51results about How to "Obvious difference" patented technology

The embodiment of the invention discloses a face image processing method and device and a server. The method comprises the following steps: obtaining a to-be-processed human face image; inputting thehuman face image into a pre-trained convolution neural network model, and obtaining the classification data outputted when the convolution neural network model responds to the input of the human faceimage, wherein the convolution neural network model takes a loss function as a constraint condition, so as to enable the feature cosine value of each class of the classification data to approach to one; obtaining the classification data, and carrying out the content understanding of the human face image according to the classification data. The cosine value between a feature vector and a weight value of the loss function is enabled to approach to one, so as to achieve the convergence of inner-class distance. The convergence of inner-class distance enables the inner-class distance of the classification data to be increased, and the increase of the inner-class distance enables the classification data to be different more apparently. The increase of the robustness of data also can improve theaccuracy of content understanding.

The invention belongs to the field of epoxy resin coatings and particularly relates to a water-based flame-retardant epoxy resin coating. The water-based flame-retardant epoxy resin coating is prepared from a component A, a component B and a component C in a mass ratio being 5:1:1, wherein the component A is prepared from raw materials in parts by weight as follows: 90-100 parts of bisphenol A epoxy resin, 10-20 parts of polyoxyethylene alkyl ester and 10-25 parts of polyacrylic ester; the component B is prepared from raw materials in parts by weight as follows: 20-25 parts of triethylene tetramine, 3-5 parts of titanium dioxide, 8-10 parts of pigments, 100-110 parts of talcum powder and 30-40 parts of water; the component C is prepared from raw materials in parts by weight as follows: 1-3.2 parts of a wetting dispersant, 1.2-1.6 parts of a defoaming agent, 1.2-1.8 parts of a flatting agent, 15-20 parts of a phosphorus-nitrogen composite flame retardant, 5-6 parts of a hypophosphite flame retardant and 5-6 parts of phosphate. The epoxy resin coating has the advantages of being environment-friendly, high in film hardness, high in anti-impact strength, good in glossiness, good in flame retardance and the like.

The invention discloses a T91-steel ageing ranking method based on a transmission electron microscope. The T91-steel ageing ranking method comprises the following steps of: (1) acquiring transmission electron microscopic graphs of a sample to be ranked and the original sample; (2) carrying out statistics on the number and the sizes of crystal-boundary carbide particles in the transmission electron microscopic graphs of the sample to be ranked and the original sample; and (3) ranking the ageing degree of the sample to be ranked according to the size difference of the crystal-boundary carbide particles in the transmission electron microscopic graphs of the sample to be ranked and the original sample. The T91-steel ageing ranking method disclosed by the invention has the advantages that the ageing condition of T91 in use can be accurately evaluated, and intuitive and scientific basis is provided for maintenance and replacement of parts using T91 steel.

A method for producing a print includes a step of thermally bonding a thermally transferrable image-protecting layer onto an image surface of an image-receiving sheet so as to form a protective layer. In this step, thermal energy applied is changed so that the resulting protective layer has a first glossy region and a second glossy region having a glossiness lower than that of the first glossy region. The first glossy region has a glossiness a, which is a 20° mirror-surface glossiness measured according to Japanese Industrial Standard Z8741. The second glossy region has a glossiness b, which is a 20° mirror-surface glossiness measured according to Japanese Industrial Standard Z8741. The value obtained by subtracting the glossiness b from the glossiness a is 60 or more.

The invention provides an indoor unit of an air conditioner. The indoor unit comprises a shell and a door body assembly installed on an air outlet. The door body assembly and the outer surface of the shell are in smooth transition when the door body assembly seals the air outlet. The door body assembly further comprises two decorative blocks and a door sheet rotationally installed between the two decorative blocks. The indoor unit further comprises a telescoping mechanism and a rotating mechanism, wherein the telescoping mechanism is installed on the air outlet and used for pushing the decorative blocks to conduct opening and closing reciprocating motion, and the rotating mechanism is used for driving the door sheet to rotate relative to the two decorative blocks. The working process of the indoor unit includes the steps that the telescoping mechanism pushes the decorative blocks out, the rotating mechanism drives the door sheet to rotate by a preset angle, the telescoping mechanism retracts, and the rotating mechanism drives the door sheet to swing back and forth. Through opening, closing and swing of the whole door body assembly, the whole indoor unit of the air conditioner has the sense of science and technology and is fashionable in appearance; in addition, the door sheet has both the dustproof function and the swing function, and compared with an existing indoor unit of the air conditioner, the indoor unit is better in comfort and more obvious in differentiation.

