120results about How to "Hard to get" patented technology

The invention discloses a fault diagnosis-based rail transit vehicle bearing residual life prediction method. The method includes the following steps that: S100, multi-angle feature extraction and fusion are performed; S200, cyclostationary theory-based fault diagnosis is performed; S300, multi-variable model training is performed, so that multi-variable model-based fault isolation is realized; and S400, based on a semi-supervised algorithm, a BP neural network algorithm and a support vector regression algorithm are used in combination to perform fault diagnosis-based residual life prediction. With the fault diagnosis-based rail transit vehicle bearing residual life prediction method of the invention adopted, fault detection, fault isolation and residual life prediction at the early stage of the fault of a bearing are realized, and a basis is provided for the formulation of vehicle bearing maintenance decision making.

The invention relates to a software defect predication method based on cost sensitivity and semi-supervision. The software defect predication method comprises the following steps: S1, collecting a source code file of software to be predicated through a version control tool; S2, extracting a measurement element value from the source code file, wherein the source code file obtained by S1 is composed of I modules; S3, obtaining a sampling result set through a measurement element value selecting module in a sampling manner; S4, constructing a training set by a labeled sampling result set and a label-free sampling result set; S5, extracting a target function and solving a classification function enabling the target function to be minimum; and S6, predicating a module to be predicated through the classification function and outputting a predicated result. The method provided by the invention fuses semi-supervision and cost sensitivity thoughts to construct a software defect predication model and two problems in software defect predication that defect data is difficult to obtain and types are not balanced are solved; and the accuracy of the predicated result is extremely improved.

The invention discloses a method for splitting traffic trip modes on the basis of mobile phone data.The method comprises the steps that according to mobile phone behavior characteristics, on the basis of the mobile phone data, a home address community and a work community of a mobile phone user are predicted, a commuting O-D matrix between the home address community and the work community of the mobile phone user is constructed by means of communication records, produced in the home address community and the work community, of the mobile phone user on working days; the commuting trip time, trip distance and trip speed are calculated by means of time information and position information which are contained in the mobile phone data, a speed and distance rank algorithm is built through trip characteristics of different trip modes, urban traffic network information data is introduced, a Logit model based on the trip time is built to achieve large scale urban traffic trip mode splitting, and a novel economical and feasible method is provided for traffic demand prediction among the traffic modes.

A hair trimmer includes a main body; a cutter block having a stationary blade and a movable blade, both of which are comb-shaped and are slidably in contact with each other, the cutter block being protrudingly installed at one end of the main body; and a comb attachment having a plurality of comb tooth-shaped sections and covering the cutter block, wherein the comb tooth-shaped sections are provided with cut-off hair deflectors to prevent hairs cut by the cutter block from outwardly scattering away from the comb attachment, cut-off hair deflectors located between two neighboring comb tooth-shaped sections are disposed at relatively inner and outer positions.

The invention discloses a method for finely classifying polarized SAR images based on Freeman entropy and self-learning. The problems that in existing supervised classification, surface feature labels are difficult to obtain, and shadow regions and mixing scattering regions are difficult to distinguish are mainly solved. The implementation process of the method comprises the steps that (1) eigenvalue decomposition is carried out on a polarization coherence matrix to obtain three characteristic values; (2) decomposition is carried out on a polarization covariance matrix to obtain three kinds of scattered power; (3) characteristic vectors are input according to the three characteristic values and a volume scattered power structure; (4) spectral clustering is carried out on the input characteristic vectors of random sampling points; (5) SVM classification is carried out according to the sampling points and the clustering marks of the sampling points; (6) MRF iteration is carried out on a classification result; (7) spectral clustering is carried out on wrongly-classified pixel points, and the fine classification surface feature categories of the polarized SAR images is obtained. Compared with an existing SAR image classification method, the method does not need manual label defining, the classification result is more precise, and the method can be used for target detection and classification recognition of the polarized SAR images.

