1988results about How to "Diversity guaranteed" patented technology

The invention, which belongs to the field of intelligent traffic sign recognition technology, relates to a traffic sign recognition method based on an asymmetric convolution neural network. With the method, problems of slow recognition speed and poor robustness during traffic sign recognition can be solved. According to the method, two convolution neural networks with different structures are used for carrying out feature mapping and extraction concurrently; the features are combined; and a full connection layer and a classifier are used for completing the whole classification process. The two convolution neural networks with different structures employ a random pooling operation and a maxout unit respectively, thereby guaranteeing diversity of the image features, improving the recognition precision, and accelerating the network operation speed. According to the invention, the structure of the traditional convolution neural network is modified and the two convolution neural networks with different structures are used for replacing the traditional convolution neural network structure. Therefore, the image feature diversity is guaranteed; the recognition precision is improved; and the network operation speed is accelerated.

The invention discloses a method and device for optimizing multi-constraint quality of service (QoS) routing selection. The method comprises the following steps: acquiring the topological structure and link parameters of an existing network in accordance with information of a prediction model; creating a corresponding multi-constraint QoS routing model in accordance with the determined topological structure and link parameters, and constructing penalty functions to transform multi-constraint conditions, as well as constructing fitness functions for evaluating paths; using a depth-first search method to acquire initial feasible paths and initializing particle swarms; calculating the fitness value of each particle, and finding out the optimal fitness value of the particle adjacent to each particle; using the generation algorithm and the genetic algorithm-particle swarm optimization (GA-PSO) to carry out iterative solution at the beginning of the initial feasible paths, and carrying out natural selection and variation operations; and finding out paths which meet conditions and are provided with the optimal fitness values, realizing optimal routing selection under the multi-constraint condition, and executing in accordance with the found routings.

The invention relates to a battery management system performance test platform and a testing method based on semi-physical simulation and belongs to the field of systems. According to the battery management system performance test platform and the testing method based on semi-physical simulation, the problems existing in the aspects of effectiveness, practicality, accuracy, generality and comprehensiveness of an existing battery management system testing device and method are solved. A BMS to be tested and a simulation control and emulating unit of the test platform are connected with a CAN bus. A signal output end of an insulation resistance testing device, a signal output end of an insulation voltage resistant testing device, a signal output end of a 24-channel single-body voltage simulator, a signal output end of a high-voltage source, a signal output end of a current source and current reversing module, a signal output end of a high-accuracy temperature environment box, a signal output end of an insulation resistance simulator and a signal output end of a direct current power source are respectively connected with corresponding signal input ends of the BMS to be tested. The BMS to be tested is arranged in a high-low-temperature operational testing box. The testing method comprises the steps of safety testing and the comprehensive testing of state parameter measuring accuracy, SOC estimating accuracy, a battery fault diagnostic function, a heat management function and environment adapting performance and is used for testing the battery management system.

The invention discloses a visual detection method of shockproof hammer defect detection. The visual detection method comprises the steps that denoising and anti-shaking preprocessing is performed on an aerial photographing image so as to obtain an original image to be detected; the existing original image is expanded by using the method of geometric transformation, scale change and contrast transformation so as to generate more data similar to the original image; samples are acquired, a shockproof hammer in the aerial photographing image is acquired and the size of the shockproof hammer is mainly acquired; a network model to be trained is determined, and the sample data are inputted to perform forward propagation and reverse propagation to adjust the weight so as to obtain the optimized detection network model parameters; the image to be detected is identified by using the trained model and the position of the hammer of the shockproof hammer is determined; and the lead in which the hammer is located is determined, and shockproof hammer defect discrimination is performed according to the relative position of the lead and the shockproof hammer and the constraints of respective directions.

The present invention discloses a deep learning-based insulator identification method. The insulator identification method comprises the steps of pre-processing an aerial image, and secondly, extending the data via the methods, such as the geometric transformation, the contrast enhancement, an analog noise adding method, etc.; acquiring the insulator samples, aiming at the insulators of different types, classifying to acquire; determining a to-be-trained model structure; inputting the samples in the to-be-trained model, and continuously adjusting the weights and the bias parameters by the forward propagation and backward propagation methods, and finally determining an optimal model parameter, based on the trained model, taking a to-be-detected image as an input signal, and by the network multi-layer convolution, pooling and full-connection operations, obtaining a final detection identification result. According to the present invention, by a deep learning method, the insulator characteristics are learned continuously, a learning network model is determined, the different insulators are identified under different background environments, and support is provided for the electric power maintenance decisions.

