35results about How to "Maintain structural features" patented technology

The invention discloses a SAR image segmentation method based on wavelet pooling convolutional neural network. The scheme is: 1. Construct the wavelet pooling layer and form a wavelet pooling convolutional neural network; 2. Select image blocks and input them into the wavelet pooling convolutional neural network for training; 3. Input all image blocks into the training Test in a good network to get the first category of SAR image; 4. Segment the SAR image with superpixels, and fuse the result with the first category of SAR image to get the second category of SAR image; 5. According to Markov random The airport model obtains the third category of SAR image, and fuses it with the result of superpixel segmentation to obtain the fourth category of SAR image; 6. According to the SAR image gradient map, the second category and the fourth category of SAR image fusion to get the final segmentation result. The invention improves the segmentation effect of the SAR image and can be used for target detection and recognition.

The present invention provides a dynamic network edge sampling and its visualization method, comprising the following steps: 1) select any group of node pairs in the dynamic network, and use the kernel density estimation method to calculate the probability density function of the edge between the pair of nodes; 2) according to the node The probability density function of the paired edges constructs a suitable reference distribution function, which approximates the real probability density distribution; 3) For each edge of the target node pair, use a 0‑1 uniform distribution to obtain a random value, and calculate the edge The ratio of the probability density at each moment to the value of the reference distribution function constructed in the second step, compare the relationship between the ratio and the random value, and judge whether to accept the sample; 4) Traversing all node pairs in the dynamic network, repeat steps 1) to 3 ), to obtain the edge sample set of the dynamic network after sampling. The invention can basically maintain the structural features of the original dynamic network while reducing the scale of the dynamic network.

The present invention discloses a process for producing a nanometer gold granule containing diamond-like carbon thin films. Said process comprises the steps of preparing a poly(phenylcarbyne)-chlorauric acid containing tetrahydrofuran precursor solution, coating a monocrystalline silicon surface with the mixing solution through spin-coating method, resulting in formation of chlorauric acid-poly(phenylcarbyne) thin film at silicon face after volatilization of the organic solvents; processing the silicon chips coated with thin films for 1-2 hours at a temperature of 500-700 DEG C under argon atmosphere, and then naturally dropping to room temperature to obtain gold nanometer granules containing diamond-like carbon thin films. The present invention is characterized by its easy preparing course, wherein the metal nanometer granules containing diamond-like composite thin films are prepared through one-step synthesis. Said thin film is hopeful to find uses in fields of mechanical friction, electrochemistry electrode, metal catalysis, optics and the like.

The invention discloses a preparation method of a heteropoly acid supported metal oxide nano fiber. The fiber takes metal oxide a supporter and heteropoly acid as an active component, wherein the weight ratio of the oxide supporter and the heteropoly acid is 1:(0.4-1). The preparation method comprises the following steps: a) spinning a compound precursor containing heteropoly acid and metal acetate to prepare a nano fiber containing the heteropoly acid / metal acetate precursor; and b) performing roasting treatment on the nano fiber containing the heteropoly acid / metal acetate precursor at 300-450 DEG C to obtain the heteropoly acid / metal oxide composite nano fiber. The fiber keeps the structural features of the heteropoly acid itself, and combines the characteristics of the nano fiber such as large specific area, high dispersity and the like; and the preparation method is simple and low in cost.

The invention discloses a large graph sampling visualization method based on graph representation learning, and belongs to the fields of graph visualization and graph sampling. The method uses the node2vec algorithm to convert the nodes in the original network into high-dimensional vectors, and then uses the dimensionality reduction algorithm to project the high-dimensional vectors of the nodes into a low-dimensional space, and the distance between the projected points can effectively express the corresponding nodes in the network space. Semantic structure similarity in . Then, a multi-objective sampling model of adaptive blue noise sampling is designed to effectively maintain the topology of the original network; a measurement index based on network attribute characteristics is proposed, and different sampling algorithms are quantified and evaluated, and the graph sampling result evaluation is obtained, which is presented by a visualization method Graph sampling result evaluation. The method samples the nodes in the representation space, well simplifies and preserves the context structure of the original network, and effectively maintains the topology structure of the network while reducing the scale of the nodes.

The invention relates to a specific animation face generation method, terminal device and storage mediumanimation, and the method achieves the color pixel style migration and preliminary animation stylization from a real face to a animation face through combining with a conventional neural style migration technology based on a convolutional neural network. In combination with an existing style migration technology based on a generative adversarial network, domain style migration from a preliminarily animation stylized real face to a specific animation face is realized, and a animation real face image of a specific animation figure is obtained. According to the method, color features can be extracted, facial structure features can be maintained, and facial deformation is reduced; and a required image can be generated for a specific animation character.

The invention provides a preparation method for a spherical cellulose aerogel, which relates to a preparation method for an aerogel. The objective of the invention is to overcome the problems that the phenomena of structure collapse and volume contraction of a conventional cellulose aerogel frequently occur and the specific surface area of an obtained aerogel is small. The method comprises the following steps: 1, putting vegetable fiber in a mixed solution of urea and NaOH and carrying out stirring after refrigeration for formation of a solution so as to obtain a vegetable fiber / alkaline urea solution; 2, mixing trichloromethane with ethyl acetate and then adding glacial acetic acid so as to obtain an organic solution; 3, titrating the vegetable fiber / alkaline urea solution into the organic solution of glacial acetic acid and carrying out ageing so as to obtain a gel; 4, putting the gel into a glacial acetic acid solution for solidification; and 5, replacing a hydrogel successively with absolute ethyl alcohol and tert-butyl alcohol and carrying out refrigeration and freeze-drying so as to obtain the aerogel. According to the invention, the phenomena of structure collapse and volume contraction of an aerogel in the process of drying are overcome, and the prepared cellulose aerogel has low density and a specific surface area of more than 240 m<2> / g.