlight control unit, and display data expansion circuits. The liquid crystal driving circuit generates a liquid crystal driving signal to be supplied to a liquid crystal display panel in response to display data. The backlight control unit reduces driving current of the light emitting diode as a light source of the backlight module to illuminate the liquid crystal display panel, in response to the temperature rise of the liquid crystal display panel. The display data expansion circuit, in response to the temperature rise of the liquid crystal display panel, performs the data expansion of the display data, and compensates the contrast lowering of the liquid crystal display panel due to the dimming of the backlight module with the temperature rise of the liquid crystal display panel.

The invention discloses a small data set face recognition algorithm based on machine vision, which comprises the following steps: 1) construction and pretreatment of the data set; 2) generating a virtual image for a category with insufficient images by using a generated antagonistic network; 3) carrying out data expansion on all kinds of images by using a data enhancement algorithm; 4) using convolution neural network to construct a model suitable for data; 5) training the design model; 6) using a camera to obtain a target image, importing the trained model, and carrying out face recognition.The algorithm of the invention can carry out face recognition with high accuracy for application scenes lacking face data.

The invention discloses a machine learning training system and method based on edge cloud computing. A user terminal continuously collects new data expansion and optimal training samples, an output model is trained by a machine learning platform, the trained output model is quickly issued to the user terminal, the user terminal collects the application status of the model and feeds back to the machine learning platform to complete the retraining of the model, a virtuous circle is formed to continuously improve the training ability of the machine learning platform and model accuracy, computingresources are closer to users, and rapid deployment and unified management of the machine learning platform is achieved with the help of resource pooling and resource management capabilities of a cloud computing platform.

The invention provides a method for generating and expanding the pedestrian re-identification data based on a generative network. The method includes the steps of generating a new pedestrian video frame sample by utilizing a video prediction network; generating the end-to-end pedestrian background transformation data by means of a deep generative adversarial network; expanding the breadth and therichness of the pedestrian data set by using different data generation methods; and sending the expanded data set to the feature extraction network, extracting the features and evaluating the performance through the Euclidean distance. According to the method, intra-class and inter-class data expansion of pedestrians is taken into consideration as well, more abundant samples can be generated by combining different generative networks, the expanded data set has good diversity and robustness, the problem of performance loss caused by insufficient number of samples and background interference canbe well solved, and the method has general applicability, and the expanded data set can achieve better performance and efficiency in the next step of pedestrian recognition.

The present invention pertains to a computerized system and method that provides for the secure storage and retrieval of electronic digital information; and, more particularly, to such a computerized system and method that provides for multiple access levels of such secure information; provides for secure access to portions of secure information dependent upon access privileges of the authorized user; provides virtually limitless data expansion capabilities; and provides for rapid access to such secure information by authorized users.

The embodiment of the invention provides a training sample data expansion method and device based on a variational auto-encoder, and relates to the technical field of big data. The method comprises the steps of obtaining an original sample; inputting the original sample into the encoder of the variational autoencoder, wherein the encoder of the variational autoencoder comprises two neural networks, the two neural networks output Mu and sigma respectively, and Mu and sigma are both functions of the original sample; according to the square of the Mu and sigma, namely sigma 2, generating a randomnumber of corresponding Gaussian distribution; randomly sampling the standard normal distribution to obtain a sampling value epsilon, and determining a sampling variable Z according to the sampling value epsilon and the random number of the Gaussian distribution; and inputting the sampling variable Z into a decoder of the variational autoencoder, decoding the sampling variable Z by the decoder ofthe variational autoencoder, and then outputting similar samples of the original samples, and taking the similar samples as extended samples. Therefore, the technical scheme provided by the embodiment of the invention can solve the problems that the time and labor are wasted and the efficiency is low when sample data is manually expanded in the prior art.

The invention discloses an image segmentation method and system based on a wide-residual pyramid pooling network. The method comprises: inputting a to-be-segmented image; standardizing the to-be-segmented image; acquiring a WRN-PPNet model; carrying out pretreatment on a training image and increasing a mode and number of training images by using a data expansion method to obtain a training image set; carrying out model training based on the WRN-PPNet model and training image set to generate a WRN-PPNet segmentation model; and according to the to-be-segmented image, obtaining an image segmentation result by the WRN-PPNet segmentation model. According to the method, the image can be segmented automatically based on the WRN-PPNet to realize target object segmentation without the restriction of the type of the to-be-segmented image; the adaptability is high and the model performance is good; and the accuracy and convenience of image segmentation are improved effectively.