The invention discloses an urban storm flood monitoring and traffic controlling and guiding system and method. The urban storm flood monitoring and traffic controlling and guiding system comprises a water depth data collecting and rectifying unit, a unit for analyzing and predicting water depth and on-off control of a drainage pump station and an early warning issuing and traffic controlling and guiding processing unit. The water depth data collecting and rectifying unit is used for performing comparison and check analysis by collecting water depth data of different types at water accumulating spots and obtaining water depths at the water accumulating spots. The unit for analyzing and predicting water depth and on-off control of the drainage pump station is used for performing trend analysis and rolling correction on the water depths of the water accumulating spots by combining zone rainfall data of surrounding rainfall stations on the basis of the water depths obtained by the water depth data collecting and rectifying unit and achieving automatic on-off of a drainage pump unit according to an accumulating water depth predicting curve. The early warning issuing and traffic controlling and guiding unit is used for processing obtained traffic controlling and guiding information and sending the information to a controller and early warning and instructing pedestrians and vehicle drivers in various manners according to safety pass standards of water-accumulating roads. The urban storm flood monitoring and traffic controlling and guiding system and method have the advantages of being reliable in data, stable in communication, comprehensive in function, professional in handling and the like.

The invention provides an image processing method, processing device and processing equipment. Image conversion is realized by using a generative neural network with combination of the image content characteristics and the style characteristics, and the resolution of the converted image outputted by the generative neural network is enhanced accordingly by using a super-resolution neural network sothat the high-resolution converted image can be acquired. The image processing method comprises the steps that an input image is acquired; a first noise image and a second noise image are acquired; image conversion processing is performed on the input image by using the generative neural network according to the input image and the first noise image so as to output the converted first output image; and high-resolution conversion processing is performed on the first output image and the second noise image by using the super-resolution neural network so as to output a second output image, wherein the first noise image and the second noise image are different.

The present invention discloses a laser and vision-based hybrid location method for a mobile robot. The mobile robot comprises a laser radar and a vision sensor. According to the technical scheme of the invention, the weight of each particle at a predicted position is updated based on the collected data of the laser radar and the collected data of the vision sensor. After that, particles of higher weights are re-sampled, so that the real location distribution of the mobile robot at the moment t can be obtained. Compared with the prior art, the above technical scheme integrates the high accuracy of the laser radar with the information integrity of the vision sensor, thus being wider in application range. Meanwhile, the defect that the visual location is unstable is overcome. In addition, in one embodiment of the present invention, a conventional particle filtering sampling model is improved, so that the diversity of particles is ensured.

The invention discloses a group protein structure prediction method based on Rosetta local reinforcement. The prediction method includes the steps: firstly, dividing searching processes of a whole algorithm in structure prediction into four stages, setting fragment length for each stage, assembling fragments, and selecting different energy functions to measure weight of conformation individuals; secondly, generating testing conformations by the aid of different mutation strategies and loop area information based on secondary structure information, randomly exchanging the loop area information to achieve cross processes, keeping population diversities, and executing Rosetta local reinforcement for testing conformations and target conformations of the stages; finally, extracting characteristic vectors of the conformations to measure diversities of the conformation individuals, taking the energy functions as main measurement indexes, taking the diversities as auxiliary measurement indexes, and guiding conformation groups to update. The prediction method is high in searching capability and high in prediction accuracy, and group diversities can be kept by energy.

The invention provides a method, a system and a device for converting a website into Web App for displaying. The method comprises the following steps that a cloud server acquires the domain name of the website, and captures a plurality of webpages in the website; the cloud server analyzes the webpages respectively to generate a plurality of front end templates of the website, and stores in a database; the cloud server receives an access request message sent by a user through a mobile terminal; the cloud server transmits a website shell architecture and a label of the website to the mobile terminal; and the cloud server transmits a front end template which corresponds to the access request message and content data to a mobile terminal end template, renders the front end template to convert a webpage requested by the user into a Web App page, and displays to the user. Due to the adoption of the method, webpages are rapidly converted into Web App which is suitable for browsing on the mobile terminal in a nondestructive way, the flow electric quantity is optimized fully, webpages can be switched smoothly, the waiting time is shortened, and user experience is improved. The invention further discloses a system for converting a website into Web App for displaying, a cloud server and a mobile terminal.

