55results about How to "Keep the original information" patented technology

The invention provides an unmanned aerial vehicle airborne LiDAR point cloud filtering method based on adaptive gradient. The method comprises the following steps: obtaining point cloud data in a test region and carrying out filtering on the point cloud data; 2) establishing index for the point cloud data; 3) carrying out calculation and keeping an optimal plane in a grid; 4) selecting an elevation lowest point in a preset window as an initial seed point of a ground point; 5) calculating the distance between the optimal plane and the ground point and gradient from the ground point to the center of the optimal plane, and obtaining a ground point set according to gradient threshold; 6) serving a point in the initial ground point set as a base point, and generating a digital terrain model through an inverse distance weighted interpolation method; and 7) through the newly-generated digital terrain model, obtaining update value of the gradient threshold, repeating the steps 4)-6) and obtaining a final ground point set. The filtering method based on adaptive gradient improves algorithm operation efficiency.

The invention discloses a parking payment system, which is characterized in that a vehicle owner terminal registers on a parking platform, logs in a platform account and opens an ETC payment functionof a specified license plate number; a toll collection system comprises a license plate recognition camera and a control server, the license plate recognition camera monitors the motion condition of avehicle on a lane, captures a vehicle head when finding that the vehicle is about to enter a parking area, acquires a vehicle image, performs image recognition on the vehicle and a license plate, generates a license plate number and sends the license plate number to the control server; the control server calculates the parking duration according to the vehicle parking time and generates a parkingfee payment order, and the control server sends the license plate number and the payment order to the parking platform for ETC payment; and the parking platform matches the license plate number withthe platform account, judges whether the platform account corresponding to the license plate number opens an ETC payment function, and judges whether the license plate number is in an ETC blacklist.

The invention provides a multi-focus image fusion method using morphology and a free boundary condition active contour model. The method comprises the following steps: (1), constructing an initial definition distribution diagram of an image by using a gradient feature; (2), carrying out calculation by using the initial definition distribution diagram to obtain a rough definition distribution diagram ad then determining a final definition distribution diagram; (3), carrying out processing on the final definition distribution diagram based on small-area removing operation of the morphology and open-close operation of the morphology to obtain an initial fusion decision image; (4), extracting a boundary, serving as an initial value of a free boundary condition active contour model, between a focusing area and an out-of-focus area from the initial fusion decision image; (5), obtaining a boundary image based on the free boundary condition active contour model and obtaining a final fusion decision image according to the boundary image and the initial fusion decision image; and (6), on the basis of a decision image of multi-focus image fusion and a set fusion rule, generating a final fusion image with all clear parts. THE method can be widely applied to various image processing application systems.

The invention relates to a novel noise image fusion method based on a cycle spinning (CS) technology and Contourlet domain context hidden Markov model (CS-CT-CHMM). First, a source image containing a certain level of Gauss white noise is de-noised by use of the CS-CT-CHMM, wherein the structure of the context is calculated based on neighborhood entropies of a father node, two nearest cousin nodes and a current node of a Contourlet decomposition coefficient; then, Shearlet transform is performed on the de-noised image, and a fusion rule is designed by adopting a weighted average fusion strategy for low-frequency sub-band coefficients and adopting an improved pulse-coupled neural network (IPCNN) model for high-frequency sub-band coefficients; and finally, reverse Shearlet transform is performed to obtain a fused image. The CS-CT-CHMM de-noising method can protect detailed information of the image and suppress the pseudo Gibbs effect while effectively removing noise, while the fusion method combining Shearlet transform and IPCNN is advantageous in terms of image contrast enhancement and the amount of information. Compared with the traditional fusion method, the quality of the fused image is greatly improved.

