137results about How to "Deepen the depth" patented technology

The invention discloses a video enhancement and transmission method based on deep learning. Downsampling is carried out on a high-definition video at a video source end, thereby obtaining a low-definition video; the low-definition video is compressed in an existing video coding mode; and the compressed low-definition video is transmitted. The size of the video data is greatly reduced through downsampling and video coding, so the video traffic needing to be transmitted is correspondingly reduced, and an effect of reducing the bandwidth cost is achieved. At a user receiving end, a user receives the low-definition video, reconstructs the low-definition video by employing a super-resolution image reconstruction method of the deep learning and restores the low-definition video into a high-resolution video for the user to watch, so the video transmission bandwidth cost is effectively reduced. According to the method, the video transmission bandwidth cost is reduced by at least 50%; the resolution of the live broadcast video is improved; and the watching experience of a user is improved.

The invention relates to a hyperspectral image classification method based on an auto-encoder and a 3D deep residual network. Compared with the prior art, the defect that remote sensing hyperspectralimage classification is difficult to carry out is overcome. The method comprises the following steps: obtaining a training sample; preprocessing the hyperspectral remote sensing image data to be trained; building and training a stack auto-encoder neural network model; constructing and training a 3D deep residual network; acquiring a hyperspectral remote sensing image to be classified; carrying outpreprocessing and dimensionality reduction of the hyperspectral remote sensing images to be classified; and obtaining a hyperspectral remote sensing image classification result. According to the method, a stack auto-encoder neural network model is built, dimensionality reduction is carried out on an original hyperspectral remote sensing image, and redundant information is eliminated; a residual error network module is introduced through a designed 3D convolutional neural network to properly increase the depth of the network, a 3D deep residual error network is established, space-spectrum joint information of a hyperspectral remote sensing image is extracted more effectively, and the problems of gradient disappearance and network degradation are avoided.

In accordance with the present invention, a light-emitting element mounting board is provided including an enamel layer which covers the surface of a core metal, wherein reflective cup portions for mounting light-emitting elements are provided on one face of the board, and wherein a heat radiating part is formed by at least partially removing the enamel layer on the other face of the board, and exposing the core metal. Further, a light-emitting element module including light-emitting elements mounted thereon, a lighting device including the light-emitting element, a display device, and traffic signal equipment are provided.

The invention provides a behavior recognition system and method based on a multi-convolution kernel residual error network, a medium and equipment, and the method comprises the steps: collecting the motion physiological data or environment state data of a user, and sequentially carrying out the transmission, preprocessing and marking of the collected data to obtain a training set and a test set; constructing a behavior recognition model based on the multi-convolution kernel residual error network; classifying the training samples according to the prediction error of the training set and carrying out self-adaptive weight distribution on each class; calculating an overall loss function and partial derivatives of each parameter of the model, iteratively updating the model parameters to obtaina trained recognition model, testing by using the test set, storing the model parameters reaching the test standard, and generating a final deep learning model; inputting to-be-recognized data into the deep learning model to obtain a user behavior recognition result. According to the invention, multi-scale data features can be effectively utilized.

The invention discloses a focal plane polarization image denoising method based on a deep neural network. The method comprises the following steps: creating a polarization image data set; designing adeep neural network PDRDN, wherein the network learns a residual image between the noise image and the noiseless image; optimizing the network parameters by using a customized loss function until themodel parameters are stable; and inputting an image to be denoised into the trained network to obtain an output residual image, and then predicting a clean image by using the noise image and the residual image. The method can achieve a better denoising effect, is used for improving the quality of a polarization image, and reduces the impact on polarization information from noise.

The invention provides a high-speed PB tube single screw, which comprises a feeding section, a separation-type melting section and a mixing-type extruding section, wherein the separation-type melting section is provided with a main screw edge and two auxiliary screw edges, the depth of a screw channel of the feeding section is 7mm to 8mm, the lead of the main screw edge is 105 to 110 mm, and the lead of the auxiliary screw edges is respectively 107 mm and 84 mm. The high-speed PB tube single screw solves the phenomenon of the prior PB tube screw that the productivity is difficult to be improved at the high speed, the plasticized quality is poor and the temperature of the fuse is too high, and can obtain the PB melting plastics with good plasticized quality and high productivity (300kg/hour).

