45results about How to "Optimize Time Complexity" patented technology

The invention relates to a program code similarity quick comparison method based on an abstract syntax tree, which relates to the technical field of code reuse. The program code similarity quick comparison method based on the abstract syntax tree is characterized by constructing a program abstract syntax tree, extracting program code features based on the abstract syntax tree, hashing code features, and finally judging code similarity by calculating Hamming distance of code feature hash values. This method takes the abstract syntax tree of program code as the comparison object, and combines Simhash and inverted index technology to transform the similarity comparison of program code into the comparison of hash values of code features. Under the premise of ensuring high accuracy and recall rate, this method can not only realize the fast similarity comparison of program code, but also meet the needs of fast similarity comparison for massive code. This method has better time and space complexity, and can be used to compare the similarity of massive codes in application scenarios, thus providing support for software code reuse and traceability.

The invention discloses a method for calculating the author influence based on a citation propagation network. The method comprises the following steps of step 1, establishing a citation relationship network and a colleague relationship network: establishing an author citation relationship network and the colleague relationship network based on an essay public information database; step 2, establishing the citation propagation network, wherein the citation propagation network is an integration network of the author citation relationship network and the colleague relationship network; step 3, calculating the author influence based on the citation propagation network, wherein the calculation formula is as shown in the figure. The method for calculating the author influence based on the citation propagation network can obtain more reasonable author influence indexes, and is easy to be implemented and simple and convenient for calculation.

The invention discloses a linear deviation degree method-based agent path planning method, relates to an agent path planning method, belongs to the field of artificial intelligent researches, and aimsat solving the problems that the existing graph search type algorithms are high in calculation time complexity under complicated environments and the global path planning of the mobile robots is lowin working efficiency and instantaneity in known environments. According to the method, a model which is scaled down at a same proportion is established according to a practical environment and a coordinate system is correspondingly established; and intelligent nodes are added, paths unrelated to the added nodes are deleted, modeling is carried out on the basis of a linear deviation degree, screening is carried out according to a deviation angle, and finally a path node is searched to determine a finally found optimal path. The method is mainly used for the path planning of agents.

The invention discloses a vehicle route optimization method. The vehicle route optimization method includes that defining a vehicle route problem into a graph model, solving an inter-cluster cost route from a global view to acquire a feasible solution space, and optimizing the feasible solution space through a Monte Carlo method, genetic manipulation, a quantum rotating gate adaptive strategy and the like. The vehicle route optimization method is capable of solving the vehicle route problem in the global connection, optimizing the time and space complexity in the problem solving process, and avoiding premature convergence. The vehicle route problem refers to that a certain number of customers have different numbers of goods demands, a distribution center provides goods for the customers, one motorcade is responsible for distributing goods, and proper driving routes are organized to meet the demands of the customers and achieve the aims of shortest journey, lowest cost, shortest time consumption and the like under a certain constraints.

The invention discloses a training sample grouping construction method used for support vector regression (SVR) short-term load forecasting, and belongs to the field of intelligent computing and machine study. The training sample grouping construction method comprises a step of analyzing correlation, wherein the correlation degree of the load of each time interval and the loads of other time intervals is analyzed through the Tangs correlation degree of the grey correlation degree to form a correlation degree matrix; a step of grouping prediction problems, wherein the time intervals with high load correlation degree are divided into one group according to the correlation degree matrix; a step of constructing a reference load matrix; a step of selecting a reference load to construct a training sample, wherein linear function fitting is carried out on each row of the loads in a load variation rate matrix in a least square fit mode, and fitting variance is calculated; and a step of selecting the load of the time interval with small fitting variance to serve as the forecasting reference load of the group. The training sample grouping construction method used for the SVR short-term load forecasting is capable of improving the load forecasting accuracy, and avoids the problem of high time complexity. The experiment result shows that a short-term load forecasting model trained by the training sample constructed through the method has good performance in forecasting accuracy and time complexity.