It includes display layer, data storage / procession layer (SP) and system interface layer (SI). SP includes data collect module (DL), data pre-processing module (DP), data comparison module (DM), data sync, module (DS) and assistant module. Via SI, the business backup and bearing nets (BBB) realize mutual comm. between net elements (NE). From BBB, DL collects NE data requiring keeping coincident. DP / DM pre-processes and compares these data. Non-coincident data are recorded in the error list. According to availability test rule, DS verifies availability of data difference, generates corresponding synchronization data against valid data difference base on sync. rule and sends to related NE to keep coincidence of data between NEs. This invention raises data comparison synchronization level, system running efficiency and resource utilization. It ensures accuracy and coincidence of system data. It extensively is applied in multi-NE cooperation fields, such as comm., insurance and banking.

The invention discloses a PID controller parameter setting algorithm based on an improved PSO (particle swarm optimization) algorithm, and the algorithm comprises the following steps: 1, initializing the algorithm parameters; 2, switching to an iterative loop, and carrying out the updating of the position and speed of each particle; 3, randomly searching a new position in the neighborhood of a current position; 4, calculating the adaptability difference between two positions, and judging whether to accept the new position or not through a simulated annealing mechanism when the adaptability of the new position is inferior to the adaptability of an original position but is superior to the adaptability of a global optimal position; 5, updating the global optimal position of a population, carrying out the natural selection operation, carrying out the arrangement of all particles according to the adaptability values, and employing the information of a part of better particles to replace the information of the other half particles; 6, judging whether to stop the iteration or not; 7, outputting PID controller parameters or executing step 2 again. The method can achieve the automatic setting of control parameters, irons out a defect that a conventional PSO algorithm is very liable to be caught in local optimization, achieves the complementation of the simulated annealing operation and a natural selection strategy, improves the convergence precision of the algorithm under the condition that the number of convergence times of the algorithm is guaranteed, is higher in robustness and precision, and enables the PID controller to generate a more excellent control effect.

The invention provides a mobile application pushing method. The mobile application pushing method includes determining more than one mobile application category with the highest relevancy to mobile applications designed by a user according to relevancy of pre-generated mobile application categories; calculating recommendation levels of the mobile applications under the mobile application categories according to pre-generated weighted values of the mobile applications; extracting the mobile application with higher rankings of the recommendation levels under each mobile application category; and selecting more than one mobile application which has the highest recommendation level as a recommendation result according to the number of pre-set recommendation results, and pushing the more than one mobile application to users. The invention further provides a mobile application pushing system. According to the mobile application pushing method and the system, diversity of the recommended mobile applications can be effectively improved.

The invention discloses a cascade reservoir optimal operation method based on the adaptive particle swarm optimization algorithm. According to the method, aiming at the defect of the particle swarm method in cascade reservoir optimal operation, fixed initialization improvement is conducted firstly on particle random initialization to enable the algorithm to have the possibility of approaching the optimal value at the beginning, large-scale dead zones do not exist, convergence speed is increased, and the stability of the algorithm is improved; then according to the group cooperation idea and the cluster ecological niche idea, an initialized group is dynamically divided into three subgroups, optimization and parameter selection are conducted on each subgroup in an adaptive mode according to the difference of particles, and in this way, the particle diversity is improved, the information exchange model is changed, and local optimum of the algorithm is avoided. According to the improved algorithm, the function problems of nonlinearity and multiple local minima can be well solved, and an effective and feasible solution is provided for cascade reservoir optimal operation.

The invention relates to a Gaussian distribution resampling Rao-Blackwellized particle filter based mobile robot simultaneous localization and mapping method. The method comprises the following steps: S1, robot initial pose is estimated according to robot pose and mileometer control information; S2, a scan matching method is executed according to a map; S3, particle sampling is carried out in proposal distribution of trajectory; S4, weight of each particle is calculated and weight of each particle is updated; S5, particle resampling is carried out on the basis of Gaussian distribution: specifically, by sorting particle weight, high-weight particles are dispersed to obtain resampled new particles; and S6, the map is calculated according to robot pose and observation information, and map revision is carried out. By the method, reliable grid map precision can be obtained.