Provided is a panorama image splicing method based on edge vertical distance matching. The method comprises the following steps: (1), extracting the edges of two images by use of a CANNY algorithm; (2), screening a matched pixel column; (3), performing matching according to variances of edge pixel ordinates; (4), performing grouping according to variance values and reflected image relative positions; (5), performing the edge vertical distance matching on each group, and determining that a smallest group distanced from statistical values is an image splicing position; and (6), performing image stitching processing. According to the invention, through matching the edge vertical distances of two neighboring images, automatic splicing of a panorama image is realized. A matching position is screened in advance so that computing amount is reduced, computing time is saved, and computing precision is improved. A vertical distance coupling algorithm can help to find out an optimal seam, i.e., the seam position is disposed at a position with smaller information content in an image, and an information key position is avoided as likely as possible; and an original image is completely reserved in most area of a splicing image so that information of the original image can be reserved to the maximum.

A method used for extracting an unearthed bamboo slip and silk character pattern image and based on digital image processing belongs to the technical field of computer image processing of unearthed bamboo slip and silk. The method is characterized in that Photo shop digital image processing software is used for extracting a digital image of the unearthed bamboo slip and silk, and steps of capturing of single character pattern, re-distribution of gray level between each character stroke ink mark and a bamboo slip and silk background according to original ratio, distinguishing of radical selecting area of a single character and a stroke selecting area in each radical by using an original basic stroke table and a basic radical table area, reserving of all stroke images in each radical selecting area and deleting of non-stroke ink mark image are conducted sequentially, and extracting of a single character pattern is achieved. By means of the method, basic radical separation is used as a basis, thorough separation of the unearthed bamboo slip and silk character patterns and a bamboo slip and silk background is achieved, character pattern images which reserve character original information and are clear and distinguishable are presented, and the technical problem of great distortion of images caused by copying the unearthed bamboo slip and silk through a traditional writing method at present is solved.

The invention provides a credit card simulation test method under multiple transaction channels. The method comprises the following steps: receiving credit card information configured on a preset configuration interface; generating a simulation message according to configured credit card information; determining a target transaction channel and a target communication strategy according to a presetrule; packaging the analog message according to the target communication strategy to obtain a response message; obtaining a test case set corresponding to the target transaction channel, wherein thetest case set is a test case library formed after desensitization processing according to credit card transaction data in a production environment; performing simulation test on the response message according to the test case set; and outputting a test result. The invention also provides a credit card simulation test device under multiple transaction channels, a terminal and a storage medium. According to the invention, a plurality of credit card transaction channels and communication strategies under different credit card transaction channels are configured through the automatic test platform, and simulation of credit card transaction information under various transaction channels can be supported at the same time for the first time.

The invention discloses a method for detecting a P wave and T wave in an ECG signal based on Gaussian function fitting. Firstly, baseline drift removal is carried out on an original signal, and then according to QRS wave group labeling of the existing ECG signal, the signal is divided into a series of single heartbeat signals; secondly, for each heartbeat cycle, search intervals where the P wave and the T wave roughly and possibly appear are divided respectively through approximate ranges of the P wave and T wave, the wave crest direction and suspected wave crest appearance point of each search interval are determined through numerical integration; according to the information, a Gaussian function is used and appropriate initial parameters are configured for fitting; finally, the wave crests of Gaussian function fitting results and starting and stopping points of the wave crests are calculated and taken as the wave crests and starting and stopping points of the P wave and T wave of a current heartbeat cycle. The method for detecting the P wave and T wave in the ECG signal based on Gaussian function fitting has the advantages that the interference of noise signals is effectively avoided, and the robustness of an algorithm is enhanced, so that the accuracy of the detection of the P wave and T wave is improved.

The invention discloses a detection method for odor substances in exhaust gas of the perfume and essence industry. According to the invention, exhaust gas discharged by the perfume and essence industry is directly or indirectly acquired, then solid-phase micro-extraction (SPME) or solid-phase adsorption-thermal desorption technology are employed for extraction of a sample; and extraction temperature and extraction time are adjusted and GC-MC is cooperatively used for analysis of the exhaust gas sample. The detection method is high in efficiency, simple to operate and applicable to analysis of odor substances with any concentration. An airbag method and negative pressure sampling technology are cooperatively used, an appropriate sampling point is selected and a representative gas sample is acquired, so original information of the sample is maintained to a greatest extent. A sampling bag has good airtightness, and a sample in the bag can be maintained for more than one month. SPME pretreatment technology is employed, so the advantages of greenness, rapidness, no pollution and repeated usability can be obtained. The detection method is applicable to both exhaust gas discharged in the perfume and essence industry and to detection of exhaust gas of industries like petroleum, the chemical industry, food, medicine and cosmetics.

