47results about How to "Accuracy advantage" patented technology

An advertisement filter system comprises a content input interface, a feature analysis module, a decision calculating module, a data recording module, an information base, a command output interface, a manual operation input interface and a machine learning module, wherein the content input interface is used for receiving user generating content from internet interactive products; the feature analysis module is used for analyzing the user generated content, extracting the multiple features of the user generated content and calculating the feature value according to the feature history condition and a manual operation record so as to generate feature vectors; the information base is used for storing various feature data of the user generated content; the decision calculating module is used for comprehensively judging whether the user generated content is filtered or not according to the feature vectors generated by the feature analysis module; the data recording module is used for writing the feature data, categorical data and the manual operation record into the information base; the command output interface is used for arranging a result judged by the decision calculation module into a display / shielding operation command, and synchronizes the display / shielding operation command to the internet interactive product; and the manual operation input interface is used for receiving and analyzing the operation of a filter result modified manually; and the machine learning module utilizes each analysis result and the manual operation record to learn and upgrades the decision calculating module according to the learning.

The invention provides a silicon-based IGBT and silicon carbide Schottky diode hybrid grid driving system. The grid driving system comprises a driving input stage, a logic enabling circuit, a current source circuit, an IGBT grid voltage-to-time change rate detection circuit, an IGBT collector current-to-time second-order differential change rate detection circuit, an IGBT, a Schottky diode D1 and a sampling resistor Rx. All stages of the IGBT starting process are mastered in real time by sampling and detecting the grid voltage VG of the IGBT, the change rate dVG / dt of the grid voltage and the second-order change rate d(dIc / dt) / dt of the collector current, after judgment of the logic enabling circuit and control over the current source circuit in the grid driving circuit, the size of the driving current in the IGBT starting process can be changed, and then the purposes of inhibiting current oscillation to reduce current overshoot and starting loss in the IGBT starting process are achieved.

The invention provides a business quality analysis method and system under a micro-service architecture, and the method comprises the steps: obtaining all micro-service call record data in real time,carrying out the analysis of the call record data, and obtaining the topological relation between a business and all micro-services; obtaining an occurrence probability index of each micro-service inthe business based on the topological relation, and obtaining the weight of each micro-service in the business based on the probability index; and calculating a performance index of the micro-service,and obtaining a real-time health degree index of the service based on the weight of the micro-service in the service and the performance index of the micro-service. According to the influence dependence degree of each micro-service on the business link and the performance index of each micro-service, the efficient and accurate real-time monitoring and analysis capability of the quality of the business process is realized, the cluster aggregation identification is realized through track slicing and vector conversion and through a feature extraction algorithm and a secondary similarity algorithm, and thus a large amount of manpower is prevented from being invested to carry out the combing process.

The invention discloses a phase encoding characteristic and multi-metric learning based vague facial image verification method. The phase encoding characteristic and multi-metric learning based vague facial image verification method comprises (1) a training phase, namely, partitioning sampling images and extracting multi-scale primary characteristics of every image block, performing fisher kernel dictionary learning through the above characteristics to generate into partitioning fisher kernel coding characteristics, performing multi-metric matrix learning on the above coding characteristics to generate a plurality of metric matrixes and obtain the metric distance after training samples are performed on multi-metric matrix projection, calculating the average metric distance and variance of positive samples and negative samples to a set and confirming a final classification threshold through a probability calculation formula of Gaussian distribution and (2) a verification phase, namely, partitioning input facial images and extracting multi-scale primary characteristics, generating partitioning fisher kernel coding characteristics, obtaining the final metric distance through the multi-metric matrix and comparing the distance and the threshold to obtain a facial image verification result. The phase encoding characteristic and multi-metric learning based vague facial image verification method has the advantages that the identification rate is high and the universality is strong.

The invention discloses a method and a system for assisting in aiming. An object the angle between which and the current orientation is minimum is firstly selected as an attacking target, the trajectory of the target is predicated by combining the movement law of the target, the position of the trajectory in which a bullet collides the target is calculated, the orientation is adjusted by the position and skills are released. Compared with traditional method of manually determining the attacking target and the attacking direction, the method and system for assisting in aiming provided by the invention have obvious advantages on accuracy, convenience and efficiency, and the game experience of players is greatly enhanced.