The invention discloses a detection method for a lamination lithium ion battery. The method includes the following steps: carrying out statistics of a variance of each line of a lamination battery gray level image in the horizontal direction and carrying out threshold processing on a statistical result so as to determine a calculated characteristic area; carrying out angular point detection on an image of the determined characteristic area and calculating the similarity of each point of the image and calculating a direction change intensity and according to the direction change intensity, setting a threshold to filter image angular point positions, that is the angular point detection; through statistics of a maximum occurrence probability of the angular points in the horizontal direction, classifying the angular points; and after a position of a maximum occurrence probability of anode and cathode angular points is determined, using the maximum position as a reference point to screen anode and cathode piece angular points and then carrying out three-cubic fitting after the screening and carrying out compensation processing on a position at which opening is larger during the fitting. The detection efficiency and precision of the lamination lithium ion battery are improved.

The invention provides a model-independent genome structure variation detection system and method, wherein a model-independent structure variation detection theory is used as a core, and structure variation detection without depending on any variation model is achieved through a variation signal extraction module, a frequent maximum subgraph mining module and a classification module. According tothe system, a frequent variation pattern mining module is used for capturing the characteristics of structural variation left on a genome, and judging a potential structural variation region only by mining abnormal points in a large amount of normal data; and according to different genome disturbance modes of different variation types, different arrangement sequences of variation signals are further caused, and the different variation types are classified on the basis of the different arrangement sequences in combination with a deep learning model with a memory function. According to the invention, the system does not depend on any variation model, so that the variation detection sensitivity and error rate are greatly reduced; and the system is suitable for detection of complex variation types, and an additional structural variation model does not need to be established.

The invention discloses a multi-target vehicle tracking method based on a Markov decision process (MDP), and the method comprises the steps: obtaining a target vehicle through a video sequence, detecting the target vehicle, and enabling the target vehicle to enter an activated state; wherein the state of the target vehicle comprises an activated state, a tracking state, a non-activated state and alost state; for the target vehicle in the activated state, training a binary classifier in an offline mode, and judging the state transition of the target vehicle according to the binary classifier;for the target vehicle entering the tracking state, an appearance template is constructed for the target vehicle in an online mode, and state transition of the target vehicle is judged according to the appearance template; for a target vehicle in a lost state, judgment of state transition of the target vehicle is realized through data association binary classifier reinforcement learning; the method can overcome the main technical difficulties faced by the vehicle tracking technology, reduce the tracking error probability and improve the tracking effect.

The invention discloses a search algorithm of all spanning trees of a directed graph and an undirected graph. Based on an MATLAB (matrix laboratory) platform, root nodes are found in the directed graph and the undirected graph, wherein the root nodes include all the spanning trees of given sub-trees. The search algorithm comprises the adjoining of branches by the nodes, the expansion of the branches and the seach of all the spanning trees. The search of all the spanning trees is realized by calling the adjoining of the branches by the nodes and the expansion of the branches. The search algorithm is very good in time complexity and space complexity. All the spanning trees of the directed graph and the undirected graph can be reliably and quickly searched. The search algorithm is ingenious and novel in thinking, thereby being very good in application prospect.

The invention discloses a PSPNet-RCNN-based dome camera monitoring anomaly detection method, and the method specifically comprises the steps: collecting multi-period exposure images at the same position, fusing the exposure images to generate an HDR image, carrying out the SIFT feature registration of the HDR image and an original position image, obtaining a registered HDR image, and carrying out the enhancement processing; the voidage of void convolution of a PSPNet network pyramid pooling module is increased, coding and decoding convolutional neural network branches are added to PSPNet, a residual convolutional network is added to an entrance of a feature extraction module for inputting double images, an improved multi-scale PSPNet network is obtained, the improved multi-scale PSPNet network is fused with a Mask R-CNN network for extracting target information in an HDR image, and the target information in the HDR image is extracted. And a PSPNet-RCNN image anomaly detection network is obtained and is used for extracting an image anomaly region. According to the invention, the universality and accuracy of the dome camera monitoring anomaly detection method are improved, and the real-time performance and robustness of the detection result are enhanced.

