69results about How to "Reduce the impact of recognition" patented technology

The invention relates to an ultrasonic human face distinguishing method and a distinguishing device. The distinguishing method comprises the following steps: (1) an ultrasonic signal is transmitted to an object to be distinguished; (2) an echoed signal is collected; eigenvector is drawn from the echoed signal; (3) according to the eigenvector obtained from the step (2), the distinguishing result is obtained from distinguishing and comparing identities in the established information database of the ultrasound human face; the distinguishing device comprises an ultrasonic sounder, an ultrasonic receiver, a feature extraction module, an ultrasound human face information database and an identify distinguishing module. The invention has the main advantages that very high spatial resolution is obtained; ample human face information can be extracted; the influence of the background on distinguishing humane face can be reduced; a 3D model and the human face can be separated; the deception of a picture and a video can be overcome; the data quantity can be reduced; the distinguishing speed can be increased; the device has higher discrimination; the needed ultrasonic human face database is characterized by little data quantity and is convenient for establishing large scale ultrasonic human face database.

The invention provides a method for recognizing and collecting microseism events in a well and belongs to the field of microseism monitoring. The method comprises the steps that (1) orientation rotation is conducted on each receiving point recorded in the even monitoring process; (2) for records obtained after rotation is conducted, a characteristic function value CFs of each receiving point at each moment is calculated; (3) an event superposition energy function L(t) is calculated through a time difference scanning technique; (4) the seism starting time ti of events is automatically detected through self-adaptive threshold values; (5) the peak value time of S waves of a set of events is collected and checked, and the jumping time of the S waves is collected and checked; (6) for each receiving point, a characteristic function value CFp is calculated for each time point within a fixed time interval Tspmax before the jumping time of the determined set of S waves; (7) the peak value time of each receiving point of P waves of the set of events is collected and checked, and the jumping time of the P waves is collected and checked.

The invention discloses a face expression identification method based on a partially shielded image. The method comprises the following steps that 1) face images of N types of expressions included in a known type of face expression library are preprocessed; 2) the similarity between a sample to be tested in a test set and a training sample in a training set is calculated, and an image in the same type with the sample to be tested and nearest to the sample to be tested is obtained; 3) a shielded part of the sample to be tested is reconstructed; 4) PWLD features are extracted from the reconstructed sample to be tested and the training sample in the training set; and 4) an SVM classifier is used to identify all the samples to be tested in the test set in a classified manner. According to the invention, an image matching method is used to effectively reconstruct the shielded part of the image, the problem that features cannot be represented completely when only non-shielded parts are extracted is solved, a three-layer pyramid structure is used to extract global and local features of the image, and the accuracy of feature representation is improved.

The invention discloses a greenhouse intelligent mobile robot vision navigation path identification method. Original image information is converted from an RGB color space to an HSI color space, H, S and I component information pictures are extracted respectively, the H component information picture is subjected to denoising processing, a K-means algorithm is used to carry out clustering segmentation on the H component information picture, and the segmentation effect picture of a road is obtained. A morphological erosion method is employed to carry out secondary denoising processing, the picture which is subjected to the erosion process is subjected to gray conversion, and the complete road information is obtained. The Candy operator edge detection is employed, edge discrete points are extracted, navigation discrete points are converted and acquired, the navigation discrete points are fitted to obtain final navigation path information which is subjected to coordinate transformation, and the navigation angle of a mobile robot is calculated. The robustness of navigation path identification to illumination inequality is effectively raised, the subsequent image processing operation is facilitated, and the rapid real-time performance of the whole path identification system is raised.

The invention provides a device parameter online identification method based on multi-section hybrid measurement information. The method includes the steps of firstly, filtering multi-section device measurement data; secondly, acquiring a device model; thirdly, judge whether the device model is a power transmission circuit or a transformer winding, performing the fourth step if the device model is the power transmission circuit, and otherwise, performing the fifth step; fourthly, identifying and estimating power transmission circuit parameters; and fifthly, identifying and estimating transformer parameters. SCADA (supervisory control and data acquisition) measurement information of a scheduling technology support system and phase measurement information of a WAMS (wide area measurement system) are fully utilized to build a hybrid measurement system, suspicious device parameters is estimated through multi-time sections, and identification accuracy of device parameters is improved.