The invention belongs to the field of petroleum exploration and development and particularly relates to a system and a method for detecting by applying a Raman spectrum technology in the gas detection process of oil-gas logging. A Raman spectrum gas detection system comprises a sample pretreatment device, a Raman gas analyzer, a data signal processing module, a data communication module, a signal output display unit and a master control unit, wherein the sample pretreatment device is connected with the Raman gas analyzer and is used for inputting the pretreated gas in the Raman gas analyzer for analyzing components; the Raman gas analyzer is used for inputting the analyzed result to the data signal processing module; the data signal processing module is used for outputting the analysis result to the signal output display unit through the data communication module; and the master control unit is used for controlling the inputting of sample gas.

The invention discloses a ship feature re-identification method, application method and system based on deep learning, and According to the method, a deep feature extraction network with the efficientprocessing capability in the field of image perception is used to extract the deep features with the distinction degree, the deep features with the high-level perception semantics is autonomously extracted, the change of pixel levels is not depended on, and the overall features are considered, so that the problems of overwater supervision, low ship information query efficiency and the like causedby inaccurate matching and high detection error rate are solved, and the target ship discrimination can be effectively carried out. The PCB partitioning and the matrix type operation matching have the good recognition effect for the ships only parts of which occur, the matching speed is high, the problems of ship transformation, hidden identity escape supervision and the like are solved, and theintelligent auxiliary effects on the maritime traffic management, accident investigation, water conservancy attack illegal sand mining, navigation channel ship gate passing charging, customs attack private activities and the like are achieved. Compared with a ship identification method in the prior art, the ship re-identification method has obvious advantages in efficiency, cost and accuracy.

The invention discloses a rapid and steady image splicing method based on medical microscopic imaging, and belongs to the field of image processing. To achieve high speed, high-precision microscopic image splicing, the rapid and steady image splicing method selects an ORB algorithm to extract image features to prove that through processing, the same precision can be achieved through the method andalgorithms such as SIFT and SURF, and the calculated amount can be reduced on feature extraction. Besides, in order to eliminate the mixed gap to generate a seamless image, the rapid and steady imagesplicing method adopts a gradual-in gradual-out weighted average strategy, and in the final fusion stage, the rapid and steady image splicing method adopts the image indexing operation, so that multiple times of copying and moving of pixels are avoided to the maximum extent, and the time is further shortened. In addition, for the rapid and steady image splicing method, the whole algorithm processis accelerated by using a GPU, and the algorithm processing time is reduced to the minimum. Through calculation, for the rapid and steady image splicing method, accurate seamless splicing can be completed within about one second by using a 3 * 3 image with the resolution of 1360 * 1024.

The invention discloses an image matching method based on random sampling hash representation. The method includes the following steps: forming an original data set through n images, extracting visual characteristics of all the images to generate a characteristic space, randomly selecting m images from the original data set, randomly extracting p visual characteristic subsets in the characteristic space to obtain a sample subset, learning t main characteristic vectors of the obtained sample subset to serve as a Hash projection function, generating a t-bit two-value hash code, repeating the steps for k times to obtain k sections to t-bit two-value hash code, conducting cascading to obtain the k*t-bit two-value hash code to serve as the matching characteristic, acquiring two-value hash codes of the images to be matched and each image in the original data set, conducting similarity measurement based on the obtained two-value hash codes to obtain of a matching result of the images to be matched. By means of the method, the accuracy of a similar adjacent searching method based on the hash codes can be improved, and the method is applicable to image retrieval, image matching and other machine learning algorithms.

The invention discloses an equipment security level classification method, which comprises the steps of training a security classification model, predicting an equipment security level in real time and the like, wherein the step of training the security classification model comprises the following steps of establishing an equipment security level library, selecting training data, extracting security features, carrying out sample SMOTE (Synthetic Minority Oversampling Technique) sampling, carrying out data preprocessing, and training a security level multi-classifier; and the step of predicting the equipment security level in real time comprises the steps of obtaining real-time data, extracting the security features, carrying out the data preprocessing, and predicting the security level. Through a multi-sampling technology, sample distribution is improved, a classification sample imbalance problem is solved, a sample imbalance rate is lowered, improvement is carried out on a data processing level, a data imbalance rate is lowered, and the advantage of a support vector machine model is performed.

The invention relates to flood disaster risk assessment, in particular to a flood disaster risk assessment method based on heterogenous multi-scale monitoring data, which comprises the following steps: acquiring heterogenous multi-scale monitoring data, establishing an index system after analysis, screening and determining a basic risk factor set, determining each network node and causal relationship. constructing a Bayesian network risk assessment model based on each network node and the causal relationship, performing fuzzy evaluation on the network nodes, performing defuzzification, calculating the posterior probability of the target event, and performing flood disaster risk assessment based on the posterior probability of the target event. According to the technical scheme, the defects that in the prior art, spatial heterogeneity of flood risks cannot be effectively handled, and the problems of risk dynamic changes and risk multi-level in the flood disaster breeding, occurrence and development process cannot be effectively solved can be effectively overcome.

