258 results about "Neural network identification" patented technology

The invention relates to a visual recognition and positioning method for robot intelligent capture application. According to the method, an RGB-D scene image is collected, a supervised and trained deep convolutional neural network is utilized to recognize the category of a target contained in a color image and a corresponding position region, the pose state of the target is analyzed in combinationwith a deep image, pose information needed by a controller is obtained through coordinate transformation, and visual recognition and positioning are completed. Through the method, the double functions of recognition and positioning can be achieved just through a single visual sensor, the existing target detection process is simplified, and application cost is saved. Meanwhile, a deep convolutional neural network is adopted to obtain image features through learning, the method has high robustness on multiple kinds of environment interference such as target random placement, image viewing anglechanging and illumination background interference, and recognition and positioning accuracy under complicated working conditions is improved. Besides, through the positioning method, exact pose information can be further obtained on the basis of determining object spatial position distribution, and strategy planning of intelligent capture is promoted.

A method controls a welding apparatus by using a neural network to recognize an acceptable weld signature. The neural network recognizes a pattern presented by the instantaneous weld signature, and modifies the instantaneous weld signature when the pattern is not acceptable. The method measures a welding voltage, current, and wire feed speed (WFS), and trains the neural network using the instantaneous weld signature when the instantaneous weld signature is different from each of the different training weld signatures. A welding apparatus for controlling a welding process includes a welding gun, a power supply for supplying a welding voltage and current, and a sensor for detecting values of a plurality of different welding process variables. A controller of the apparatus has a neural network for receiving the welding process variables and for recognizing a pattern in the weld signature. The controller modifies the weld signature when the pattern is not recognized.

The invention discloses a license plate detection method based on a convolutional neural network. The method specifically includes the steps that an Adaboost license plate detector based on Haar characteristics detects license plate images to be detected, license plate roughing regions are acquired, a convolutional neural network complete license plate recognition model recognizes the license plate roughing regions, a final license plate candidate region is acquired, the final license plate candidate region is segmented through a multi-threshold segmentation algorithm, license plate Chinese characters, letters and numbers are acquired, a Chinese character, letter and number convolutional neural network recognition model recognizes the license plate Chinese characters, letters and numbers, and then a license plate recognition result is acquired. License plate images under different conditions can be accurately recognized through the Adaboost license plate detector based on the Haar characteristics and the convolutional neural network complete license plate recognition model, meanwhile, characters are segmented through the multi-threshold segmentation algorithm, character images can be more easily and conveniently segmented, and the good effect is achieved in engineering application.

The invention discloses an automatic insect image identification method based on a depth convolutional neural network. The method comprises the following steps: (1), collecting an original image and carrying out pretreatment to form a training set, and calculating a mean value image of the training set; (2), constructing a depth convolutional neural network; (3), collecting a sub image block randomly from a training sample of the training set and carrying out pre training on the depth convolutional neural network by using the sub image block; (4), training the depth convolutional neural network by using the training set and combining a mini-batch-based random gradient descent algorithm; and (5), carrying out pretreatment on a to-be-measured insect image to form a test sample, and using the trained depth convolutional neural network to identify the test sample after subtracting the mean value image of the training set from the test sample. Therefore, the identification precision is high; the identification types are diversified; the insect within-class change robustness is enhanced; and the insect inter-class similarity sensitivity is high.

The system is presented to recognize visual inputs through an optimized convolutional neural network deployed on-board the end user mobile device [8] equipped with a visual camera. The system is trained offline with artificially generated data by an offline trainer system [1], and the resulting configuration is distributed wirelessly to the end user mobile device [8] equipped with the corresponding software capable of performing the recognition tasks. Thus, the end user mobile device [8] can recognize what is seen through their camera among a number of previously trained target objects and shapes.

The present disclosure relates to systems, non-transitory computer-readable media, and methods for generating and utilizing a touchpoint attribution attention neural network to identify and measure performance of touchpoints in digital content campaigns. For example, a deep learning attribution system trains a touchpoint attribution attention neural network using touchpoint sequences, which include user interactions with content via one or more digital media channels. In one or more embodiments, the deep learning attribution system utilizes the trained touchpoint attribution attention neural network to determine touchpoint attributions of touchpoints in a target touchpoint sequence. In addition, the deep learning attribution system can utilize the trained touchpoint attribution attention neural network to generate conversion predictions for target touchpoint sequences and to provide targeted digital content over specific digital media channels to client devices of individual users.