The invention relates to a ground moving vehicle target classification and recognition method and system based on a high-resolution distance image, and the method comprises the steps: carrying out the averaging and energy normalization of multiple collected continuous original HRRP echoes of a ground moving vehicle target; calculating the HRRP echo power spectrum features according to the HRRP echoes after averaging and energy normalization; calculating the distances between the HRRP echo power spectrum features and various types of preset power spectrum feature templates, and obtaining a plurality of distances; comparing the distances, and determining the class of the ground moving vehicle target according to a comparison result. The method calculates the distances between the HRRP echo power spectrum features and various types of preset power spectrum feature templates through a nearest neighbor classifier, recognizes the class information of the ground moving vehicle target through comparing the distances between the HRRP echo power spectrum features and various types of preset power spectrum feature templates, achieves the precise classification and recognition of the ground moving vehicles, and provides support for the classification and recognition of military moving vehicles on the ground.

The invention discloses a fundus illumination multi-disease detection system based on a regional feature set neural network. The system comprises a computer memory, a computer processor and a computerprogram which is stored in the computer memory and can be executed on the computer processor, wherein a trained multi-fundus disease detection network model is stored in the computer memory, and themulti-fundus disease detection network model comprises a feature extraction network, a semantic segmentation sub-network and a plurality of classifiers; when the computer processor executes the computer program, the following steps are realized: obtaining a to-be-detected original fundus picture, inputting the to-be-detected original fundus picture into the multi-fundus-disease detection network model, obtaining the probability of each type of lesion, and realizing fundus lesion identification. According to the invention, a single model can be used for detecting various fundus diseases, and the detection precision is high.

The invention provides a method and a system for identifying traffic signs in a street scene image. The method includes the following steps: receiving a street scene image, preprocessing the street scene image, and enhancing the traffic sign features of interest in the street scene image; extracting the traffic sign features of interest in the street scene image by combining a HIS color space threshold and a non-color region threshold; using shape features to optimize and screen the traffic sign features of interest extracted based on color features, and removing false regions in the street scene image; and sending the traffic sign features of interest to a multilayer neural network, identifying the features of interest, and outputting an identification result. According to the invention, the street scene image is processed, the traffic sign features of interest are extracted, and traffic signs are identified using a trained multilayer convolutional neural network. The identification rate is improved, and the influence of the factors of the street scene image on identification is reduced.

The invention provides a method and device obtaining wonderful moment frames in game direct broadcasting; the method comprises the following steps: S1, obtaining a position of a target image in game direct broadcasting, wherein the target image containing wonderful moment broadcast words is obtained from the game direct broadcasting in advance; obtaining a contrast image from the position of the present frame in the game direct broadcasting; S2, using a perception Hash algorithm to obtain a Hash value of the target image and the contrast image; S3, if the distance between the hash values of the target image and the contrast image is smaller than a first preset threshold, using the present frame on which the contrast image is located as the wonderful moment frame. The method and device can real time obtain wonderful moment frames on one hand, and can only use the target image containing wonderful moment broadcast words to identify on the other hand, thus reducing influences by the background on the wonderful moment frame identifications, and improving the wonderful moment frame obtaining precision.

The invention relates to a people number counting method combined with convolution neural network and track prediction. The method comprises following steps of using the frame difference method to segment crowd blocks included in a video; distinguishing a sparse crowd block mass and a dense crowd block mass; for the sparse crowd block mass, converting an RGB image through a formula so as to obtainHSV color spaces; in two different color spaces, using a selection searching algorithm to pre-determine pedestrian positions; combining and removing repeated regions in the two spaces so as to obtainpedestrian region positions; using the convolution neural network to extract characteristics and selecting grid loss function Grid Loss to train the network based on the blocks, thereby achieving recognition of local positions including faces and bodies of shielded pedestrians; for the dense crowd block mass, extracting characteristics of a crowd density distribution graph, establishing a multi-regression model and estimating people number; and using the Markov model chain to predict walk tracks of the shielded pedestrians, locking the positions of the shielded pedestrian and counting the pedestrians.