The embodiment of the invention provides an address fuzzy matching method and device. The method comprises the following steps: respectively obtaining a plurality of virtual record pairs correspondingto each target user in the standard data block, wherein each standard data block is used for storing the same standard address tables of the first three layers respectively, each virtual record pairis used for storing a corresponding user unique identifier and a corresponding virtual address respectively, and the virtual address is acquired in advance based on detailed address information in a fourth layer of the corresponding standard address table or a layer larger than the fourth layer of the corresponding standard address table; generating a target directed graph by applying each virtualrecord pair; and carrying out message transmission in opposite transmission directions twice in the target directed graph to obtain the address similarity between the virtual record pairs with similarity. According to the method, the efficiency of user address fuzzy matching can be effectively improved, the calculation amount of user address fuzzy matching can be effectively reduced, and then theefficiency of obtaining users with high address similarity by enterprises can be effectively improved.

The invention discloses a passive type two-person behavior recognizing method based on WIFI background noise. The method is characterized by comprising the following steps that firstly, RSS data of combinations of six behaviors are collected through a WLAN card, secondly, wireless signal sample data are preprocessed to obtain denoised sample data, thirdly, a classification system model of the combinations of two-person behaviors is established according to the denoised sample data, and fourthly, data to be classified are recognized through the classification system model of the combinations of the two-person behaviors. According to the method, single person behavior recognition can be expanded to two-person behavior combined recognition, and a tested person can freely move indoors without carrying devices.

The invention belongs to the technical field of natural language processing, and provides a sensitive word detection method based on a decision tree and variant recognition. The method comprises the following steps: S1, constructing a sensitive word dictionary and updating according to a preset period, wherein Chinese character patterns and corresponding pinyin are added into the sensitive word dictionary according to an initial sequence, and constructing a decision tree according to the sensitive word dictionary; S2, inputting a text into a sensitive word detection model, wherein the sensitive word detection model detects sensitive words in the text through a matching algorithm and a matching standard on the basis of the decision tree, and positions the sensitive words. Through direct comparison and similarity comparison of character patterns and pinyin, the problems that sensitive words are not included and detection is avoided by modifying character patterns, pinyin, or English of the sensitive words, can be effectively solved.

The invention provides a network page efficient and accurate deduplication system based on cloud computing, and aims to solve the problems that most of web pages searched by an existing search engineare static web pages, due to the existence of a large amount of transshipment and plagiarism, the main content of a large number of web pages is repeated, and for the search engine, the repeated web pages virtually increase the burden of index storage, and meanwhile, more retrieval time can be consumed; the webpage deduplication system based on the Hadoop cloud platform is designed and realized bycombining an open source framework, other modules of a search engine can be better connected by adopting a mode of detecting and judging duplicate in real time after a spider program captures a webpage; and in a massive webpage collection stage, the network page efficient and accurate deduplication system based on cloud computing can preprocess the web pages in advance, then web page similarity detection and discovery are carried out, repeated web pages or web pages with high similarity are removed, and therefore index quality is improved, retrieval results are optimized, and good search experience is provided for users.

The embodiment of the invention provides an RLC SDU segmentation processing method and device, and a terminal, and relates to the technical field of communication. The method comprises the following steps: receiving a first radio link control RLC service data unit SDU segment; determining a hash value of a sequence number SN of the first RLC SDU segment; and according to the hash value, storing the first RLC SDU in a linked list corresponding to the hash value in a segmented manner. According to the embodiment of the invention, the SDU segment is stored in the linked list corresponding to thehash value of the SN of the SDU segment, so that the operation is simple.

The invention discloses an Internet of Vehicles system performance optimization method based on Nakagami-m fading. The method comprises the following steps: in a time slot when a group of V2V and V2Iusers complete one-time communication, acquiring sending power and link state information of respective links, and establishing a link outage probability and traversal reachable rate compact upper bound model according to the acquired information; assuming that the total transmission power of the link is a constant value, establishing an optimization model by taking maximization of the traversal reachable rate of the V2V link and the V2I link as an objective function under the constraint of the link outage probability, and searching for a power distribution coefficient enabling the upper boundof the total reachable rate of the V2V link and the V2I link to be maximum. According to the method, the compact upper bound of the reachable rate is used for replacing complex accurate solution calculation, and the time complexity of the algorithm can be greatly improved while the error range is ensured. The total reachable rate of the system is maximized under the constraint of the outage probability of the system, and the performance of the Internet of Vehicles system is comprehensively improved.