The invention discloses an individualized diabetic diet recommendation method by introducing Adaboost probability matrix decomposition. The method comprises the following steps: 1, establishing a foodpreference characteristic set U={u1, u2, ..., un} of a diabetic patient and a food attribute characteristic set V={v1, v2, ..., vm}, recording diets of the diabetic patient, extracting preference characteristics and food attribute characteristics, and forming a food preference matrix U belong to RK*M of the diabetic patient and the food attribute characteristic V belong to RK*N; 2, determining association strength between the food preference of the diabetic patient and the attribute characteristics of the foods by using association degree quantification between the food preference of the diabetic patient and the attribute characteristics of the foods; 3, performing weight distributing on the association degree to obtain basic classification, updating the weight distribution by a trainingdata set, endowing all the association degrees with the weights to be classified, excluding unnecessary foods, and obtaining the final following association degree classification shown in the description; 4, classifying according to conditional probability and the association degree classification, thereby obtaining the individualized diet.

The invention discloses a three-dimensional human body standard skeleton extraction method for continuous frame point cloud, and the method comprises the following two steps: step 1, collecting imagesof a multi-view moving human body, and reconstructing a dense point cloud model by using each view image of each frame; for each frame of point cloud model, carrying out downsampling and surface reconstruction, and using a three-dimensional human body standard skeleton extraction algorithm based on model segmentation to extract a standard skeleton model; step 2, performing inter-frame alignment on the extracted standard skeleton and matching the standard skeleton with corresponding points; constructing a skeleton point sequence of a continuous frame standard skeleton; establishing a continuous frame skeleton point position optimization model to optimize the obtained skeleton point sequence; The finally obtained three-dimensional human body skeleton sequence oriented to the continuous frame point cloud has more advantages than a skeleton extracted by a traditional method in integrity, fitness with an original model, accuracy and standardization on the premise that manual intervention is almost avoided, and has higher practical value and significance.

The invention discloses a neural network model pruning method and system based on adaptive batch standardization. Randomly sampling the number of floating points to serve as the pruning rate of each layer, generating pruning rate vectors (r1, r2,..., rL) as pruning strategies under the limitation of preset calculation resources, and pruning the model based on the pruning strategies to form a pruning model candidate set; updating statistical parameters of batch standardization layers of the pruning models in the candidate set by using a self-adaptive batch standardization method; evaluating andobtaining the classification accuracy of the neural network model with the updated statistical parameters, and finely adjusting the model with the highest classification accuracy on the training setuntil convergence to serve as a final pruning model. According to the method, the candidate sub-networks are quickly and accurately evaluated by adjusting the batch standardization layer, and the parameters of the final pruning network are obtained by finely adjusting the winning pruning strategy in the quick evaluation method, so that huge time consumption required for finely adjusting all pruning networks is avoided, and meanwhile, the accuracy rate also has the advantage.

The invention discloses a lithium battery thermal runaway detection system, which comprises a plurality of first temperature sensors, a plurality of voltage sensors, a first pressure sensor, a secondpressure sensor and a controller, wherein the first temperature sensors are respectively and correspondingly arranged on the surface of each battery cell of each battery cell module of a lithium battery so as to detect the first temperature data of each battery cell; the plurality of voltage sensors are respectively and correspondingly arranged on the surface of a high-voltage copper bar of each battery cell of each battery cell module of the lithium battery and are used for detecting voltage data of each battery cell; the first pressure sensor is arranged at a front end plate in the lithium battery shell and is used for acquiring first voltage data; the second pressure sensor is arranged at a rear end plate in the lithium battery shell and is used for acquiring second voltage data; and the control module is connected with the first temperature sensors, the voltage sensors, the first pressure sensor and the second pressure sensor to receive the first temperature data, the voltage data,the first pressure data and the second pressure data transmitted by the sensors, and judge whether thermal runaway happens to the lithium battery or not based on the data. Correspondingly, the invention also discloses a lithium battery thermal runaway detection method which is implemented by the lithium battery thermal runaway detection system.