The invention discloses a spectroscopic imaging method for multi-frequency electrical impedance tomography which can be used for early detection of cerebral stroke and belongs to the technical field of electrical impedance tomography. The spectroscopic imaging method includes firstly, using data at all frequency positions in a frequency band for imaging, and reconstructing electrical impedance tomography spectral images reflecting tissue impedance change along with frequencies according to impedance spectral characteristics of tissues; secondly, according to impedance spectral specificity of the tissues, obtaining independent electrical impedance images from the electrical impedance tomography spectral images so as to separate cerebral stroke tissues from normal tissues; finally, selecting the electrical impedance images capable of reflecting the cerebral stroke tissues according to spatial information and position information of the cerebral stroke tissues. The spectroscopic imaging method is capable of detecting focuses of the cerebral stroke from experiments which are based on a real cranium brain structure so as to lay the foundation for applying the multi-frequency electrical impedance tomography to rapid early detection of the cerebral stroke.

The invention discloses a transfer learning picture classification method and device based on principal component analysis, and the method comprises the following steps: S100, carrying out the featurevector dimension reduction of each sample in a small sample data set through employing a principal component analysis method, and obtaining a feature vector after dimension reduction; S200, traininga full connection layer neural network in the deep convolutional neural network by using the feature vectors after dimension reduction to obtain a classifier; S300, extracting feature vectors of the to-be-classified pictures, projecting the feature vectors of the to-be-classified pictures to a low-dimensional space to obtain a dimension reduction result, inputting the dimension reduction result into a classifier, and outputting the classification result; and the total number of each type of samples in the small sample data set is less than or equal to 15. According to the method, the sample feature vectors are projected to the low-dimensional space, so that the sample density is increased, the influence of noise on image recognition is reduced, and the classification accuracy is improved.On the other hand, the invention further provides a transfer learning picture classification device based on principal component analysis.

The invention provides a license plate recognition method and device and electronic equipment. The method comprises the following steps: obtaining coordinate information and coarse classification information of the license plate in the target image based on a YOLO model; obtaining a license plate area image of a license plate in the target image by using the coordinate information, correcting thelicense plate area image based on an STN model, and obtaining a character recognition result in the corrected license plate area image by using an attention model; further, determining fine classification information of the license plate based on the character recognition result and the coarse classification information. According to the license plate recognition method and device, the license plate area image is corrected through the deep learning method, the influence of the imaging quality of the license plate on license plate recognition is reduced, more accurate coarse classification information and character recognition results are obtained accordingly, fine classification information is obtained through further processing on the basis, and therefore the recognition rate of license plate recognition is increased.

The invention provides a speech recognition method, intelligent equipment and an intelligent television. The method comprises the steps of separating original speech data into one or more paths of speech data; determining a confidence level of each path of separated speech data; and performing speech recognition on the speech data with the highest confidence level. According to the method, influences of noise on the speech data can be reduced, the problem that at present speech recognition accuracy is not high due to existence of interference sound such as noise and the like is solved, the speech data can be accurately recognized, and more effective human-computer interaction can be performed.

The invention discloses an automatic detection system device for apparent diseases of a highway tunnel and a subway tunnel. The device comprises a carrier vehicle provided with a power system, a positioning system and an identification system, an image acquisition system, and a lighting system. The power system drives the carrier vehicle to move forward; the positioning system acquires and recordsdisease location information of the tunnel lining; and the identification system recognizes a tunnel lining crack picture and calculates the opening and length of a crack. The image acquisition system collects apparent image data of the tunnel lining; and the lighting system covers the tunnel lining surface by light and eliminates a shadow caused by the uneven surface of the tunnel lining. Therefore, the simultaneous detection of the tunnel lining crack and the leakage water diseases is realized; and the detection equipment has characteristics of high monitoring efficiency, less lighting shadows, high detection accuracy, non-contact mode and the like.