The invention provides a detecting method for residual stress of a steel structure. Furthermore improvement is realized based on an original hole drilling method. Specific parameters in residual stress of the steel structure are further optimized. Through accurately setting parameters such as hole diameter, hole depth, hole distribution condition, hole inclination angle and distance between holes and a strain gauge, more accurate testing structure is realized. Particularly through researching the inclination angle of each hole, a defect of drilling only vertical holes in traditional technology is overcome. Through drilling oblique holes which incline in an angle range, stress change at the periphery of the drilling hole can be more accurately acquired. Furthermore hole depth is increased as possible. Depth in the vertical direction is reduced. Damage of a tested component is reduced.

The invention relates to the field of stereo garage devices, in particular to a cylinder-type stereo garage. The cylinder-type stereo garage comprises a cylindrical garage shaft, an entrance gate house system arranged on a shaft mouth of the top of the garage shaft, more than two layers of annular parking layers vertically distributed along the garage shaft, a lifting rotation platform arranged at the center of the garage shaft and used for fetching vehicles and a traction mechanism for driving the lifting rotation platform to vertically lift. The cylinder-type stereo garage is reasonable in structure, stable to operate, efficient and stable and is suitable for fetching vehicles.

The invention provides a thermal power plant condenser vacuum degree prediction method based on multilayer LSTM. Firstly, a sample data set is constructed, input data and output data in a sample are subjected to standardization processing respectively, then the input data sequentially enter a multi-layer long-short-term memory neural network, and model training is performed by using an adaptive moment estimation algorithm; and finally, to-be-predicted data is input into the trained prediction model, and the vacuum degree of the condenser is predicted. According to the method, the long-term andshort-term memory neural network structure is applied to condenser vacuum degree analysis, deep mining of big data is achieved, the accuracy and speed of data prediction are improved, and use of system resources is optimized. According to the two-layer LSTM structure designed by the invention, the model depth can be increased, the model capacity can be increased, the prediction precision can be improved, and meanwhile, the adaptive moment estimation algorithm is used for optimizing the traditional gradient descent algorithm, so that the problem of gradient explosion is avoided.

The invention discloses a thermal treatment method of a spline shaft head of a transmission shaft. The method is characterized by sequentially comprising a normalizing process and a medium-frequency induction quenching process, wherein the normalizing process sequentially comprises two steps of: (I) heating the spline shaft head by a push disc-type heating furnace, wherein the heating temperature is 850+/-10 DEG C, and the heating time is 120 minutes; and (II) performing air cooling for 2-3 hours; the medium-frequency induction quenching process comprises two steps of: (I) quenching, wherein the surface quenching area of the spline shaft head comprises a spline part and a full-oil seal table part, and the depth of the hardening layer is 4-6mm; and (II) low-temperature tempering. Compared with the same kind of hardening and tempering and medium-frequency induction quenching technology (needing tempering at a high temperature of 680+/-10 DEG C for 150 minutes), the method disclosed by the invention greatly saves resources on the premise of guaranteeing the static torsional strength and fatigue life of the spline shaft head, reduces the production cost, has positive social significance of low carbon and environmental protection, and remarkably improves the product competitiveness.

The invention discloses a vehicle multi-attribute detection method based on single-network multi-task learning. The method comprises the following steps of collecting and screening pictures; making adata set; carrying out network design on the basis of a Darknet deep learning framework, designing a network structure by adopting an end-to-end and one-stage non-cascade mode according to the characteristic of multiple attributes of the vehicle, and constructing a network model; carrying out model training, setting and adjusting model parameters, training a vehicle multi-attribute data set according to a designed network model, and carrying out data enhancement and multi-scale training during training; and carrying out model testing and model evaluation. The Darknet-based deep learning framework platform is designed according to the Darknet-based deep learning framework platform; a network model is built and is of an end-to-end one-stage non-cascade structure, the network improves the detection effect of multiple attributes of a vehicle by adopting the technologies of data enhancement, convolution kernel separation, multi-scale feature fusion and the like, and better real-time performance is achieved while the higher detection accuracy and recall ratio are achieved.