The invention discloses an attribute-level sentiment analysis method based on a hierarchical attention mechanism and a gate mechanism, and aims to make full use of the relationship between contexts and attribute words to make the contexts and the attribute words fully associated. Highlighting the importance of different words in the context and the attribute words so as to improve the attribute-level sentiment analysis precision; the method enriches information in context and attribute word representation, and adopts the scheme that comment corpora are preprocessed, and a context word embedding matrix and an attribute word embedding matrix are obtained through GloVe word vector indexing; inputting the context and the attribute words into the GRU to obtain a context hiding state and an attribute word hiding state; obtaining a context vector representation 1, an attribute word vector representation and a context vector representation 2 through self-attention; splicing to obtain an overall vector representation, obtaining distribution of sentiment polarities, corresponding to contexts, of attribute words through a classifier, analyzing differences, and adjusting parameters in the model; the method belongs to the field of sentiment analysis in natural semantic processing.

The invention relates to a method for rapidly predicting macromolecule energy gap. The method is based on a first principle in combination with a support vector machine, and comprises the following steps: looking up the structure of macromolecule and an experiment value of an energy gap from literatures; computing the energy gap according to different first principle methods, and selecting an optimal method; looking up 4 module macromolecule structures of repetitive units from the literatures and optimizing the 4 module macromolecule structures; computing the energy gap of the 4 module macromolecules; generating a descriptor through Dragon software, and dividing data into a modeling set and a testing set; performing variable screening through maximum correlation minimum redundance algorithm, and establishing a model through the support vector machine; rapidly predicating the energy gap of the macromolecules to be detected through the established model. The method is based reliable literature data and modeling methods, and the predicating model for the macromolecule energy gap has the advantages of being convenient, rapid, low in cost and free of pollution.

According to the pavement use performance weight analysis method based on the modified analytic hierarchy process, firstly, all single items are subjected to importance degree overall sorting, the importance degrees of adjacent indexes are compared in sequence, then a judgment matrix is constructed, the judgment matrix of all the indexes is calculated, and the weights of all the single items of indexes are obtained. The judgment matrixes obtained through the improved analytic hierarchy process are all consistency matrixes, and the situation that consistency inspection is not passed is avoided.The result shows that the importance degree of each single index in experts in different technical fields is comprehensively evaluated by analyzing the weight of the pavement use performance index byadopting an improved analysis hierarchy analysis method, and the obtained weight result is more accurate, comprehensive and reliable, and more conforms to the actual situation of the weight of the analyzed local pavement use performance index.

The invention relates to an automatic liquid feeding method for dyeing processes of multi-component dyes. In the dyeing process of the multi-component dye A, concentrations of all the dyes in dye liquid are monitored online and real-timely; after the percentage of the mass of each dye to the total mass of the dyes is calculated according to the concentrations of all the dyes in the dye liquid, thepercentages of all the dyes are input into the same BP neural network model simultaneously, K0 values of all the dyes are output by the BP neural network model, and an absolute value of the difference obtained by subtracting the minimum value from the maximum value of the K0 values is D<max>; the change rate of D<max> relative to D<max><0> is calculated, the change rate is compared with a threshold value, and if the change rate is smaller than the threshold value, no dye liquid is replenished; otherwise, the dye liquid is replenished till the concentrations of all the dyes are equal to initial concentrations. The automatic liquid feeding method is simple, by monitoring the concentrations of all the dyes in the dye liquid online and real-timely, the dyeing result approaches to the target effect to the utmost extent, on-line feedback control is achieved, the quality of dyed products is improved, the dyeing one-time success rate is increased, and the better dyeing effect is achieved.