The invention discloses a service data monitoring method and device, a terminal device and a storage medium. The service data monitoring method comprises the steps that multi-dimensional service datain a big data platform is acquired, and a monitoring strategy configured by a user is acquired, wherein the monitoring strategy comprises at least one monitoring index and at least one monitoring dimension; target service data is acquired from the multi-dimensional service data on the basis of the monitoring dimensions, and whether all the target service data conforms to the monitoring indexes ornot is judged; if the target service data does not completely conform to the monitoring indexes, it is determined that the target service data is abnormal data, and a monitoring result is obtained. When the service data monitoring method is used for service data monitoring, the effects that the service data monitoring efficiency is higher and service data monitoring results are more comprehensivecan be achieved.

The invention discloses a rolling bearing fault diagnosis method based on a convolutional neural network (CNN). By aiming at problems of rolling bearing characteristic components such as easy submergence and difficulty in extraction and combining with rolling bearing signal own and large monitoring data quantity and other characteristics, the CNN is introduced in the rolling bearing fault diagnosis. By short time Fourier Transform, a motor vibration signal is converted into a time frequency spectrogram to be adapted to a CNN network training sample format, and then mass sample data having labels used to express different faults is established, and therefore sample diversity is guaranteed, and network overfitting is prevented. The CNN network having a proper layer number is established, and parameters are initialized, and then the preprocessed samples are input in the CNN for forward propagation. By combining with predetermined label calculation errors, a network weight is adjusted by using an error reverse propagation algorithm, and then after a plurality of times of iterations, the network used for the interconnection between the signal and equipment is established, and therefore the rolling bearing fault accurate diagnosis is realized.

The invention discloses a collision avoidance planning method for mobile robots based on deep reinforcement learning in a dynamic environment, and belongs to the technical field of mobile robot navigation. The method of the invention includes the following steps of: collecting raw data through a laser rangefinder, processing the raw data as input of a neural network, and building an LSTM neural network; through an A3C algorithm, outputting corresponding parameters by the neural network, and processing the corresponding parameters to obtain the action of each step of the robot. The scheme of the invention does not need to model the environment, is more suitable for an unknown obstacle environment, adopts an actor-critic framework and a temporal difference algorithm, is more suitable for a continuous motion space while realizing low variance, and realizes the effect of learning while training. The scheme of the invention designs the continuous motion space with a heading angle limitationand uses 4 threads for parallel learning and training, so that compared with general deep reinforcement learning methods, the learning and training time is greatly improved, the sample correlation isreduced, the high utilization of exploration spaces and the diversity of exploration strategies are guaranteed, and thus the algorithm convergence, stability and the success rate of obstacle avoidance can be improved.

The invention provides a retinal fundus vessel segmentation method based on a deep multi-scale attention convolutional neural network. An internationally disclosed retinal fundus vessel data set DRIVEis adopted to perform validity verification: firstly, dividing the retinal fundus vessel data set DRIVE into a training set and a test set, and adjusting the picture size to 512*512 pixels; then, enabling the training set to be subjected to four random preprocessing links to achieve a data enhancement effect; designing a model structure of the deep multi-scale attention convolutional neural network, and inputting the processed training set into the model for training; and finally, inputting the test set into the trained network, and testing the model performance. The main innovation point ofthe method is that a double attention module is designed, so that the whole model pays more attention to segmentation of small blood vessels; and a multi-scale feature fusion module is designed, so that the global feature extraction capability of the whole model on the segmented image is stronger. The segmentation accuracy of the model on a DRIVE data set is 96.87%, the sensitivity is 79.45%, thespecificity is 98.57, and the method is superior to classical UNet and an existing most advanced segmentation method.

The invention relates to a ship intelligent control system test platform with virtual and real fusion. The ship intelligent control system test platform with virtual and real fusion comprises a real navigation database, a real-time simulation system for generating a numerical twinning fused traffic scene according to data in the real navigation database, a ship simulation system for simulating a ship in the test scene, and a tested intelligent control system for controlling and simulating the ship according to the data provided by the ship simulation system. The experimental environment constructed by the ship intelligent control system test platform with virtual and real fusion disclosed by the invention comes from real data collection, which ensures the validity of the test; and the based object experimental ship and experimental process are still based on numerical simulation, which reduces the risk and cost of the test; that is, the experimental object ship is based on numerical simulation, and the experimental scene completely comes from the perception fusion of the real world.