The embodiment of the invention discloses a wild animal image automatic recognition system based on big data and deep learning. The system takes a wild animal monitoring image database as a main body,and achieves processing and analysis of wild animal data through combining a super-resolution image recovery technology and a wild animal automatic recognition algorithm; a front-end interface of thesystem is built through data visualization. Specifically, the wildlife monitoring image database is used for storage management of wildlife images and the related information, the super-resolution image recovery technology is used for achieving recovery and reconstruction of low-quality compressed images, and the automatic recognition algorithm is used for realizing automatic recognition and monitoring analysis of the wildlife images through deep learning; therefore, classification processing of the images and elimination of invalid images are realized. According to the method, efficient classified storage of wild animal images and automatic analysis and visual display of wild animal data can be realized, and the intelligent level of wild animal image management is improved.

The invention discloses a method and system for processing a signal of a channel self-adaptation dynamic peak artificial cochlea. The method comprises the following steps frequency division processing is carried out on a transmitted sound signal, and the signal is decomposed into A frequency bands; the A frequency bands are combined into N channels according to the number N of usable electrodes, and envelope energy of each channel is stored at the same time; the target dynamic range of each channel is calculated, and the envelope energy of each channel is adjusted to be in the target dynamic range; the N channels are ranked according to the size of the envelope energy of the adjusted channels, and M channels with the largest energy are selected; sound-electric stimulus compression is carried out on the envelope energy of the M channels, and the channel needing to be stimulated among the M channels is determined, and the stimulus information and energy of the channel needing to be stimulated are transmitted. Therefore, by means of the method and system, the effect of background noise on recognition can be effectively reduced, and a normal audition environment effect can be guaranteed.

The invention discloses a clothing attribute tag identification method based on fine classification of a neural network model, which comprises the following steps: (1) cutting a collected clothing picture by using a target detection algorithm Faster RCNN, and removing redundant irrelevant backgrounds; (2) separating different background type pictures into a tiled picture and a model picture; (3) carrying out data augmentation on the obtained picture, including reflection, rotation and random stretching operation, increasing the number of pictures, and obtaining augmented training data; (4) taking the obtained augmented training data as a training set, training a NasNet improved multi-classification network, and training an inception resnetv2 improved multi-classification network; and (5) fusing the network models trained by NasNet and inception netv2, and outputting the attribute tags of the clothes by adopting a weighted average fusion method. The method can be widely applied to application scenarios such as clothing image retrieval, tag navigation and clothing matching.

The invention belongs to the technical field of advanced aided driving. On the basis of a deep convolutional neural network theory, the invention discloses a traffic identification method based on a color space fusion network and a space transformation network. Traffic sign images of an RGB color space, an HSV color space, a YIQ color space and a grayscale image space are fused by constructing thecolor space fusion network so as to enrich features extracted by a convolutional neural network; improving the spatial invariance of the convolutional neural network to the input image by using a spatial transformation network; the global maximum pooling is used to replace the full connection layer to reduce the parameter quantity of the convolutional neural network. Experiments show that the traffic sign recognition method based on the color space fusion network and the space transformation network designed by the invention can obtain a higher recognition rate under the condition of fewer parameters.

The present disclosure provides a double-station labeling and sorting system. The system includes a conveyor, a console, a printer, a visual identification device, a controller and a tray storage cabinet; the conveyor is used for transmitting a to-be-labeled tray; the console is used for manual labeling operation; the printer is used for printing a label; the visual identification device is used for identifying the to-be-labeled tray and transmitting the information of the to-be-labeled tray to the controller; the controller is used for sending instructions to the printer and the visual identification device and receiving and converting information transmitted by the visual identification device; and the tray storage cabinet is used for storing a labeled tray.

The invention discloses a protein function module identification method for a label propagation algorithm on the basis of edge driving. The protein function module identification method comprises thefollowing steps: S1, randomly distributing a unique integral value m as the tag of an edge to an absolute value E of edges in a PPI (Protein-Protein Interaction) network; S2, according to a defined label update rule, through iteration, changing the label of each edge in the PPI network until the label in the network is not changed. According to the protein function module identification method, the interference of noise in the PPI network can be reduced, the stability of an algorithm identification process is improved, and therefore, a more accurate and effective protein function module division result is obtained in the larger-scale PPI network.