Embodiments of the present invention provide an image processing method, a processing device, and processing equipment, which implement image conversion by using a generated neural network in combination with image content features, so that the converted output image includes both the conversion features and the consistency with the input image. The image processing method includes: acquiring an input image; performing image conversion processing on the input image according to the input image by using a generated neural network to output the converted output image, wherein the generated neural network includes a plurality of processing levels, wherein The output result of the i-th processing level is input to the i+1th processing level and the jth processing level, and the jth processinglevel further receives an output result of the j-1th processing level, and the output result of the j-1th processing level is The output of the i-th processing level has the same size, where: i is less than j-1, and i and j are positive integers.

The invention discloses a method for extracting a modal parameter from viscous damping vibration signals. The method for extracting the modal parameter from the viscous damping vibration signals comprises the steps that firstly, Nyquist uniform sampling is conducted on the vibration signals; secondly, an autoregression matrix equation is established through sampled signals, the least square solution of the autoregression coefficient vector of the autoregression matrix equation is worked out, a Prony polynomial is established with all the elements of the least square solution as coefficients, and the polynomial is solved so that the solution vector of variables can be obtained; thirdly, the solution vector is corrected, and an inherent frequency vector containing a false modality and an inhehrent damping ratio vector are worked out according to the corrected solution vector; fourthly, a matrix is established with the corrected solution vector as the basis, sparse optimization solution is conducted on the projection equation on the matrix through the sampled signals, and then a sparse vector is obtained; finally, the false modality is removed through the sparse vector, wherein all the nonzero elements in the sparse vector are vibration mode coefficients. The method for extracting the modal parameter from the viscous damping vibration signals has the advantages that the noise immunity is high, and the extracted modal parameter is high in precision and high in stability.

The invention relates to a single image rain removing method and device based on dark channel and fuzzy width learning. The method comprises the steps: S1, carrying out defogging preprocessing on an original image; S2, carrying out high-low frequency separation on the image after defogging pretreatment, taking a high-frequency part to carry out color space conversion, and converting the RGB colorspace into a YCbCr color space; S3, taking the Y channel of the YCbCr color space corresponding to the image of the training set part as the input of fuzzy width learning to carry out model training;S4, taking the Y channel of the test set part as the input of fuzzy width learning to obtain a rain-free graph of the Y channel; S5, combining the high-pass layer rain removal graph with the low-passbase layer to obtain a preliminary rain removal effect graph; and S6, based on the defogged and preprocessed image, carrying out optimization processing on the preliminary rain removal effect pictureto obtain a final rain removal effect picture. Compared with the prior art, the method has the advantages of high color rendition degree and the like.

The invention relates to a phase attention mechanism network model-based circuit breaker residual life prediction method, which comprises the following steps of: firstly, acquiring a vibration signal in an opening process, then optimizing a VMD algorithm, decomposing the vibration signal by using the optimized VMD algorithm, and selecting a modal component with relatively high kurtosis for reconstruction; then, according to the energy-entropy ratio, a contact breaking vibration segment is extracted from the reconstructed vibration signal; and finally, a prediction model fusing a stage attention mechanism is established, the prediction model takes a one-dimensional convolutional neural network and a GRU network as a trunk network, the stage attention mechanism is divided into two stages, the first stage is a distributed attention mechanism applied to the one-dimensional convolutional neural network, weighting is performed on an input sample in time and feature dimensions, and the second stage is a distributed attention mechanism applied to the GRU network. And in the second stage, weighting is carried out on the time dimension again by applying a time step attention mechanism of the GRU network. According to the method, the contribution degree of important information on the time dimension and the feature dimension to the prediction result is enhanced, and the prediction precision is improved.