The invention discloses a mechanical equipment fault intelligent diagnosis method based on a generation model under small sample data. The mechanical equipment fault intelligent diagnosis method comprises: carrying out zero-mean-value standardization preprocessing on a small number of obtained mechanical signals; Establishing a composite network for mechanical signal generation; Training a generative adversarial network model in an adversarial manner in combination with a Wasserstein distance and a gradient punishment method; Establishing a deep convolutional neural network model for classifying the operation states of the mechanical equipment by using the mechanical signals; And by combining the generative adversarial composite neural network model and the deep convolutional neural network model, training two networks by using a small amount of real mechanical signals, and finally realizing intelligent fault diagnosis of the mechanical equipment under small sample data. The method hasthe advantages of being good in mechanical signal feature extraction effect, high in state classification accuracy and good in mechanical signal data expansion performance.

A picture taking apparatus and a chromatic aberration correcting method are provided in which a satisfactory correction processing can be performed even when the camera shake correction is simultaneously performed.

An output signal from a camera-signal processing circuit 4 is selected by a selector switch 5 and is supplied to a chromatic aberration correcting unit 6. On the other hand, an angular velocity due to the camera shake is detected using sensors 7P, 7Y and the detected signal is supplied to a camera shake correcting vector calculating unit 9 in control microcomputer 8. A driving state of a camera lens 1 such as a zoom focal length and focal position is supplied to a conversion-ratio calculating unit 10. An optical axis centered shift vector of camera lens 1 is obtained from the camera shake correcting vector and is supplied to the above unit 6. A conversion ratio of each color is supplied to the above unit 6. A signal corrected by the above 6 is compressed by data compression circuit 15 and supplied to a recording medium of a recording and reproducing unit 17 for recording. A signal reproduced therefrom is decompressed by a data expansion circuit 18 and supplied to the selector switch 5.

Accordingly the picture-quality deterioration caused in the miniaturized camera lens can be compensated by processing the captured image signal, and also a satisfactory correction processing can be performed even when the camera shake correction is performed simultaneously.

The invention relates to the technical field of image processing, more specifically, to a depth convolution neural network image denoising method based on an Inception model. Building a training network with one-layer Inception model; setting the parameters of network solution and choosing the appropriate supervisory framework; the parameters of the network structure are optimized, and the obtained network is used for denoising. The depth convolution neural network image denoising method based on the Inception model of the invention is based on the depth convolution neural network and combinesGaussian white noise to simulate unknown real noise and the network containing the Inception model to reduce the parameter quantity on the basis of maintaining the original denoising effect; throughthe way of data expansion, the content of training data is enriched, and the neural network can learn the inner structure of image distribution or noise. By introducing the residual structure, the network converges more easily and the effect is the best.

The invention provides an image high-level semantic recognition method based on multi-feature fusion of a deep network. A global color histogram extracts the color features of an image, the LBP algorithm extracts the texture feature of the image, the deep object network extracts the object features of the image and the deep emotion network extracts the deep emotion features of the image to be fused to identify the composite emotion of the image and the main object contained, Finally, for the input image, the network model can generate descriptive phrases with high-level semantic information, including emotional semantics and object semantics. The invention starts from the aspect of depth learning for small data sets. In this invention, a new method of data expansion is proposed, which combines pre-extracted low-level features such as color and texture statistics, and proposes a multi-feature fusion method to identify the emotion of images and the high-level semantic information of objects. This method improves the accuracy of the experimental results and makes the selection of experimental images more scientific.

The present invention is made such that data in a buffer can be transferred even when read back commands are continuously issued so as to completely eliminate the need for a mechanical operation required for each data block in ROR processing, resulting in considerably reducing a processing time. For this purpose, a method of controlling a magnetic tape unit of the present invention includes the steps of, when the read back command is received from a host processor, reading plurality of data blocks, when a read position of an MTU meets a predetermined condition, reading in a forward direction a plurality of data blocks in the range between the position and a start position of the read back command so as to write the data blocks into the data buffer, storing read commands concerning the data blocks into a command queue in reverse order to how the data blocks are read, and thereafter, in response to a read command from the host processor, reading the data blocks in the data buffer in the order in which the read commands are arranged in the command queue so as to transfer the data blocks to the host processor after data expansion. Further, the invention is applied to a magnetic tape unit to write data compressed through, for example, an EDRC system onto a magnetic tape, and read the compressed data therefrom.

