42results about How to "Reduce data noise" patented technology

A human assisted method of debugging training data used to train a machine learning classifier is provided. The method includes obtaining a classifier training data set. The training data set is then debugged using an integrated debugging tool configured to implement a debugging loop to obtain a debugged data set. The debugging tool can be configured to perform an estimation and simplification step to reduce data noise in the training data set prior to further analysis. The debugging tool also runs a panel of prediction-centric diagnostic metrics on the training data set, and provides the user prediction based listings of the results of the panel of prediction-centric diagnostic metrics.

This document describes a general system for noise reduction, as well as a specific system for Magnetic Resonance Imaging (MRI) and Nuclear Quadrupole Resonance (NQR). The general system, which is called Calculated Readout by Spectral Parallelism (CRISP), involves reconstruction and recombination of frequency-limited broadband data using separate narrowband data channels to create images or signal profiles. A multi-channel CRISP system can perform this separation using (1) frequency tuned hardware, (2) a frequency filter-bank (or equivalent), or (3) a combination of implementations (1) and (2). This system significantly reduces what we call cross-frequency noise, thereby increasing signal-to-noise-ratio (SNR). A multi-channel CRISP system applicable to MRI and NQR are described.

The invention discloses a deep learning-based picture sentiment polarity analysis method, and relates to the technical field of image content understanding and big data analysis. In a conventional picture sentiment analysis method, the final prediction precision is non-ideal due to simple models and features. At present, a deep learning method is used to perform training in a large-scale training set, but the noises of the training set are excessively high, so that the final performance is limited. In the deep learning-based picture sentiment polarity analysis method, a mode of obtaining data directly from a network is adopted, and slave data scale is large. Only sentiment polarity information of common words needed to be obtained during data preparation is possibly needed to be manually annotated. Later, the whole image obtaining and cleaning work can be automatically finished, so that the required labor cost is very low. In the data obtaining stage, two data cleaning processes are introduced, so that a large portion of noises due to inconsistence of pictures and tags can be eliminated. According to the method, priori knowledge is used for filtering the training set, so that the noises of the training set are reduced; and an improved network structure is used assistantly, so that the picture sentiment prediction accuracy is improved.

The invention provides a method and system for revising a user word bank. The method comprises checking whether current input contents are completely or partially same or similar with input contents on input codes and different on characters; and/or checking whether the current input contents are completely or partially same or similar with the input contents on the characters and different on the input codes; revising data in the user word bank based on the current input contents and error correction contents if conditions are met; and enabling the error correction contents to be a part of the input contents corresponding to the current input contents. The method and system can intelligently record user input information, avoids learn misinput words as much as possible and reduces data noise in the user word bank. The method and system does not need more limitation on user editing actions, greatly expand application range and depth of word bank revising, and can better remove the data noise which cannot be found in the prior art.

The invention relates to the technical field of computer time series data analysis and prediction, in particular to a time series data filling and restoring method based on machine learning. The method includes: filling the missing value by using a domain-based median and mean value filling method; estimating a true value of an expected sampling moment through a linear rule; detecting wave crestsand wave troughs of the time sequence, and smoothing abnormal values; taking hundreds of thousands of collected real data as samples, designing and generating time sequence characteristics, taking real results as labels, and training a machine learning model based on an XGBoost (Extreme Gradient Boost) for predicting a large number of unknown data. According to the method, the problems of multiplemissing values, large volatility, error accumulation and the like of specific time sequence data are solved, and the accuracy of data filling and restoring is effectively improved; moreover, the complexity of a machine learning model is well controlled, the filling and restoration of hundreds of millions of data records can be completed within an hour level, and the method has a high practical value.

Data lines (D0, D1) are shared by a first storage portion (MA) and a second storage portion (MB), and furthermore, a first transistor (MC0) coupled to a first comparison data portion (CD0) and a second transistor (MCA) coupled to the storage node of a first storage portion are connected in series to form a first comparing circuit (11), and a third transistor (MC1) coupled to a second comparison data line (CD1) and a fourth transistor (MCB) coupled to the storage node of the second storage portion are connected in series to form a second comparing circuit (12). Consequently, it is possible to enhance a symmetry in the layout of a diffusion layer and a wiring layer and to achieve the easiness of a layout in which a memory cell is line symmetrical with respect to a center line passing through a center thereof. Thus, a manufacturing process condition can easily be optimized and a variation in a manufacturing process can be reduced so that the microfabrication of the memory cell can be achieved.

The present invention discloses an oil-gas microorganism gene exploration method. The method comprises the following steps: respectively collecting samples of surface shallow layers above a known oilwell, a gas well and a dry well in an exploration area, performing high-throughput sequencing after extracting DNA, establishing a microbial community composition pattern diagram in the exploration area according to a sequencing result, respectively screening out characteristic microorganisms in surface soil above the oil/gas well in the exploration area according to the pattern diagram, and designing primers according to attribute characteristics and carrying out a fluorescent quantitative PCR detection on the samples in the whole exploration area to detect the number of the characteristic microorganisms. A contour line of the characteristic microorganisms obtained by the oil-gas microorganism gene exploration method has a high coincidence rate with the known well, has a good coincidencerate with a trap line, and can carefully describe an oil area.