The invention discloses an automatic driving 3D modeling method, device and system. The invention relates to the technical field of automatic driving, and solves the problems that data modeling is incomplete due to single data sampling, or the relevance between data is poor, and an accurate 3D environment model cannot be created. According to the invention, the method comprises the steps of: obtaining and fusing data collected by the multi-element sensor to form a composite view, extracting inter-frame target parameters of the same detection target in a plurality of adjacent frames acquired bydifferent sensors in the comprehensive view; comparing and analyzing data acquired by different sensors; adopting a stepped threshold method to eliminate singular values and redundancy of data in different sensors, achieving accurate judgment of target parameters, and when the inter-frame target parameters of the same detection target in a plurality of continuous adjacent frames are within a preset threshold, determining the inter-frame target parameters to be inter-frame correlation dynamic data so as to construct a 3D environment model. According to the method, more and more accurate associated dynamic data can be extracted, and the establishment of an accurate 3D environment model is facilitated.

The invention discloses an image processing method and device, a medium and equipment, relates to the field of artificial intelligence, and is used for removing a moving target in an image, and the method specifically comprises the following steps: extracting original image information of three channels from a blurred image; performing Laplace transformation and mean square error removal calculation on the original image information of the three channels to obtain intermediate image information corresponding to the three channels; averaging the intermediate image information respectively corresponding to the three channels to obtain the ambiguity of the blurred image; and selecting a corresponding deep learning network model to perform deblurring operation on the blurred image according to the blurring degree of the blurred image.

The invention relates to a method for rapidly detecting the potassium content of tobacco leaves based on electronic nose-artificial neural network. The method comprises the following steps: 1, collecting several tobacco leaf samples in different producing areas, preprocessing the tobacco leaf samples, and carrying out electronic nose scanning to obtain electronic nose data of every tobacco leaf sample; 2, detecting the potassium content of every tobacco leaf sample by adopting flame photometry; 3, carrying out dimension reduction on the electronic nose data by adopting partial least squares to obtain the dimension reduction data of every tobacco leaf sample; 4, establishing a rapid forecasting model of the potassium content of tobacco leaves by adopting an artificial neural network algorithm with the dimension reduction data of every tobacco leaf sample as an independent variable and the potassium content of every tobacco leaf sample as a dependent variable; and 5, rapidly forecasting the potassium content of a tobacco leaf kind to be detected according to the established rapid forecasting model of the potassium content of tobacco leaves and the electronic nose data of the tobacco leaf kind to be detected. The method has the advantages of simplicity, rapidness, low cost, comprehensive and accurate data, no pollution and simple test.

The invention discloses an image processing method and device, equipment and a storage medium. The method comprises the steps of performing down-sampling on an original high-resolution image based on wavelet transform to obtain low-frequency information and high-frequency information; inputting the low-frequency information into a target inverse neural network to obtain a low-resolution image; inputting the high-frequency information into the target inverse neural network to obtain an information distribution diagram; and sending the low-resolution image and the information distribution diagram to a second terminal, so that the second terminal generates a target high-resolution image according to the low-resolution image and the information distribution diagram. Through the technical scheme of the invention, the original information of the image can be better reserved in the down-sampling process, and the low-resolution image is restored by using the information in the up-sampling process, that is, a high-resolution image with more complete information can be restored only by transmitting the low-resolution image and adding the information distribution diagram.

The invention relates to the technical field of computer software, in particular to a heterogeneous software workload estimation method based on deep learning, which comprises the following steps: S1,establishing data sets including a source data set and a target data set; wherein the target data set is a data set owned by the user; wherein the source data set is a data set having heterogeneity with the target data set; S2, training an auto-encoder by using the source data set and the target data set; and S3, extracting data features from the auto-encoder, inputting the data features into a convolutional neural network training predictor, and generating a predicted value of the software workload by using the predictor. The convolutional neural network can automatically extract the high-level meaning of the data, so that the software workload estimation work is more efficient, and resources are saved.