The invention discloses a terminal reconfiguration method based on a fuzzy optimization theory in software radio environment. The terminal reconfiguration method is characterized by comprising the following steps of: (1) acquiring attribute parameters of a coverage network in an area where a current terminal is located from a network reconfiguration manager (NRM) by using a terminal reconfiguration manager (TRM) of the current terminal; (2) determining the weights of the attribute parameters of the coverage network by using an expert scoring method, dividing all reconfiguration schemes into high-class schemes and low-class schemes by configuring a characteristic value index membership degree matrix, and calculating the goodness and badness of each reconfiguration scheme; and (3) calculating the membership degree of each scheme according to the goodness and the badness, and determining an optimal reconfiguration scheme according to a membership degree maximization principle. By adoption of the method, scheme difference can be obviously embodied, and the uniformity of an evaluation result is avoided. The method has the advantages of simplicity in calculation, practicability and wide application prospect. The evaluation process and the evaluation result have obvious physical meanings.

The invention belongs to the technical field of neural networks, and especially relates to a method and a system for constructing a large-scale trapping scene based on cloud computing. The method comprises the following steps that 1, a plurality of clouds which are the same as an application layer, a transmission layer, a network layer, a link layer and a physical layer of a local end are established, and each cloud is regarded as a mirror image bait container of the local end; 2, bait resource data of a local end are collected, wherein the bait resource data at least comprise user behavior data, application use data, network environment data, login credential data, file data and flow data; and for the collected local end data, data preprocessing is performed to obtain preprocessed data, and data classification is performed on the preprocessed data. According to the method, the bait resource data is subjected to classified learning to ensure that different bait resource data can be subjected to more adaptive model training, so that the generated bait is more fraudulent.

A target base-specific primer for use in discrimination of the base type of a target base in a nucleic acid in a nucleic acid sample, which is capable of discriminating the target base clearly with less possibility of false positive, and a target base discrimination method using the target base-specific primer, are provided. The target base discrimination method of the present invention uses a target base-specific primer for use in discrimination of the base type of a target base in a nucleic acid having a target nucleotide sequence including the target base in a nucleic acid sample, in the case where a plurality of base types can be enumerated for the target base, wherein the primer has a target base-corresponding base which is a base complementary to a base predicted as the target base, and the base type of a mismatch base is a specific base type when the mismatch base refers to a base adjacent to the 5′ side of the target base-corresponding base, and a mismatch-corresponding base refers to a base adjacent to the 3′ side of the target base in the target nucleotide sequence.

The invention relates to a method and a device for detecting offset of a code spraying position. The method for detecting the offset of the code spraying position comprises the following steps that more than one marker are arranged on one side of a prefabricated code spraying position of a prefabricated part; the code spraying position is adjusted to be consistent with the corresponding prefabricated code spraying position; a marking material is sprayed to the markers; a detection device analyzes the reflection of the marks, and an alarm is given out if the offset exists; and according to theoffset condition, the spraying direction of spraying heads is adjusted. The method has the beneficial effects that the markers on one side of the code spraying position of the prefabricated part havea photosensitive function, and a detection part of the detection device is used for detecting the photosensitive markers; because the markers are of circular truncated cones or conical bulges, the detection part can identify the offset direction according to the color change of photosensitive parts on the bulges, and further fine-tune the position of the spraying heads; and because the detection is carried out according to the change of the photosensitive brightness, the arrangement of bulge structure makes the difference of the photosensitive brightness more obvious and the method is more suitable for staff to identify.

The invention relates to a method for defining an interface (32, 42, 52, 62) of a welding power source (31, 41, 51, 61) for communication with an external machine (26) that is connected to the interface. A freely configurable interface is used in the welding device, said interface being configured and activated via software directly via the operating panel of the welding power source or an external device that can be connected to the welding power source. A parameter of the welding device is allocated to an internal and/or external connection (101...105). Optionally, a link is established by stringing together graphic symbols of the software. The invention further relates to a welding power source (31, 41, 51, 61) and to a computer program which are set up for performing or storing the process according to the invention.

The invention relates to an image segmentation method and device, a readable medium and electronic equipment, and the method comprises the steps: obtaining the prediction object center point feature information of each pixel point in a to-be-segmented image, the prediction object center point feature information being used for representing the credibility that the pixel point is an object center point; determining center point position information of an object in the to-be-segmented image according to the predicted object center point feature information of each pixel point; and performing image segmentation on the to-be-segmented image according to the central point position information. According to the technical scheme, the central point position information of the object is considered,so that the area where the object is located is more prominent in the to-be-segmented image, the difference between the area where the object is located and the background area is more remarkable, the foreground object can be segmented more accurately when the to-be-segmented image is subjected to image segmentation, interference of the background area is effectively avoided, and the image segmentation accuracy and the segmentation effect are improved.