A man-machine interface which provides tactile feedback to various sensing body parts is disclosed. The device employs one or more vibrotactile units, where each unit comprises a mass and a mass-moving actuator. As the mass is accelerated by the mass-moving actuator, the entire vibrotactile unit vibrates. Thus, the vibrotactile unit transmits a vibratory stimulus to the sensing body part to which it is affixed. The vibrotactile unit may be used in conjunction with a spatial placement sensing device which measures the spatial placement of a measured body part. A computing device uses the spatial placement of the measured body part to determine the desired vibratory stimulus to be provided by the vibrotactile unit. In this manner, the computing device may control the level of vibratory feedback perceived by the corresponding sensing body part in response to the motion of the measured body part. The sensing body part and the measured body part may be separate or the same body part.

The invention discloses an assimilation evapotranspiration and LAI (leaf area index) region soil moisture monitoring method, and the method comprises the specific steps: S1, calibrating a crop growth model, and completing the determination of spatial parameters; S2, generating an MODIS ET and LAI time sequence curve through superposition; S3, constructing an LAI time sequence curve through filtering; S4, calculating the one-order differential monotonicity of the curve; S5, operating an SWAP model grid by grid, outputting the ET and LAI, and calculating the one-order differential monotonicity; S6, building a cost function according to the difference between the one-order differential monotonicity obtained at step S4 and step S5, taking an seedling emergence date and irrigation time as to-be-optimized parameters, and obtaining an optimal value through employing an optimization algorithm; S7, inputting the optimal parameter into a model, and simulating soil moisture grid by grid. The method estimates the soil moisture through a data assimilation method, introduces two types of remote sensing data, improves the precision of large-area monitoring of soil moisture, and is suitable for the estimation of the soil moisture of regional-scale of farmland.

The invention, which belongs to the technical field of multi-sensor information fusion, discloses an evidence-synthesis-based information-fusion target recognition method. A plurality of sensors are used for carrying out attribute information collection on a to-be-identified target and a feature attribute is extracted from the collected attribute information; data having the feature attribute aredivided into training data and testing data, wherein the training data are used for constructing a neural network model and the testing data are used for obtaining a basic probability assignment value; and then evidences are synthesized based on an improved evidence synthesis method and the synthesized result is used as the target recognition basis. According to the invention, the basic probability assignment values of evidences are obtained accurately and a synthesis problem of high-conflict evidences is solved. The basic probability assignment values of evidences are obtained by using the neural network and the neural network has the high nonlinear mapping capability and is capable of mapping the intrinsic relationship between the target feature data, so that the accuracy of the basic probability assignment values is ensured, the conformance to the real scene is realized, and the practical significance is good.

The invention discloses a human body fall detection method based on an acceleration sensor and relates to the technical field of biomedical signal processing. The human body fall detection method is based on a double check human body fall detection algorithm, wherein primary fall detection comprises the following steps: extracting a threshold value set according to pre-acquired activities of dailyliving (ADL) and fall acceleration sample data; then acquiring ADL acceleration data of a human body in real time, and extracting variance of the data as a dynamic threshold value for partially updating a pre fall behavior threshold value specific to a user; and setting the threshold value as a human body fall judgment standard, and performing primary detection on a human body fall behavior; in aprimary fall detection process, once a fact that a pre fall behavior happens is detected, human body behavior data t seconds before the pre fall behavior is sent onto a server near a data source, andsecondary fall detection judgment based on SVM is performed; and finally according to a dual fall judgment result, a human body fall event is determined, and the threshold value set of a fall detection algorithm is dynamically updated to be used for follow-up fall detection on the user.