The invention belongs to the technical field of additive manufacturing and discloses an online detecting and optimizing system for a selective laser melting forming defect. The online detecting and optimizing system comprises a selective laser melting manufacturing platform, an image acquisition module and a processing module; the selective laser melting manufacturing platform comprises a supporting baseplate, a forming device and a laser scanning device; the image acquisition module comprises a three-axis motion platform and a CCD camera; and the processing module comprises an image processing module and an automatic process parameter adjusting module, and the image processing module comprises an image pre-processing module, an image threshold segmentation module, a neural network identifying and counting module and the automatic process parameter adjusting module. By using the online detecting and optimizing system, the surface defect condition of every layer of a part in a selective laser melting manufacturing process can be rapidly detected on line by using a machine vision technology and a digital image processing technology, and no human eye detection limitation exists, so that the accuracy of every defect detection is guaranteed.

The invention relates to the image and text identification field, and specifically relates to a complex image and text sequence identification method based on CNN-RNN. The complex image and text sequence identification method includes the steps: utilizing a sliding sampling box to perform sliding sampling on an image and text sequence to be identified; extracting the characteristics from the sub images obtained through sampling by means of a CNN and outputting the characteristics to an RNN, wherein the RNN successively identifies the front part of each character, the back part of each character, numbers, letters, punctuation, or blank according to the input signal; and successively recording and integrating the identification results for the RNN at each moment and acquiring the complete identification result, wherein the input signal for each moment for the RNN also includes the output signal of a recursion neural network for the last moment and the vector data converted from the recursion neural network identification result for the last moment. The complex image and text sequence identification method based on CNN-RNN can overcome the cutting problem of a complex image and text sequence and the problem that the identification result relies on a language model, thus significantly improving the identification efficiency and accuracy for images and text.

The invention discloses a wireless communication modulating signal identification method based on deep learning for mainly solving the problem that the identification effect depends too much on manualmodulating signal feature extraction in the prior art and improving the drawbacks of low identification accuracy in the case of low signal to noise ratio in the prior art. The method comprises the following steps: sampling captured to-be-identified modulating signals; performing normalization on a sampling sequence obtained by sampling, and drawing a two-dimensional histogram of the modulating signals according to the normalized sampling sequence; constructing a deep convolutional neural network; training the deep convolutional neural network by using training examples; and identifying wireless communication modulating signals by using the trained deep convolutional neural network. By adoption of the wireless communication modulating signal identification method disclosed by the invention, the identification effect of the modulating signals does not depend on the manual feature selection and extraction, and very high identification accuracy is also ensured in the case of low signal tonoise ratio.

The invention discloses a gas insulated substation (GIS) partial discharge online monitoring system and a fault mode identifying method thereof. The monitoring system comprises an ultrahigh frequency sensor, a signal preprocessing subsystem and a data processing subsystem, wherein the data processing subsystem comprises a monitoring and spectrogram display module, an off-limit alarm module and a mode identifying module; and a partial discharge signal acquired by the ultrahigh frequency sensor is transmitted into the data processing subsystem after being processed by the signal preprocessing subsystem so as to realize the monitoring and spectrum display of the partial discharge and the functions of off-limit alarm and fault mode identification. The system and the method have the beneficial effects that by adopting the GIS partial discharge online monitoring system and the fault mode identifying method thereof, the long-term online monitoring function on the partial discharge of the GIS is realized, and meanwhile, the fault type of the partial discharge can be identified well by utilizing the mode that a neural network identifying method based on a fuzzy rough set.

The invention discloses a digital identification method, a digital identification device, a neural network training method and a neural network training device. The digital identification method comprises the steps of acquiring a digital image sample to be identified, and identifying a digital category corresponding to the digital image sample to be identified through a neural network after training. The neural network after training is trained by the steps as follows: acquiring digital detection samples; detecting a preset neural network by using the digital detection samples, wherein the preset neural network is a neural network trained by digital training samples; judging whether the accuracy rate of the detection results reaches a preset threshold; and when judging that the accuracy rate of the detection results does not reach the preset threshold, adjusting the number of nodes output by the preset neural network, and using an error sample in the digital detection samples to retrain the preset neural network to obtain the neural network after training. The technical problem in the prior art that the precision of digital identification is low is solved.

Systems and methods for a sensor hub system that accurately and efficiently performs sensory analysis across a broad range of users and sensors and is capable of recognizing a broad set of sensor-based events of interest using flexible and modifiable neural networks are disclosed. The disclosed solution consumes orders of magnitude less power than typical application processors. In one embodiment, a scalable sensor hub system for detecting sensory events of interest comprises a neural network and one or more sensors. The neural network comprises one or more dedicated low-power processors and memory storing one or more neural network programs for execution by the one or more processors. The output of the one or more sensors is converted into a spike signal, and the neural network takes the spike signal as input and determines whether a sensory event of interest has occurred.