The invention relates to a real-time behavior recognition system based on low-power wide-area Internet of tings and a capsule network and a working method thereof. The system comprises four parts, namely, behavior information acquisition, behavior information transmission, behavior information processing and behavior information application. A low-power wide-area network node and a low-power localarea network gateway are adopted for the transmission of a behavior information access layer, so that remote low-power behavior information transmission is realized. Behavior information indeterminacy is subjected to inconsistency and incompleteness processing in a behavior information platform layer, so that the credibility of behavior information is improved. A capsule is adopted for automatically acquiring available features and a space relationship between the features for recognition, so that the accuracy is improved greatly. An error correction mechanism is added in the behavior information application layer, so that the generalization performance of the system is improved, and an effective feasible method is provided for real-time behavior recognition. Thus, the system has certainadvantages on the aspects of practicability, adaptivity, reliability and the like.

The invention relates to a real-time behavior recognition system based on Lora and Capsule and a working method thereof. The system comprises a behavior information physical layer, a behavior information access layer, a behavior information platform layer and a behavior information application layer. The invention adopts Lora node and Lora base station to transmit behavioral information in the behavioral information access layer, and realizes long-distance and low-power consumption behavioral information transmission. By dealing with the inconsistency and incompleteness of behavioral information, the credibility of behavioral information is improved; the Capsule is used to automatically acquire useful features and that spatial relation between features for behavior recognition, and the accuracy is greatly improved.

The invention discloses a Hash code image training model based on binary weight, and a model algorithm comprises the steps: selecting a loss function, determining a target equation, and performing binary coding of a classifier and training image features; performing unified learning of a binary code, updating the binary code, and optimizing the loss function; and deducing the Hash code training model. The invention also discloses a classification learning method employing the Hash code image training model based on binary weight, and the method comprises the steps: obtaining a Hash code of a to-be-searched image through the Hash code training model based on binary weight, and solving Hamming distances between the Hash code and a classifier binary code; searching in the minimum Hamming distance from the Hamming distances, and obtaining the classifier corresponding to the minimum Hamming distance, wherein the classifier is the category to which the to-be-searched image belongs. The method can be used for the image classification for various types of images in high-latitude scenes, improves the performance of an algorithm in a large-scale data set, is precise, efficient and quick, andis small in consumption of the memory.

The invention provides a QR code region detection method for improving background prior and foreground prior. The QR code region detection method comprises the following steps: performing convex hulldetection on an input original image G; carrying out super-pixel segmentation; respectively carrying out multi-feature extraction on the image under each super-pixel scale; calculating a background saliency map of the original image G; calculating a foreground saliency map of the original image G, fusing the obtained final background saliency map and the final foreground saliency map to obtain a weak saliency map, training by adopting a multi-kernel learning enhancement method according to a training sample generated by the weak saliency map to obtain a strong saliency model, and applying themodel to all test samples to obtain a strong saliency map; and finally, weighting and fusing the strong saliency map and the weak saliency map to obtain a final saliency map, wherein the highlighted part in the final saliency map is the QR code area in the image. According to the QR code region detection method, the salient targets can be highlighted accurately and consistently, and the QR code area in the image can be detected accurately. The QR code region detection method provided by the invention has more advantages in the aspect of accuracy of salient target detection.

The invention discloses a two-dimensional encoding and decoding method of visible light locating. The method is based on visible light encoding of OOK, and can be used to realize indoor visible lightlocating. Light sources are encoded, certain carrier frequency is used to modulate the same, thus the light sources are enabled to display a light-dark-alternating polygonal pattern on a CMOS (Complementary Metal-Oxide-Semiconductor) camera, which is set with the certain scanning frequency, while illumination is carried out. An image is collected through a binocular camera fixed on mobile equipment, the obtained image is processed and decoded to obtain information which is carried thereby, to distinguish the light sources of different locations, and finally, a location of the equipment is calculated and obtained by binocular locating technology. Based on visible light two-dimensional encoding and decoding technology, the useful information contained in the image can be quickly and efficiently extracted, the different light sources can be identified, and locating on the equipment can be realized. Encoding and decoding of indoor visible light are realized, and the method has the advantages on the aspects of encoding rules, accuracy, reliability, stability, versatility and the like.

The invention provides a portable gas explosive identification device based on a Raman spectrum, aiming at solving the problem that most of existing gas explosive identification devices are inconvenient to carry and use due to too large volume. The portable gas explosive identification device based on the Raman spectrum comprises a sample pretreatment device, a first analysis unit, a second analysis unit, a data communication unit, a signal output display unit and a main control unit. The portable gas explosive identification device provided by the invention has the beneficial effects that an analysis and processing mechanism is not included in the identification device, so that the reduction of volume and the improvement of the portability of the identification device are facilitated; combustible gases such as methane can be detected, rapid, simple and noninvasive quantitative and qualitative analysis can be achieved, and the portable gas explosive identification device has obvious advantages in aspects of continuity, reliability, accuracy, sensitivity and the like; and the safety of gas explosives, namely combustible gases, in inspection and identification processes can be ensured to the greatest extent.