The invention discloses a monocular unsupervised depth estimation method based on a context attention mechanism, and belongs to the field of image processing and computer vision. According to the method, a depth estimation method based on a hybrid geometry enhancement loss function and a context attention mechanism is adopted, and a depth estimation sub-network, an edge sub-network and a camera attitude estimation sub-network based on a convolutional neural network are adopted to obtain a high-quality depth map. The system is easy to construct, and a corresponding high-quality depth map is obtained from a monocular video in an end-to-end mode by using a convolutional neural network; the program framework is easy to realize. According to the method, the depth information is solved by usingan unsupervised method, the problem that real data is difficult to obtain in a supervised method is avoided, and the algorithm operation speed is high. According to the method, the depth information is solved by using the monocular video, namely the monocular picture sequence, so the problem that the stereoscopic picture pair is difficult to obtain when the monocular picture depth information is solved by using the stereoscopic picture pair is avoided.

The invention discloses a query representation and hybrid retrieval model construction method based on a context sensing theme. The method includes the following steps that firstly, a pseudo relevance feedback document of query is obtained on the basis of a keyword set of query, and context related to query is selected from the pseudo relevance feedback document; secondly, a context sensing theme model is introduced, the context is fused into the context sensing theme model, and implicit theme information of a context window is mined on the basis of a corpus theme to obtain a corresponding theme vector; thirdly, query is represented by combining the theme vector and the keyword set, a hybrid retrieval model is constructed on the basis of the theme vector and the keyword set, and a final retrieval score is obtained.

The invention discloses a convolutional neural network (CNN) model generation method. The method comprises the following steps: constructing a CNN model structure and setting a loss function expression; A plurality of training image pairs are acquired, each training image pair comprises an input image and a target image, and the input image and the target image are images obtained by shooting thesame scene through a mobile terminal and a single lens reflex; Each training image pair is registered, so that an input image included in the training image pair is aligned with the target image; Using the registered training image pair as a training sample to train a CNN model; and continuously updating model parameters of the CNN model, calculating a loss function value of the CNN model according to the loss function expression every time the model parameters are updated, and when the loss function value is converged, stopping the training process to obtain the model parameters of the CNN model so as to generate the CNN model. The invention also discloses a corresponding image quality optimization method and computing equipment.

It is intended to provide sanitary paper which is excellent in texture such as moistness and softness and scarcely causes skin irritation or blushing even if it is brought into contact with the skin frequently. Namely, sanitary paper having an oil absorbance specified in JIS P8141 of 7 mm or less and a moisture content of from 9.50 to 15.00% (measured in accordance with JIS P8127 after conditioning in accordance with JIS P8111); carrying a solution, which contains a moistening agent, a softener, an antioxidant and so on, coated in a dose of 46.0 to 160.0 mg/cm3 of the paper base; and having a bending hardness B measured with the use of a pure bending machine of from 0.0040 to 0.0060 g·cm²/cm.

The invention discloses a fused convolutional neural network-based wear surface three-dimensional topography measurement method, which comprises the following steps of: generating a random rough surface through a two-dimensional digital filtering technology, and obtaining a luminosity image sequence of the random wear surface by utilizing Blender rendering software so as to generate a data set forneural network training; designing a feature extraction module, a fusion module and a normal vector estimation and refinement module to obtain a fused convolutional neural network applied to wear surface normal vector estimation; defining a training loss function of the neural network, and training and adjusting a network model based on the data set; and in combination with priori knowledge of the abraded surface, solving the depth information of the abraded surface based on a regularization algorithm. According to the method, the neural network method and the photometric stereo technology are effectively combined, the problem that the reflection characteristics of the abraded surface are not matched with the Lambert model is solved, and accurate reconstruction of the abraded surface is achieved in combination with priori knowledge of the abraded surface.

A heat dissipating element for a memory is composed of a set of fins which can be latched oppositely, and a hook member which can be used to clip the set of fins, wherein the aforementioned fins are hooked with a T-shape hook plate having horns and a T-shape slot having through-holes and a through-hole at its neck. The hook member is provided with perforations on its surfaces, and slant press pieces which are extended downward are located on surfaces of fins. After the fins are clipped by the hook member, the hook member is hooked with the slant press pieces through the shear sides below the perforations. In addition, lower edges of the aforementioned fins are provided with slant ribs extended outward, and by pressing the ribs, the aforementioned slant press pieces can be separated from the perforations of hook member.