The invention relates to a real-time path planning method of AUV (Autonomous Underwater Vehicle), in particular to a method for carrying out online, real-time local path planning according to an online map in an AUV real-time collision preventation process. The method comprises the steps of: setting the quantity of small populations according to the quantity of path points of the AUV, initializing; carrying out immune selection on each small population to obtain subgroups; carrying out genetic manipulation on one subgroup, carrying out cell cloning on the other subgroup; then clustering through a vaccination and an antibody to form the next generation of small population, judging whether the next generation of small population meets the conditions or not; if yes, selecting optimal individuals of the small populations; and selecting the optimal individuals from the set consisting of all optimal individuals to be used as a planning path. According to the invention, the diversity of the population is maintained by using an antibody clustering principle, the premature convergence of an algorithm is avoided, and the global optimization is facilitated. The established immune genetic algorithm is used for clustering and analyzing generated filial generations by adopting a self-regulating mechanism, and the diversity of the population is ensured.

The invention discloses a driver seat self adaptive adjusting device based on physical characteristics of a driver. The driver seat self adaptive adjusting device comprises a gravity sensor, a driver body size identification system, a driver sitting posture angle identification system, an adjusting control system, a drive system and a storage system. According to the driver seat self adaptive adjusting device based on the physical characteristics of the driver, the physical characteristic parameter affecting the driver seat adjusting mode is monitored from a plurality of aspects in real time, influence of the driver individual differences on the sitting posture angle recognition is eliminated, the precision is high, and the accuracy is better. Meanwhile, the driver seat self adaptive adjusting device has a memory function, automatically stores the physical characteristics information of the driver and a matching seat adjusting scheme, automatically match the existing adjusting program to adjust according to the physical characteristics of the driver, and has higher timeliness and convenience.

The invention provides a black box antagonistic attack defense method based on sample selection and model evolution. The method includes the following steps: 1) using a sample selector to randomly select partial samples from the multiple types of samples to be input into various attack models to generate a large number of counter samples; 2) calculating the attack effect of the counter samples, and analyzing the attack effect of different input samples and the attack model; (3) updating the number of different samples selected by different samples in the attack model and the sample selector according to the attack effect to make the newly generated counter sample have a better attack effect, meanwhile, updating the counter sample pool, storing the several counter samples with the best attack effect, and outputting the counter samples with the best attack effect in the pool to serve as the final result of the current evolution; and 4) training a large number of training output results and normal samples, so that the attack can be defended. According to the invention, the defense capability of the black box model can be improved.

The invention discloses an improved fuzzy neural network bus intelligent scheduling method based on a chaos theory, and belongs to the field of intelligent transportation. According to the improved particle swarm bus intelligent scheduling method based on the chaos theory, advantages and complementarity of various algorithms are fully utilized, a series of improvement measures are also introduced, such as conjugate gradient optimization, and inertia factor and constraint factor of the particle swarm algorithm etc., the mechanism and the search performance are researched from the theoretical and practical perspectives, problems of poor global search capability and premature convergence of the conventional optimization algorithm are fundamentally solved, the diversity of population can be obviously increased, the global search capability is obviously improved, the problem of fuzzy information can be effectively dealt with, the convergence speed is fast, and a new high-efficiency method is provided for bus intelligent scheduling.

The invention relates to a BP neural network image segmentation method and device based on an adaptive genetic algorithm, and the method comprises the following steps: 1), analyzing a to-be-segmented image, and generating a training sample of a neural network; 2), setting the parameters of the neural network and population parameters, and carrying out the chromosome coding; 3), inputting the training sample for the training of the network, optimizing the weight value and threshold value of the network through employing a new adaptive genetic algorithm, adapting to the crossing and mutation operations, and introducing an adjustment coefficient; 4), inputting the to-be-segmented image, carrying out classifying of the trained neural network, and achieving the image segmentation. The device comprises a training sample generation module, a neural network structure determining module, a network training module, and an image segmentation module. The method introduces the adjustment coefficient which is related with the evolution generations, solves a problem that the individual evolution stagnates at the initial stage of population evolution, and also solves a problem of local convergence caused when the individual adaption degrees are close, thereby obtaining the neural network which can maximize representation of the image features, and achieving the more precise image segmentation.