The invention relates to an optimal codebook design method for a voiceprint recognition system based on a nerve network. The optimal codebook design method comprises following five steps: voice signal input, voice signal pretreatment, voice signal characteristic parameter extraction, three-way initial codebook generation and nerve network training as well as optimal codebook selection; MFCC (Mel Frequency Cepstrum Coefficient) and LPCC (Linear Prediction Cepstrum Coefficient) parameters are extracted at the same time after pretreatment; then a local optimal vector quantization method and a global optimal genetic algorithm are adopted to realize that a hybrid phonetic feature parameter matrix generates initial codebooks through three-way parallel algorithms based on VQ, GA and VQ as well as GA; and the optimal codebook is selected by judging the nerve network recognition accuracy rate of the three-way codebooks. The optimal codebook design method achieves the remarkable effects as follows: the optimal codebook is utilized to lead the voiceprint recognition system to obtain higher recognition rate and higher stability, and the adaptivity of the system is improved; and compared with the mode recognition based on a single codebook, the performance is improved obviously by adopting the voiceprint recognition system of the optimal codebook based on the nerve network.

The invention discloses a content-based human body upper part sensitive image identification method and a content-based human body upper part sensitive image identification device. The method comprises the following steps of: performing human face identification on static images and removing images which do not contain human face information; identifying a texture image by using neural network identification and finding out related sensitive information positions; generating a human body skin color model of the image by using the color distribution information in a detected human face area and the default skin color information; extracting the human body skin area from the image according to the established human body skin color model; and if the sensitive information image and the human face information image are consistent with a human body upper part shape model, and the skin ratio of the sensitive information image exceeds a threshold, determining the image as the sensitive image. In the invention, human face identification, human body upper part identification and sensitive image identification are used in combination to differentiate sexy photo images from pornographic sensitive images, so that the error report rate of the sexy photo images is effectively reduced.

The invention discloses a fake plate detection method based on license plate identification and vehicle feature matching. The method comprises the following steps that: extracting a monitoring equipment frame image, and carrying out graying on a source image; adopting Sobel edge detection to position a license plate; adopting a morphological processing image to enable regions to be communicated soas to bring convenience for extracting the outline of the license plate; setting an aspect ratio to accurately extract areas; through hough transformation and vertical projection, carrying out license plate correction and character segmentation; using a neural network to identify segmented characters to obtain license plate information; migrating an AlexNet neural network frame, and carrying outclassification through the identification of the depth feature of the color; and applying a KNN (K-Nearest Neighbor) algorithm to be combined with database system information to detect a fake plate situation. By use of the method, vehicle identification accuracy is guaranteed, a high-accuracy convolutional neural network is directly migrated to serve as a basic framework, cost and expenditure aresmall, the method can be quickly realized on a computer platform, cost is small for arranging a license plate identification and vehicle identification system on a large scale, and feasibility is high.

The invention discloses a polyp image identification system. The system comprises an image processor, a video collector and a plurality of program modules, wherein the program modules include an image obtaining module, an image identification module, an algorithm processing module and a prompt processing module; the image obtaining module is used for splitting a video into a plurality of static images frame by frame; the image identification module is used for introducing the static images into a deep convolutional neural network identification engine to obtain a pixel-level probability graph of a plurality of identification targets; the algorithm processing module is used for performing targeted optimization on the input probability graph to remove environmental interference except main target features so as to judge the position of a polyp; and the prompt processing module is used for marking the judged position of the polyp. The invention furthermore discloses a polyp image identification method of the polyp image identification system. The polyp image identification system and method has high sensitivity and high specificity at the same time; the position of the polyp in an endoscopic image can be accurately identified; and the missing identification and false identification rates of polyp detection are remarkably reduced.

The invention relates to a data drive control method for minimum energy consumption of a refrigerating system on basis of SPSA. The method includes the steps of adjusting frequency of a compressor to enable chilled water supply water temperature to be constant according to changes of a system load so that refrigerating capacity can be matched with a thermal load; obtaining a relation curve between the system load and the minimum stable superheat degree of an evaporator; establishing an online neural network identification model of the system; calculating the system load according to changes of refrigerating capacity of the refrigerating system of an air conditioner under dynamic regulation of the compressor, obtaining the minimum stable superheat degree corresponding to the system load according to the relation curve between the system load and the minimum stable superheat degree of the evaporator, and using the minimum stable superheat degree as a set value of the superheat degree of the evaporator; establishing a neural network controller; completing control over the superheat degree of the evaporator through an expansion valve control loop. The method is easy to calculate and implement, the number of parameters is small, and the control effect is good.