The invention relates to a belt conveyor, in particular to a belt conveyor with a linear detector. The belt conveyor comprises support rollers and the linear detector. A shaft of each support roller is tubular. Each support roller is provided with a channel penetrating in the axial direction. The linear detector is arranged in the channels. In this way, the linear detector can directly detect the temperature or other state quantities of the support rollers.

The invention discloses an annular workpiece size measurement and defect detection method and device. The method comprises the steps that after a system is started, a circular ring workpiece enters a rotating disc; an optical fiber sensor monitors the rotating disc, and a trigger signal is generated when the workpiece passes; the trigger signal is respectively sent to the corresponding imaging and image collecting stations; and after the imaging and image collecting stations obtain images, original images are sent to a software platform, the software platform carries out detection through the algorithm detection modules according to the images of the stations. The detection speed is high, the working speed can reach 500 per minute, the detection result robustness is good, stability is high, and the device can continuously work for more than 12 hours; and at present, on an annular workpiece production line, the detection method can measure the size of the circular ring workpiece and detect the annular workpiece at a high speed, and the detection speed, accuracy, stability and repeatability have great advantages in the market.

The invention provides a phlegm resolving liquid for extracting free DNAs from a cell-free supernatant of a sputum specimen. The phlegm resolving liquid consists of physiological saline and a dithiothreitol water solution with the concentration of 0.1-0.7 mol / L; the physiological saline and the dithiothreitol water solution are independently packaged and stored before use; normal saline is mixed with the sputum specimen and the dithiothreitol water solution on the use site. The volume ratio of the physiological saline to the dithiothreitol water solution is 9:1. Reagents of a formula are easily obtained, a preparation method is simple, and wide popularization is promoted. Compared with a phlegm resolving liquid reagent which has been applied clinically, the phlegm resolving liquid has advantages in the aspects of the phlegm resolving effect, the amount of obtained cfDNAs and the accuracy of a gene change detection result. Since the amount of the obtained cfDNAs is large, different kinds of genes can be detected. The formula belongs to isotonic liquids; after the cell-free supernatant is collected through short-term treatment, a cell fixing liquid is added into a precipitate, correspondingly cell morphology cannot be damaged, and a subsequent cytological pathological diagnosis result is not influenced.

The invention discloses a box-type vehicle deformation real-time measurement method and system based on binocular vision, which are used for representing the deformation degree of a carriage in a numerical form and judging whether the carriage needs to be repaired or not. A box-type vehicle deformation real-time measurement system based on binocular vision comprises an image collection module, an image processing module and a deformation calculation module. Wherein the image collection module consists of a camera array and is used for acquiring a box-type vehicle image; the image processing module consists of data processing equipment and is used for receiving the image collected by the camera and processing the image; and the deformation calculation module is used for comparing and calculating whether the compartment of the box-type vehicle deforms and the deformation degree according to the three extracted images of the compartment and the image before deformation, and judging whether the vehicle needs to be repaired or not. The problem that it is difficult to judge whether the compartment deforms or not and detect the deformation degree is solved, the compartment can be conveniently detected at any time, and the method has the advantages in the aspects of cost, operability, accuracy and the like.

The invention discloses a method of introducing Adaboost probability matrix decomposition for diabetes personalized diet recommendation, comprising: step 1, establishing a dietary preference feature set U={u for diabetic patients 1 , u 2 ,...,u n} and the attribute feature set of food V={v 1 ,v 2 ,...,v m}, record the diet of diabetic patients, extract preference features and food attribute features, and form the dietary preference matrix U∈R of diabetic patients K×M and food attribute features V ∈ R K×N Step 2, determine the correlation strength between the dietary preference of the diabetic patient and the attribute characteristics of the food by quantifying the correlation between the dietary preference of the diabetic patient and the attribute characteristics of the food; Step 3, weighting the degree of association After the value distribution is basically classified, the training data set updates the weight distribution, and assigns weights to all the correlation degrees to classify and exclude unnecessary foods, and obtains the final correlation degree classification as follows: Step 4. According to the conditional probability and the correlation degree Categorize the personalized diet.