The invention discloses a fuzzy strong box remote identity authentication method based on a face feature. The method comprises: fuzzy strong box data is generated according to a human feature and a user password, wherein the fuzzy strong box data comprises a human feature value and key information; the fuzzy strong box data is stored into a server; a client obtains a user face image and a password when a user registers, and encrypts the obtained face feature to generate fuzzy strong box data, and sends the data to a database server; the database server obtains a user registration key and a user login key by computing the fuzzy strong box data, and compares the user login key with the registration key; if the user login key is the same as the registration key, the database server regards that the user can log in the system; and the user can select different passwords to protect the human feature, therefore, the cross-database search can be effectively prevented, the safety and privacy of the method can be effectively guaranteed, and the fuzzy strong box remote identity authentication method has application value.

Provided is a boiler combustion optimization air distribution method based on online model prediction. The method includes: constructing a boiler combustion optimization air distribution model by employing an improved BP neural network algorithm and model online update calculation; and performing optimization calculation on parameters such as the optimal oxygen amount, the primary air amount, each secondary air door and a burnout air door etc. with the combination of an adaptive genetic algorithm and dynamic optimization boundary so that a multi-target function constructed by the boiler efficiency and the pollutant discharge capacity reaches the optimal combustion range.

The invention relates to a pedestrian re-identification method fusing random batch masks and multi-scale representation learning. The pedestrian re-identification method comprises the steps of constructing a pedestrian re-identification training network; performing network hyper-parameter adjustment according to preset training parameters to obtain a learning network; shielding multi-scale representation learning and random batch mask branches to obtain a test network, and inputting the test set into the test network to obtain a corresponding test identification result; judging whether the accuracy of the test recognition result is greater than or equal to a preset value or not, if so, inputting the actual data set into the learning network, and otherwise, retraining the network; and finally, shielding multi-scale representation learning and random batch mask branches to obtain an application network, and inputting the query image into the application network to obtain a correspondingidentification result. Compared with the prior art, the method has the advantages that a random batch mask strategy, multi-scale representation learning and loss function joint training are used, moredetailed discrimination features of pedestrian images can be captured, and local important suppressed features are extracted.

The invention provides a multi-unmanned aerial vehicle (UAV) track planning method based on a culture ant colony search mechanism, which includes the following steps: (1) carrying out mesh generationon a standard space according to a grid method; (2) building a multi-UAV track planning model, including the number of UAVs, the start and end points and a threat model; (3) initializing the start point and the end point; (4) initializing an ant colony algorithm, including: initializing an ant colony and calculating a heuristic factor and a guide factor; and (5) assigning all ants to an initial node, and updating taboo knowledge; selecting next node for transfer according to the taboo knowledge and the state transfer probability until there is no optional node or a destination node is selected, updating historical knowledge, and updating pheromones according to the historical knowledge; and outputting a shortest path if the maximum number of iterations is achieved, and continuing the process until U multi-UAV optimal multi-path tracks are obtained. The problem that it is difficult to find the optimal flight tracks of unmanned aerial vehicles due to slow search and heavy computing burden is solved, and multi-UAV track planning is realized.

A method for detecting dynamic gridding instruction based on artificial immunity, is a method for detecting instruction facing to gridding which takes use the artificial immunity technique for reference. According to the dynamic and real time requirement of the instruction detection under gridding surroundings, the method takes the prior clonal selection algorithm as main body, combines negative selection, clonal selection, affinity maturation and memory detector gene bank method, so at to dynamic handle the instruction detection under gridding surroundings. The method includes a dynamic detector evolvement process and a gridding instruction detection process which are based on artificial immunity, which is characterized in by using the artificial immunity technique for reference, and combining the negative selection, clonal selection, affinity maturation and memory detector gene bank method; firstly obtaining an evolvement matured detector; and then dynamically handling the instruction detection problem in the gridding surroundings under the coordination of the artificial immunity mechanism, to complete the entire process of dynamic gridding instruction detection.