The invention discloses an image thinning and characteristic classification method used for product defect detection and quality control. A pixel of an image is subjected to binaryzation to form a pixel value set area corresponding to the image; end point removal is carried out before an image is skeletonized to eliminate noise influence; an image thinning direction is judged, image skeletonization is carried out, an image is subjected to continuous iteration thinning successively from the upper, lower, left and right pole positions of the image to the inside, and strict image skeletonization is carried out, wherein the strict skeletonization is carried out is characterized in that continuous iteration thinning is carried out successively from the north, the south, the west and the east points of the image. When the image is not thinned or reaches thinning frequencies, redundant end points are removed, and the image is smoothly processed. A round area forms a point after being thinned, and a skeleton keeps original information; the method consumes time like a Davies two-step method; and after an END point is defined, end point interference can be eliminated, and an improved thinning principle is cooperated to guarantee judgment validity.

The invention discloses a label feature near-duplicate video detection method based on convolutional neural network semantic classification, and aims to solve the problems of large feature storage space, low retrieval efficiency and the like in the existing near-duplicate video retrieval field. The method comprises the following steps: firstly, extracting dense semantic classification label features from a video by utilizing a deep convolutional neural network model; removing redundancy according to repeatability among the video frame label features to obtain semantic classification label features of the video; carrying out similarity matching on the feature vectors of the query video and the library video; and finally, measuring the similarity of the two videos by calculating a Jaccard coefficient so as to achieve the detection of the nearly repeated videos, wherein the two steps of video tag feature redundancy elimination and feature matching have two implementation modes of a videolevel and a frame level, namely, nearly repeated video detection based on semantic classification tag features can be achieved through two different levels of methods. According to the invention, near-repetitive video detection can be rapidly realized, and the method has certain robustness for video editing transformation and noise.

The invention discloses a face feature extraction method, which comprises the steps of collecting a face image and a feature extraction network, extracting features by using an initial feature extraction module and a GTFB module, fusing the features by using a bottleneck layer, amplifying a feature map, performing feature transformation by using a feature transformation sub-network and the like. The amplification subnet and the feature transformation subnet are connected to form a network, and redundant useless information is reduced by reusing important information on different scales through the GTFB module, so that information useful for face recognition can be fully extracted. The invention further discloses a low-resolution face recognition method and device. The low-resolution face recognition method comprises the steps of feature extraction, vector matching and the like, the accuracy of low-resolution face recognition is improved by efficiently extracting the face features, and the requirements of actual application are met.

The invention relates to an automatic liquid feeding method for dyeing processes of multi-component dyes. In the dyeing process of the multi-component dye A, concentrations of all the dyes in dye liquid are monitored online and real-timely; after the percentage of the mass of each dye to the total mass of the dyes is calculated according to the concentrations of all the dyes in the dye liquid, thepercentages of all the dyes are input into the same BP neural network model simultaneously, K0 values of all the dyes are output by the BP neural network model, and an absolute value of the difference obtained by subtracting the minimum value from the maximum value of the K0 values is D<max>; the change rate of D<max> relative to D<max><0> is calculated, the change rate is compared with a threshold value, and if the change rate is smaller than the threshold value, no dye liquid is replenished; otherwise, the dye liquid is replenished till the concentrations of all the dyes are equal to initial concentrations. The automatic liquid feeding method is simple, by monitoring the concentrations of all the dyes in the dye liquid online and real-timely, the dyeing result approaches to the target effect to the utmost extent, on-line feedback control is achieved, the quality of dyed products is improved, the dyeing one-time success rate is increased, and the better dyeing effect is achieved.

The invention discloses a GIS state recognition method based on principal component analysis of vibration signals, extracting 14 feature quantities of GIS vibration signals under normal and different fault states to form a 14-dimensional composite feature vector; The vector is compressed and dimensionally reduced into the principal component feature vector; then the corresponding decision function between the vibration signal principal component feature vector and the GIS state is obtained through the two-stage training of the deep belief network, and the principal component feature vector of the GIS vibration signal to be recognized is used. Carry out classification, and determine the GIS status according to the classification. Using principal component analysis to optimize the composite eigenvector not only retains the original information of the composite eigenvector, but also reduces the dimension of the eigenvector, improves the efficiency of the classifier, and effectively improves the accuracy and speed of GIS state recognition.