The invention provides a novel intelligent skylight system for environment prediction. The novel intelligent skylight system comprises a window frame, a window sash installed on the window frame, a driving device for driving the window sash to rotatably open and close, open and close in a translating mode or open and close in a folding and unfolding mode, a control system for controlling startingand stopping of the driving device and an environment prediction system, wherein the environment prediction system monitors real-time weather, calculates a real-time weather index and predicts the weather index; and when weather is changed or is about to change, different skylight control commands are sent according to the weather changes, after the skylight control commands are received, the control system controls the driving device connected with the control system to drive the window sash of a skylight to rotate, translate, fold and unfold. The novel intelligent skylight system has the beneficial effects that through data collection and data processing of the surrounding environment of the skylight, the environment for a period of time in the future is reasonably predicted, and the skylight is opened and closed in advance according to the predicted weather index, so that manpower waste is reduced, and loss of life and property is reduced.

The invention relates to a safety information supervision device applied to a high-speed rail signal system. The safety information supervision device comprises a communication processor, an RBC interface machine, a CTC interface machine, a supervision analysis server, an analysis terminal and a storage device. The communication processor and the RBC interface machine are in communication connection with the supervision analysis service through a detection network. The CTC interface machine is in communication connection with a CTC system of a station and the supervision and analysis server, and is used for acquiring information of a train in the CTC system and transmitting an analysis result of the supervision and analysis server to the CTC system through the CTC interface machine. According to the device, multi-source data fusion analysis is adopted, filtering and validity analysis are carried out on basic data, and false alarms are reduced.

The invention provides a method for automatically correcting a pitch angle of a vehicle-mounted laser radar. The method comprises the steps: screening and obtaining points of the laser radar on the ground according to a denoising strategy; calculating the current pitch angle of the laser radar for points on the ground according to a ground fitting strategy; and correcting the current pitch angle by adopting an updating strategy. According to the method, straight line fitting is carried out on the point cloud point ground data after data denoising processing according to the ground fitting strategy, the current pitch angle is estimated and obtained, the operand is small, and few operation resources are occupied.

According to the traffic sign recognition method provided by the invention, the detection effect on the road traffic sign is remarkably improved, and the requirement on real-time performance is met. A weighted bidirectional feature pyramid network with more balanced accuracy and efficiency is adopted to replace a path aggregation network, and channel features of road traffic signs are better fused. Secondly, common convolution is replaced by the cavity convolution, and the cavity convolution is combined with the space pooling pyramid, so that the receptive field is further expanded. Meanwhile, the detection scale is increased to four types, and the small target detection effect is improved; and random cutting is added in a data enhancement technology, so that the model learns more detail features. Finally, digital image operation technique is used to increase the number of instances for low precision categories.

Disclosed are an in-band pseudolite wireless positioning method, system and device, the system comprising: a base station, a pseudolite and a terminal; the base station is used to transmit identifier information to the pseudolite after correcting the transmission clock of the pseudolite, and transmit a pseudolite array and positioning correction information to a terminal; the pseudolite is used to generate a random positioning signal sequence according to the identifier information, and utilizes the positioning link having the same frequency band as the base station wireless system to transmit a positioning signal according to the transmission clock and the random positioning signal sequence; and the terminal is used to generate a random positioning signal sequence of the pseudolite according to the pseudolite array and the positioning correction information, and match the received positioning signal according to the random positioning signal sequence to obtain the arrival time of the positioning signal, and obtain through calculation the position coordinates of the terminal according to the position coordinates of the pseudolite and the arrival time.

The invention discloses a self-balancing mounting device and method for a large ground penetrating radar. The self-balancing mounting device comprises a support (1), a plurality of mounting steel wire ropes (2) and a tensioning device (3). Two mounting steel wire ropes (2) with end heads at two ends are arranged on the support (1); and the mounting steel wire rope (2) is connected with the tension device (3). Two mounting steel wire ropes (2) with ends at the two ends are arranged on a support (1), a tensioning device (3) is used for transversely tensioning the mounting steel wire ropes (2), and the radar is lifted up to be connected with the support (1) along with tensioning of the mounting steel wire ropes (2); and a rubber block (5) for shock insulation is arranged on the mounting steel wire rope (2), so that the effects of protecting the radar and reducing interference to data are achieved, the mounting of the ground penetrating radar can be completed by a single person, the labor intensity is reduced, the mounting of four hanging lugs can be completed by transverse tensioning and one-time tensioning, and time and labor are saved.

The invention provides a financial event extraction method based on combination of a pre-training language and a deep learning model. The financial event extraction method based on combination of a pre-training language and a deep learning model comprises the following operation steps: S1, data acquisition and preprocessing: crawling public financial event text corpora by using a web crawler, and performing text preprocessing on the original financial event text corpora. According to the financial event extraction method based on combination of a pre-training language and a deep learning model provided by the invention, financial field event types and templates are defined by using a mode of combining machine learning with field knowledge, so that the time cost and the labor cost of manually defining events are greatly reduced; large-scale automatic labeling of financial field event corpus data is realized by using a remote supervised learning mode; data noise is effectively reduced by using a heuristic pruning method; and the blank of lack of large-scale corpus data in the current financial event extraction field is filled.