The invention discloses a personalized recommendation system based on a user memory network and a deep model with a tree structure. The personalized recommendation system is characterized by comprising a user memory module, a commodity body module and a prediction module. Wherein the user memory module is used for capturing historical data of a user; the user memory module is composed of a context-based long and short memory network framework, and captures interest dynamics of a user through short-term memory and long-term memory. The short-term memory is used for capturing records of the userfor purchasing commodities recently, and obtaining short-term memory mapping of the user through the records; the long-term memory summarizes and records the characteristics of the commodity which the user is interested in according to the long-term purchasing habit of the user and a large number of purchasing records of the user, and long-term memory mapping of the user is obtained through the records. The commodity body module obtains mapping information of the commodities through the associated information between the commodities and historical purchase records of the user; and the prediction module performs final recommendation prediction in combination with the short-term memory mapping, the long-term memory mapping and the commodity mapping output by the user memory module and the commodity body module.

The invention discloses a picture retrieval method. The method comprises the steps of carrying out DCT calculation and DWT calculation on a processed picture to obtain a first group of hash codes, performing DCT calculation and DWT calculation on the rotationally processed pictures at 90 degrees, 180 degrees and 270 degrees to obtain a second group of hash codes, a third group of hash codes and afourth group of hash codes, calculating and constructing an NB+tree based on the four groups of hash codes, combining the NB+tree into a random forest model, and performing picture retrieval by usingthe random forest model. A block DCT and DWT technology is used for constructing a perceptual hash code of a picture, so that the hash code construction process is accelerated; a decision tree is constructed according to the pictures with different rotation angles, so that the picture retrieval accuracy is improved; a normalized B+tree is used for reducing a high-dimensional input vector to one dimension, so that the time complexity is remarkably improved; a random forest model is used, and multiple decisions such as picture rotation, DCT and DWT hash codes are combined to improve the retrieval speed and performance.

An ACT-Apriori algorithm based on a self-encoding technology comprises the following steps: firstly, preprocessing a transactional database D, neglecting a high-frequency parameter NS of each item in the transactional database D, and replacing the high-frequency parameter NS with a self-encoding bit vector SNT to form a simplified database RBD, so that the data dimension is greatly reduced; according to the method, data records are all read into a memory after being simplified, in the process of generating candidate item sets through frequent item set connection and pruning, the process of generating the candidate item sets is improved, the candidate item sets are directly generated, the database is scanned after the candidate item sets are obtained, and the support degree is calculated; when a candidate item set is searched in each record, the search of the transaction can be stopped once a value greater than a candidate item is searched. Compared with an existing brand-new association rule algorithm, the algorithm has the advantages that the calculation time is greatly shortened, the memory occupancy is greatly reduced, and the time complexity and the space complexity are obviously optimized.

The invention relates to an SMO parallel processing method orientated at a multi-core cluster. The SMO parallel processing method orientated at the multi-core cluster comprises the steps that an initial value is assigned to a local problem parameter according to a global parameter, and an initial value is assigned to an algorithm parameter; a local first boundary and a local second boundary of the local problem parameter are calculated according to the initial value of the local problem parameter; a global first boundary and a global second boundary are obtained according to local first boundary and the local second boundary; when the difference between the global first boundary and the global second boundary is not smaller than preset accuracy, a first multiplier corresponding to the global first boundary and a second multiplier corresponding to the global second boundary are calculated in an iterative model; the local problem parameter is updated in a multithreading mode after each time of iteration; when the iteration reaches a preset iteration frequency, a local solution of data to be classified is calculated according to a local sample multiplier, a global solution is obtained according to the local solution, and data classification is finished. The SMO parallel processing method orientated at the multi-core cluster resolves the traditional problems of high data classification cost, a high error rate and low response speed.

The present invention provides a decryption method and device. The method includes: determining N task cycles, where N is an integer greater than 1; and sequentially executing the corresponding stages of the decryption task in the N task cycles to obtain plaintext data, wherein the decryption task includes: At least two decryption subtasks executed sequentially, the decryption task can be divided into at least two stages in sequence, and one stage of the decryption task corresponds to one task period. Based on the decryption method and device disclosed in the present invention, on the premise of not changing the hardware of the RKE system, the load rate of the processor in the decryption process is reduced, and a large delay in responding to other task requests can be avoided.