30results about How to "Complementary effect is good" patented technology

The invention provides a traffic missing data completion method based on a bidirectional recurrent neural network, and belongs to the field of traffic. The method comprises the following steps: firstly, the time sequence characteristic of data in time are utilized; meanwhile, the influence of data before and after the completion time point on the current time point is considered, the utilization and completion precision of the data is greatly improved, and secondly, the influence of external features and adjacent sensor data on the current sensor data is considered and added into the completion model, so that the completion precision is greatly improved. According to the method, the completion precision under the condition that the data missing rate is low is greatly improved, and the completion precision under the condition that the data missing rate is high is also improved.

The invention relates to a knowledge base completion method based on multi-modal representation learning, and the method comprises the steps of giving a knowledge base KB which comprises two parts: aknown knowledge set and an unknown knowledge set; performing data preprocessing on the data in the knowledge base; proposing a knowledge base completion model ConvAt, and firstly generating multi-modal representation of a head entity and a tail entity for the acquired data; splicing the multi-modal representation of the head entity, the structural feature vector of the relationship and the multi-modal representation of the tail entity according to columns, respectively processing through a convolutional neural network module, a channel attention module and a spatial attention module, and finally multiplying by a weight matrix to obtain a score of a triple (h, r, t); and training the completion model in the step S2 by using a loss function, and performing knowledge base completion by usingthe trained model. According to the algorithm provided by the invention, external information can be fused, and richer semantic information can be utilized.

A traffic missing data complementation method based on a space-time attention mechanism comprises the steps of firstly capturing the influence degree of all road sections in a road network on the road network traffic state at the current moment in an attention mechanism mode, capturing spatial correlation information again at different moments, and improving the data complementation precision; and secondly, considering the time sequence of the traffic data, the influence degrees of the traffic data at different moments on the data at the current moment are different, capturing the inconsistent time correlation information through a time attention mechanism, retaining the most effective information when the current missing data is complemented, and improving the complementation effect of the model. And finally, while capturing the spatial-temporal correlation of the traffic data by using a spatial-temporal attention mechanism, considering that the correlation between the data is attenuated due to the increase of the spatial distance and the time interval, and adding a spatial-temporal attenuation matrix to improve the completion precision. According to the method, the complementation precision under the condition that the data missing rate is low is greatly improved, and the complementation precision under the condition that the data missing rate is high is also improved.

The invention discloses a tensor low-rank model non-smooth three-dimensional image completion method based on manifold optimization, and the method comprises the steps: setting a tensor Qnuclear normTQN and an orthogonal projection basis in a low-rank completed non-smooth three-dimensional image as learnable image dependent optimization variables through manifold optimization, and updating a datadependent orthogonal projection basis; wherein the input is a limited observation image sample of the non-smooth three-dimensional image under the action of a projection operator, and the output is ato-be-recovered non-smooth low-rank three-dimensional image, so that the low-rank recovery of the non-smooth three-dimensional image is efficiently realized. The method is used for recovering the low-rank image, the applicability of image completion is improved, and the low-rank completion effect of the non-smooth three-dimensional image is improved.

The invention discloses a massage chair with an external counterpulsation function, which does not influence the normal massage use of the massage chair and is convenient to realize household and community popularization. Massage can be carried out after counterpulsation is finished, so that the conditions of numbness, discomfort and the like of legs after counterpulsation can be overcome or reduced, and a better blood circulation effect can be achieved; besides, for a few patients with high heart rate and high blood pressure, the minority of patients can be massaged and relaxed firstly, and external counterpulsation is performed after the blood pressure is stable and the heart rate is reduced; when counterpulsation is carried out, the waist massage function of the massage chair can be selectively started, and in combination with blood circulation during counterpulsation, waist and back muscle groups and spines can be better massaged and subjected to health care. The combination of thetwo is not only the superposition of functions, but also has a better complementary effect.

The invention discloses a three-dimensional spectrum situation completion method based on a generative adversarial network, and the method comprises the steps: obtaining stored historical or experience radio monitoring data offline, and obtaining a plurality of complete three-dimensional spectrum situations or field intensity training data; performing iteration and adversarial offline training on the generative adversarial network variation to obtain the generative adversarial network variation of which the three-dimensional spectrum situation or the field intensity completion mechanism is learned; acquiring actual radio monitoring data of the current three-dimensional target area on line, and preprocessing to obtain defect spectrum situation or field intensity measured data of the three-dimensional target area; inputting the obtained measured data into the generative adversarial network variation to obtain output data of the generative adversarial network variation, and processing to obtain the current three-dimensional target area completion spectrum situation or field intensity. The method is applied to the three-dimensional spectrum situation or the field intensity in the space-air-ground information network, and the complementation error of the three-dimensional spectrum situation or the field intensity can be effectively reduced.

The embodiment of the invention discloses a data processing method and device, equipment and a medium. The method comprises the following steps: carrying out the rasterization of a target region according to the digital map information of the target region, obtaining S grid regions with the same area, and S is an integer greater than or equal to 1; determining the moving point quantity of each grid region according to the position data of the target region obtained by the base station in the preset first time period; complementing the missing fragment movement point quantities of all the gridregions in the preset first time period to obtain a complete movement point quantity of the target region, the fragment movement point quantity being a movement point quantity of one grid region in apreset duration, and the preset duration being a time period in the preset first time period.

The invention provides a vegetation index time sequence reconstruction method in combination with matrix completion and trend filtering, which comprises the following steps of: firstly, determining a missing position in time sequence data through quality marker data in a vegetation index product; the method comprises the following steps of: converting a one-dimensional vector into a two-dimensional matrix through matrix change, establishing an optimal energy equation for low-rank matrix completion for a matrix after each pixel is converted, and realizing matrix completion through an imprecise augmented Lagrange algorithm to obtain a time sequence completion matrix which does not contain data missing preliminarily; and finally, vectorizing the completion matrix, establishing an energy optimization equation of weighted trend filtering on the basis of a one-dimensional vector, and solving the model through an alternating direction multiplier method so as to further filter residual noise and obtain smooth and clean high-quality vegetation index time sequence data. And high-precision reconstruction of the long-time remote sensing vegetation index sequence is realized.

The invention provides a face incompleteness scanning completion method based on deep learning. The method comprises the following steps: taking a depth image and a color image collected by a depth camera; detecting two-dimensional face feature points in the color image, generating three-dimensional face feature points of face scanning according to an internal reference of the camera and depth information in the depth image, and roughly aligning the face scanning to a standard coordinate system where a template face is located according to the three-dimensional feature points; more accurately aligning the face scanning with the template face by using an iterative nearest point algorithm; fitting the template face to the aligned face for scanning by using a Laplace deformation algorithm; setting a distance threshold value on a fitting result, and removing a redundant surface in the face scanning; performing geometric shape completion on incomplete point cloud of a face area by using a PointNet auto-encoder to generate a complete face point cloud. According to the method and the device, the problems of different face scanning poses and redundant surfaces are solved, and geometric face shapes are complemented by using a neural network, so that a better complementation effect is obtained.