A kind of novel playing cards features that its clubs, spades, hearts and diamonds can be expressed by different 4 colors, its patterns may be stereoscopic, and its numerals can have different type-faces or the lines with different thicknesses.

The invention provides a light self-insulation prefabricated slab system. The system is characterized in that a constant-size light self-insulation prefabricated slab is covered between two adjacent purlins and is vertical to the purlins; the spacing between the two adjacent parallel purlins is matched with the length of the constant-size light self-insulation prefabricated slab; one end of the constant-size light self-insulation prefabricated slab is fixedly put on one purlin and the other end thereof is fixedly put on the adjacent purlin; gaps are reserved between the constant-size light self-insulation prefabricated slabs which are put on the same purlin, are arranged in the direction vertical to the purlins and are arranged oppositely; and the adjacent constant-size light self-insulation prefabricated slabs are butted with each other in parallel in the direction parallel to the purlins. The system is convenient to transport and maintain, is suitable for roofs or wall surfaces with complex load conditions, and has strong wind resistance and high reliability, thus improving the rigidity of the whole roof system.

The embodiment of the invention provides a classification model training method and device, computing equipment and a storage medium. The method comprises the steps: inputting a training sample into a classification model, and obtaining the prediction probability distribution of the training sample belonging to each class; and calculating target loss and a penalty term, wherein the penalty term is used for indicating the dispersion degree of the negative class in the prediction probability distribution. The sum of the target loss and the penalty term is recorded as total loss, and classification model parameters are updated according to the total loss. In traditional model training, due to the limitation of a loss function, the accuracy of the model on negative class prediction is ignored. The dispersion degree of negative class distribution is used as the penalty term to be introduced into the model loss to construct a novel loss function, the performance of the original loss function is improved, and the prediction capability of the model is improved.

The present invention relates to dry sea cucumber product with mark and its production process. The dry sea cucumber product has uneven mark, which is obvious, fast and hygienic. The present invention makes it possible for the consumer to distinguish different dry sea cucumber products, especially some famous ones, easily.

The invention relates to the field of vehicle-mounted voice control, and particularly discloses a user-defined vehicle-mounted voice skill system and method. A user-defined voice skill training engine, a user-defined voice skill execution engine, a user-defined voice skill script management public module and a user-defined voice skill display public module are arranged; customized vehicle-mounted voice skills are realized, the pursuit of vehicle-mounted voice individuation of people is met, and the sense of participation of users and the product experience are improved. Operation actions of a user are recorded through a voice training process, and the actions are converted into a unified script format, so that voice customization is not limited to TTS broadcast texts or vehicle control instruction texts, and voice control instructions have more selections.

The invention discloses an end-to-end electric energy transaction auction method based on an evaluation mechanism, and the method comprises the following steps: (1) building a multi-attribute comprehensive evaluation system suitable for a distributed power generation market from three dimensions: electric energy transaction reliability, transaction stability and transaction economy, and evaluatingthe power supply service level of each user; (2) determining evaluation coefficients of the users according to the historical energy supply service data of the users; and (3) enabling an end-to-end electric energy auction model solving algorithm based on market evaluation to give a transaction clearing strategy in each time period. According to the invention, an end-to-end electric energy transaction auction strategy between microgrid users based on market evaluation is provided, credit and qualification constraints are provided for bidders in the electric power market, the bidders with goodevaluation are enabled to be advantageous, better income is obtained, and virtuous cycle of the electric power market is promoted.

The invention provides an apple grading method and system based on a residual network. The apple grading method comprises the following steps of: acquiring an appearance image of an apple; and obtaining a grading result of the apple according to the obtained appearance image and a preset apple grading network model. The Apple grading network model is obtained by training an improved residual network; and specifically, the residual network is improved by adding a convolutional attention module and a leaky rectified linear unit. Apple grading based on appearance is realized through the improved residual network. An Otsu method is used to segment an apple image on an HSI color channel, a convolutional attention module and a Leaky ReLU activation function (a leaky rectified linear unit) are added to improve a residual network, the residual network is applied to apple grading, the grading result is compared with grading results of other convolutional neural networks, and it is proved that the improved residual network can better achieve apple grading.