The invention provides an in-depth learning-based partial discharge ultrasonic audio frequency identification method and system. The method comprises the following steps: partial discharge ultrasonic signals of an electric power device are detected, partial discharge ultrasonic audio frequency data is obtained and then converted into a sound spectrogram, a depth convolution nerve network model is built, samples are used for training a network, partial discharge ultrasonic audio frequency data to be diagnosed is input into a trained network, and a partial discharge defect type is output and obtained. According to the in-depth learning-based partial discharge ultrasonic audio frequency identification method and system, the partial discharge ultrasonic audio frequency data is converted into the sound spectrogram, a depth convolution nerve network is used for identifying the sound spectrogram, ultrasonic signals of all kinds of defects in partial discharge can be accurately and effectively identified, and a convenient and reliable diagnosis method is provided for electric power device insulation state assessment.

The invention discloses a focusing processing method and apparatus, a device and a storage medium. The method comprises the following steps: performing main body identification on a picture collectedby a camera module at present by using a preset neural network to determine a target main body contained in the currently collected picture and the position information of the target main body; determining a current corresponding target shooting mode of the camera module according to the type to which the target main body belongs; and controlling the camera module to perform focusing on the targetmain body according to the position information of the target main body, and shooting an image containing the target main body in the target shooting mode, wherein the preset neural network is trained by using all types of shooting main bodies as samples. By adoption of the method, the target main body in the collected picture is identified by using the neural network and automatic focusing is performed on the target main body according to the position of the target main body, thereby not only improving the quality of the shot pictures and effectively avoiding the phenomenon that the shot pictures are fuzzy, but also reducing manual operations, facilitating the use of users and improving the user experience.

The invention belongs to the technical field of detection of performance of electric transmission line insulators, and discloses a method for automatically identifying hydrophobic grades of composite insulators. The method includes the steps of image enhancement, image filtering, image segmentation, image feature quantity extraction, neural network identification model establishment and the like. By analyzing hydrophobic images of the composite insulators, automatic identification of the hydrophobic grades of the composite insulators is achieved. Human factor influences are eliminated, the method is high in accuracy of the judgment result, easy to operate and capable of automatically identifying the hydrophobic grades of the insulators different in voltage grade, type, manufacturer, foul degree grade, operation age limit and the like.

The invention discloses a guided filter and auto-encoder-based SAR target recognition algorithm. Aiming at the problem that the time is consumed as the recognition carried out on synthetic aperture radar (SAR) image targets via neural networks pursues high recognition rate and complex structures are designed, the algorithm applies a rapid image fusion technology to feature extraction of an SAR recognition technology, and uses a weighted guided filter (GF) to carry out two-scale fusion on the SAR images so as to generate one-dimensional image vectors, uses an auto-encoder to carry out low-dimensional feature reconstruction on the images, and uses a softmax classifier to carry out classification; and an image fusion technology of the weighted guided filter and feature extraction are combinedthrough experimental simulation and verification, so that the target recognition precision is improved, the quantity of hidden layer neurons of the auto-encoder is greatly decreased and the calculation complexity is greatly reduced.

The invention discloses a particle filter and radial basis function (RBF) identification-based neural network proportion integration differentiation (PID) control parameter self-setting method used for a control system, the object model of which is unknown and the interference of which is on-linear and non-Gaussian noise. The method comprises the following steps of: connecting the output of a PID controller and the system output to the input of an RBF neural network identification structure respectively, and connecting a particle filter part between the system output and the RBF neural network identification structure; and filtering the system output by using particle filter to obtain particle filter output, training the RBF neural network by using the difference value of the particle filter output and the RBF neural network output as a target function to obtain the RBF neural network output, then calculating Jacobian information of the system, finally training a neuron by using the deviation signal between the system reference input and the system output as a target function, guiding the neuron by using the Jacobian information, and adjusting the PID controller by a learning algorithm. At the same time of keeping the characteristics of high PID control robustness, good reliability and the like, the method can further improve the dynamic response performance and the interference resistance of the control system.

The present invention belongs to the field of machine vision, and relates to an intelligent identification method and apparatus for a numerical value of a pointer instrument. The present invention provides an identification method and apparatus, wherein a model is trained to intelligently identify a numerical value of a pointer instrument based on an anti-sample model training method by using a convolution neural network model, and a camera shooting angle adaptive adjustment technology is provided. The apparatus has a shooting angle adaptive function and can self-correct a shooting angle, without the need for performing precision adjustment on the apparatus before use. In the method, an anti-sample training convolution neural network identification model is adopted, so that the workload for original data acquisition is greatly reduced; and numerical values of various instruments can be identified at the same time, so that there is no need using multiple cameras to shoot separately, so that cost is saved. The method and apparatus provided by the present invention have the characteristics of high robustness, high accuracy, fast identification, strong operability and convenient transplantation and the like.