The invention relates to a filter. The filter comprises a filter body and an external cleaning circulating device, wherein the filter body comprises a filter tank body, a filter material, an upper pipeline and a lower pipeline; the filter material is arranged in the filter tank body, the upper pipeline comprises a liquid inlet pipe, the lower pipeline comprises a water outlet pipe, the external cleaning circulating device comprises an external cleaning device, a backwashing feeding pipe, a backwashing water outlet pipe and a backwashing pump, and the external cleaning circulating device is connected with the filter body. According to the filter, as the external cleaning circulating device is arranged, the filter material can be cycled, back-washed and regenerated thoroughly with high strength, and the difficulties that the filter material of the filter equipment has short service life and low filtering precision due to backwashing regeneration difficulty and poor anti-contamination capability of the filter material can be solved, the backwashing strength can be improved, and a problem of the prior art that the liquid containing sewage difficult to be separated from the filter material cannot be filtered is solved.

The invention provides an escalator safety monitoring method based on YOLOv3, and the method comprises the following steps: 1, intercepting a picture with a passenger taking an escalator in a historical escalator monitoring video, marking a character part in the picture as a target region, and making a data set 1 according to a PASCAL VOC data set format; classifying according to the posture of the figure in the target area, and making a data set II according to a PASCAL VOC data set format; 2, training a YOLOv3 network model 1 by using the data set 1; Using the second data set to train a second YOLOv3 network model; 3, real-time safety monitoring is conducted on the escalator, and position coordinates of a target area are output through a YOLOv3 network model I; and the target area picture is the input of the YOLOv3 network model II to identify the postures of the passengers on the escalator. According to the technical scheme, the problems that a model is too complex and the real-timemonitoring requirement cannot be met in a traditional method for recognizing pedestrian actions on the escalator are solved.

The invention discloses a heavy metal contaminated soil remediation method comprising the steps of adding alkaline pH conditioning agents to heavy metal contaminated soil, and then plowing; ditching on plowed soil, applying enzymatic biological organic fertilizer, calcium magnesium phosphate fertilizer, silicon calcium fertilizer, urea and potassium chloride to the bottoms of the ditches, fillingsoil to cover the fertilizer, and then planting sugarcane seedlings; additionally applying the urea, phosphate fertilizer and potassium sulphate after entering the sugarcane tillering and jointing stage; after the sugarcane of the current year matures, harvesting the overground part of the sugarcane of the current year, reserving the underground part of the sugarcane, and then planting the sugarcane of the next year; ditching on both sides of the underground part of the sugarcane, applying the enzymatic biological organic fertilizer, the calcium magnesium phosphate fertilizer, the silicon calcium fertilizer, the urea and the potassium chloride; and after the sugarcane of the next year matures, repeating the planting scheme of the sugarcane of the next year for planting the sugarcane of thethird year. By adopting the remediation method disclosed by the invention, the remediation effect of the heavy metal contaminated soil can be improved.

The invention provides a segmentation method, device and apparatus and a storage medium. The method comprises the steps of receiving a heart image to be segmented; processing the heart image to be segmented by using the trained hierarchical residual neural network; wherein the hierarchical residual neural network comprises at least one residual layer and at least one full convolution layer, the atleast one residual layer is used for performing encoding processing on an image training set, and the at least one full convolution layer is used for performing decoding processing on an encoded image output by the residual layer so as to output a segmentation decision value of a heart image to be segmented; and determining a contour map of the right ventricle in the heart image according to thesegmentation decision value. The method uses the coding-decoding structure of the hierarchical residual neural network to determine the contour map of the right ventricle in the cardiac image and solves the problem that the boundary of the right ventricle is fuzzy in the heart image, and can accurately draw the boundary contour of the right ventricle.