The invention relates to a CNN based non-intrusive power consumption load decomposition method. The method comprises the steps of firstly, collecting the total power of power consumption loads and thepower of each load, and expanding the selected load consumption data by using a data expansion technology so as to generate sufficient data for training and improving the generalization performance of a load data deep learning network; and secondly, constructing a CNN model, training the CNN model by the preprocessed data, automatically extracting load features, and finally performing load decomposition on data to be decomposed by using the trained CNN model so as to obtain utilization and consumption power information of each load. Compared with the prior art, the method has the advantages that the decomposition stability and accuracy are improved, and the robustness and the generalization performance of the network are facilitated to be improved.

The invention discloses an urban road scene semantic segmentation method based on deep learning. The method comprises the following steps: 1) collecting an image at the front end of a vehicle; 2) input data expansion of the annotated image and the original image: randomly cutting, splicing or adding different types of noise to the image, transforming the image through an image affine matrix, and finally maintaining the original resolution of the image through filling, cutting and other transformation to obtain a data set; 3) training a network by using the image after data expansion and the annotation image, wherein the residual U-net network comprises a down-sampling part, a bridge part, an up-sampling part and a classification part; and 4) modifying the time interval T of the acquisitionmodule, inputting subsequently obtained images into the trained deep learning model, outputting predicted semantic segmentation images, and returning different gray levels in the images to the processor. A small data set is used, too fast gradient descent can be prevented, and it can be ensured that an over-fitting problem does not occur during training.

An image processor (24) includes an image processing section (04) which performs a blurring process on a given image data. The image processing section (04) is provided with an image data reduction part (04A), a spatial filter processing part (04B), and an image data expansion part (04C). The image data reduction part (04A) reduces the given image data at a predetermined reduction rate to generate a reduced image data. The spatial filter processing part (04B) performs a spatial filter process on the reduced image data reduced by the image data reduction part (04A) to generate a processed image data. An image data expansion part (04C) expands the processed image data processed by the spatial filter processing part (04B) at an expansion rate inverse of the reduction rate.

The present invention pertains to a computerized system and method that provides for the secure storage and retrieval of electronic digital healthcare information; and, more particularly, to such a computerized system and method that provides for multiple access levels of such secure information; provides for secure access to portions of secure information dependent upon access privileges of the authorized user; provides virtually limitless data expansion capabilities; and provides for rapid access to such secure information by authorized users.

A system for performing hashing includes a controller for controlling the system and for providing a clock signal; an array of integrated circuits; in each integrated circuit, a plurality of cores for performing hashing; and in each core, a plurality of data expanders and data compressors, the data expanders and the data compressors having pipelined circuitry so that two iterations of a hashing loop are performed for each cycle of the clock signal. A method for performing hashing, includes controlling a system having an array of integrated circuits with a clock signal; performing hashing in a plurality of cores in each integrated circuit; and performing for each cycle of the clock signal, in each core, a plurality of data expansion and data compression operations, using pipelined circuitry so that two iterations of a hashing loop are performed for each cycle of the clock signal.

The invention discloses an encryption domain H.264/AVC video reversible data hiding method. Under the premise that the encryption domain H.264/AVC video reversible data hiding method is compatible with an H.264/AVC video compressed encoding standard, code words of prediction modes, Exp-Golomb code words of motion vector difference and sign bits of residual error coefficients are selected to be encrypted, computation complexity is low, the application demands of real-time videos are met, and encryption safety is high. Meanwhile, influences of encryption on the code rate of code streams of H.264/AVC videos are quite small, and the problem of data expansion in the video encryption process is solved. A data hiding person can embed private information in the H.264/AVC videos of an encryption domain directly, and thus the problems of video content safety and privacy disclosure can be solved effectively. The hidden data can be extracted effectively from the encryption domain and can also be extracted effectively from a decryption domain, namely, data extraction and data decryption are separated completely, and practicality is high. In addition, the encryption domain H.264/AVC video reversible data hiding method is completely revisable, and original videos can be restored without damage after the hidden information is decrypted and extracted.