An embodiment of the invention provides a breast mass automatic detection method and system. The method includes acquiring a to-be-detected breast image and acquiring a mass image from the breast image, wherein the candidate mass image is a part of sub images in the breast image; obtaining a detection result about whether a breast position corresponding to the candidate mass image has a mass or not by taking the candidate mass image as the input of a pre-constructed mass recognition model. The method can implement segmentation of the to-be-detected image more accurately. At the same time, by utilizing a neural network recognition model constructed in advance according to a large amount of sample data for recognizing the image subjected to segmentation, the breast mass is acquired. The method provided by the invention has beneficial effects of good adaptability and accurate detection.

The invention provides an artificial lateral line pressure detection method which comprises the following steps: step 10, selecting a streamlined underwater robot, and determining the pressure trace of the underwater robot by a simulation mode; step 20, arraying pressure sensors along the pressure trace and around an axial line; step 30, putting the underwater robot in different water flow states, acquiring the pressure data of the current state by utilizing each pressure sensor, and uploading the pressure data to an analysis center; step 40, using the analysis center to import the pressure data into a neural network identification tool, and training to obtain a neural network model with an identification ability; and step 50, importing the acquired underwater data into the neural network model so that the flow field / motion state of the position can be determined. According to the invention, a set of artificial lateral line system is developed by utilizing the MEMS technology, and machine learning is carried out through the pressure data acquired by the sensors in the system, thereby realizing development of the ability of the underwater robot for environment identification.

The invention relates to an UHV DC transmission line fault cause identification method based on multi-source information fusion, which comprises the steps of: acquiring electrical quantity fault dataand non-electrical quantity information of an UHV DC transmission line; extracting electrical characteristic quantities of the UHV DC transmission line, and constructing an electrical characteristic input vector; extracting non-electrical characteristic quantities of the UHV DC transmission line, and constructing a non-electrical characteristic input vector; constructing a comprehensive neural network identification model for lightning strike, mountain fire, foul, wind deviation and bird damage respectively, wherein a number of layers of hidden layers of a neural network model is selected by adopting a method with minimum error; a self-learning method is adopted for identifying specific fault causes. The UHV DC transmission line fault cause identification method combines the electrical quantity information and the non-electrical quantity information, searches for the characteristic law, fuses the multi-source information by using the fusion property of the neural network, utilizes theidea of a big data method, utilizes the multi-source fault information, combines with a neural network algorithm to perform identification of the causes, and fully guarantees the precision of cause identification.

The embodiment of the invention provides an image content identification method and device and a terminal. The method includes: inputting sample images to a convolutional neural network in a process of training the convolutional neural network, wherein the sample images are used for iteratively training the convolutional neural network; determining the number of already passed training iterationsof the convolutional neural network; adjusting a loss function on the basis of the number of the already passed training iterations to obtain a target loss function; carrying out iterative training according to the target loss function to obtain a target convolutional-neural-network; and carrying out content identification on a to-be-identified image through the target convolutional-neural-network. Through the convolutional-neural-network training scheme provided by the embodiment of the invention, distribution of the complex image samples can be better fitted, the number of sample images of intermediate probability value distribution can be decreased, and thus a recall rate of samples can be increased in a case of ensuring an identification result accuracy rate of the convolutional neuralnetwork.

The invention discloses a method for identifying an electroencephalogram image based on a deep convolutional neural network. The method comprises the following steps: carrying out baseline eliminationpreprocessing on a collected motor imagery electroencephalogram signal; Dividing each lead signal into a plurality of time windows, and carrying out fast Fourier Transform on each window MI-EEG model, carrying out fast Fourier inverse transform on the EEG models respectively, and calculating corresponding time domain power values of the EEG models; Calculating a mean value of the time domain power values obtained by each window to obtain time domain power characteristics; Performing interpolation imaging on the extracted three-frequency-band power characteristics in a data matrix to obtain apseudo RGB image of the MI-EEG signal; Designing the DCNN model into five segments of convolution, and after each segment of convolution is finished, replacing a maximum pooling layer with a convolution layer to carry out data dimension reduction; And performing evaluation on the test set by using the trained DCNN model to complete a classification test. MI-EEEG images have the advantages in the aspect of feature expression, and are matched with 30 layers of DCNN with higher model fitting capability, which has great significance for the improvement of the MI- EEG signal feature expression andclassification precision .