The invention relates to a tri-tube tri-high-pressure jet grouting pile construction method, and belongs to the field of constructional engineering. The tri-tube tri-high-pressure jet grouting pile construction method comprises the following steps: a first step of measuring, paying off and positioning, wherein measurement is performed according to a designed construction drawing, and a construction axis is paid off; a second step of guiding a drilling machine to be in place, wherein the drilling machine is mounted on a designed pile position, and a drill bit of the drilling machine is alignedto the center of the pile position; a third step of drilling, protecting walls and forming holes, wherein the drilling machine is used for drilling holes, and the walls are protected with mud until hole depth achieves designed elevation; a fourth step of separately forming a mud outlet, a water outlet and a gas outlet in the bottom end of a grouting rotary-jetting tube, wherein a pressure value ofair sprayed out by the gas outlet ranges from 1.6 MPa to 2.2 MPa, the water outlet is positioned above the mud outlet and the gas outlet, and the water outlet in the rotary-jetting grouting tube is drilled for washing and reaming while the jetting grouting tube carries out grouting; a fifth step of flushing; and a sixth step of moving the machine. A relatively great gap is formed between the grouting tube and the hole wall, so that pressure in holes is conveniently released, and high water pressure is reduced.

The invention discloses a double-focusing field electromagnetic tweezer implementation method based on a time reversal spatial and temporal focusing mechanism, and belongs to the technical field of electromagnetic wave non-contact operation and control. According to the implementation method, a TR electromagnetic wave spatial and temporal focusing transmission mechanism is used as foundation, and two parallel focusing spots are generated through a planar time reversal mirror; and a hollow region with extremely weak electromagnetic field strength generated by an electromagnetic interference cancellation mechanism is generated between the two focusing spots to form an electromagnetic field gradient trap to generate required thrust and gradient force, so as to catch or control tiny particles with electric neutrality or weak current. Compared with a single-spot electromagnetic wave focusing field, the double-focusing field can remarkably lower scattering force, improve the gradient force of the electromagnetic field trap, and overcome the shortcomings of relatively weak dynamic action of the electromagnetic wave at frequency band of less than optical frequency band; in addition, the trap depth of the electromagnetic field can be increased, and the stability of the electromagnetic field at frequency band of less than optical frequency band in catching and controlling particles can be improved; and in addition, a large amount of particles with electric neutrality or weak current can be constrained in the narrow and long hollow region at the same time.

The invention relates to a retractable cup saucer which comprises a vertically retractable cup saucer body and a fixing assembly. The fixing assembly is used for fixing the cup saucer body onto an outer plate and comprises an upper fixing plate, a lower fixing plate and a collar, the collar is positioned between the upper fixing plate and the lower fixing plate, and the cup saucer body is nested in the collar; the edge of an opening in the upper end of the cup saucer body is turned out to form a flange; elastic parts and a boss are arranged on the inner wall of the collar, and the relative retraction position of the cup saucer body can be constrained by the elastic parts and the boss. Compared with the prior art, the retractable cup saucer has the advantages that the placing depth of a cup can be effectively increased, so that the placing stability of the cup can be improved, the cup is not easy to topple over, and the retractable cup saucer not only is generally applicable to dining tables, but also is applicable to commercial office tables with requirements; the cup saucer extends downwardly, so that an effective service space can be saved for a user, and the cup saucer can be collected when not used; the retractable cup saucer is fixed by the aid of rivets and accordingly is firm; components of the retractable cup saucer are simple in structure, and accordingly additional casting moulds can be omitted.

The invention aims to provide a liquid soap bottle. By adoption of the liquid soap bottle, liquid soap can be conveniently sucked out as much as possible, and waste of the liquid soap is reduced. Theliquid soap bottle comprises a bottle body, a pressing liquid outlet mechanism and a liquid suction pipe arranged inside the bottle body. The pressing liquid outlet mechanism is arranged at a liquid outlet of the liquid suction pipe. The liquid soap bottle is characterized in that the liquid suction pipe is made of a flexible material, and a load-bearing part is arranged at the bottom end of the liquid suction pipe; a liquid suction opening of the liquid suction pipe is located in the load-bearing part; and the length of the liquid suction pipe is larger than or equal to the distance from thetop end of the liquid suction pipe to any side of the bottom of the bottle body. At the last using stage, due to the small amount of the liquid soap, the depth of the liquid soap is small after the liquid soap is flatly laid at the bottle bottom. At the moment, the bottle body can be tilted, the liquid soap remaining inside the bottle can be gathered to one corner of the bottom of the bottle body,and therefore the depth of the liquid soap can be increased, the load-bearing part carries the liquid suction opening to be sunken into the bottom of the liquid soap automatically so as to suck the liquid soap, the liquid soap can be sucked as much as possible, and waste is reduced.