The invention relates to an image data expansion method for deep learning model training and learning, and belongs to the technical field of computer medical image calculation. The method comprises the following steps: firstly, judging a data type, and identifying CT or MRI image data; For the image data, judging whether an ROI (Region Of Interest) is defined or not, and selecting a correspondingmethod to complete the construction of an image data set in combination with the size of a tumor region; Training the image data set by adopting a basic image transformation method to obtain a preliminary training data set; And finally, carrying out data expansion on the preliminary training data set, carrying out deep training by adopting a network model, and finally, carrying out probability prediction. Based on artificial intelligence deep learning, a series of data expansion methods are applied to learning of deep model training in the field of medical image processing, the influence of medical image data heterogeneity on abnormal data is solved, computer-aided diagnosis is facilitated, and the diagnosis efficiency and accuracy are improved.

The invention discloses a small data set finger vein recognition method based on a neural network. The method comprises the following steps: constructing and preprocessing a data set; carrying out data expansion on all types of images by utilizing a data enhancement algorithm; utilizing a convolutional neural network to construct a model suitable for data; training the designed model; acquiring atarget image by using acquisition equipment for preprocessing, and outputting a finger vein feature vector of the preprocessed image through model output; and comparing with a stored finger vein feature vector template to determine the category of the input finger vein. According to the method, through combination of a neural network and data enhancement, features with robustness are extracted; and through Euclidean distance comparison between different image output feature vectors, different types of finger vein images are distinguished according to the size of the distance, so that finger vein recognition can obtain high accuracy on the premise of a small data set.

The invention relates to an early-warning detection method for emergencies in smart city video monitoring. The method comprises the steps that an early-warning detection system is established, and in order to ensure monitoring timeliness, a video is uploaded through double channels; a client preprocesses a monitoring video stream, and a local host only extracts foreground features in key frames through adoption of a background model and through key frame extraction, background separation and foreground feature extraction, the local host packages and compresses the foreground features and uploads the foreground features to a cloud server; moreover, the video stream is uploaded to the cloud server through another channel, a classifier is established through frame interception, object detection segmentation and object feature extraction and through adoption of an SACBA adaptive association rule classification algorithm, and through continuously increasing original video data expansion and through updating of an abnormal feature library, and the accuracy of detecting various emergencies is gradually improved; and the cloud server compares the received packaged feature data packet information with the abnormal feature library, carries out pre-warning judgment and feeds back a result to the client. According to the method, the early-warning monitoring timeliness and accuracy can be ensured.

The invention provides a postgreSQL database mass data management method and system. The method is implemented based on a Linux platform and includes the steps that mass storage management configuration initialization is performed, and system timing scheduling is started; stale data partitions beyond a reserved time quantum are deleted, and then partition data, within a reserved time quantum and beyond an actual storage cycle, in non-stale data partitions is cleared away; storage space checking is performed, whether a storage space supports pre-expansion or not is judged, if yes, partition pre-expansion operation and index establishment are performed, and otherwise, an adaptive simplification mechanism is triggered. According to the postgreSQL database mass data management method and system, the validity, availability and expandability of data are guaranteed, a numerical basis is provided for data expansion, the maintenance efficiency of database administrators is effectively improved, the database management and maintenance cost is greatly reduced, and the postgreSQL database mass data management method and system have significant market value.

The invention discloses an intelligent detection and early warning method and system for an outer envelope in a prostate operation. The method comprises the following steps: 1) acquiring outer envelope image data in a prostate operation video; 2) carrying out gray processing and singular value decomposition on the outer envelope data, and extracting a principal component characteristic value of the image; 3) performing image enhancement on the outer coating image after the image preprocessing in the first step by adopting a deep bilateral learning method; 4) neural network training; 5) detection and early warning. According to the invention, the latest artificial intelligence image recognition technology is utilized to carry out target detection on the prostate envelope, so that an intelligent early warning function is provided for the operations. Different from the existing static medical image recognition technology, the method is characterized in that dynamic images of an operationsite video need to be recognized, early-warned and analyzed. By means of image preprocessing measures such as data expansion, principal component analysis and image enhancement, speed and precision are balanced, and the actual application requirements of prostate operation auxiliary early warning are met.