The invention discloses an SSVEP electroencephalogram signal recognition method based on MVMD-CCA. The method can decompose an electroencephalogram signal into a plurality of multivariate modulation components, reduces the impact of non-correlated brain activity and artifacts in the electroencephalogram signal, and improves classification precision. The method comprises the following steps: collecting a multichannel SSVEP electroencephalogram signal as an electroencephalogram signal to be recognized; and setting the number K of to-be-decomposed components, constructing a variational problem, solving the variational problem by adopting an ADMM algorithm, and decomposing the electroencephalogram signal to be recognized into K multivariate modulation components; defining a reference signal according to the visual stimulus frequency for inducing the electroencephalogram signal to be recognized; solving a weighted correlation coefficient of the specific stimulus frequency fi; and taking thefrequency corresponding to the maximum weighted correlation coefficient as the inductive stimulus frequency of the electroencephalogram signal to be recognized.

The invention relates to a detection and identification method for spores in gynecological micro-ecology, equipment and a storage medium, and the method comprises the following steps: inputting a to-be-detected image into an effective target AI detection model, detecting a corresponding target object, and secondarily determining whether the target object is a spore through a machine vision technology. The effective target detection AI model generation method comprises the following steps: collecting a multi-scene microscopic image; carrying out manual category labeling according to a conventional method; expanding the labeling area of the epithelial cell nucleus by introducing hyper-parameters; and making an effective training set with a large inter-class gap and a small intra-class gap, and inputting the effective training set into a convolutional neural network for training based on a backbone network architecture and in combination with a deep learning target detection framework to obtain an effective target AI detection model. According to the method, an effective target detection AI model can be trained by introducing a hyper-parameter to expand a labeling area of an epithelial cell nucleus; Whether the target object is the spore or not is secondarily determined through a machine vision technology, so that the false detection rate and the omission ratio of the spore are effectively reduced.

The invention discloses a tray positioning detection method and device and an intelligent forklift. According to the method, an active light source is adopted to irradiate a scene; stable image features can be obtained; enhancing environmental anti-interference capability, according to the method, the images of the scene with the external light source and the scene without the external light source are subjected to image difference operation and binarization processing, the influence of different image backgrounds on recognition can be effectively reduced, and the recognition efficiency and the recognition success rate of the intelligent forklift on the tray can be remarkably improved, so that the working efficiency and the intelligent degree of the forklift are improved.

The invention discloses a facial expression recognition method based on an inter-class difference strengthening network, and the method comprises the following steps: collecting a data set, and carrying out the preprocessing of the data set; analyzing an expression similarity relationship; constructing a parallel branch network by utilizing similarity information, and respectively extracting global features and fine-grained features similar to distinguishing similar expressions; the extracted features are respectively sent to a full connection layer for dimension reduction, feature fusion is carried out, and expression categories are output through a classifier; and adding a class balance weighted loss function so as to enlarge the class spacing. According to the method, expression similarity information is fully utilized, fine-grained features are extracted, and the face expression classification effect is effectively improved.

A deep neural network and an underwater acoustic target voiceprint feature extraction method, the deep neural network includes an input layer, a hidden layer and an output layer, used for the extraction of underwater acoustic target voiceprint features, and the number of nodes in the input layer is the underwater acoustic target signal The number of frequency points of the original signal spectrum, the number of frequency points of all frequencies within the value range of the fundamental frequency and the sum of harmonic orders, the number of nodes in the output layer is the number of frequency points of the original signal spectrum, and the number of nodes in the hidden layer is less than the number of nodes in the input layer number; the underwater acoustic target voiceprint feature extraction method includes the signal acquisition step, the fundamental frequency and harmonic acquisition step and the reconstruction step. The accurate extraction of the fundamental frequency and harmonics and the reconstruction of the original signal spectrum weaken the noise line spectrum contained in the original signal spectrum, purify the original signal spectrum, and reduce the impact of the interference line spectrum on the final identification of the final ship target individual. And can adapt to frequency drift.

The invention discloses a human face living body detection method and device. The method comprises the steps of collecting a to-be-detected human face image; performing feature extraction processing on the to-be-detected face image to generate a plurality of feature images; inputting the to-be-detected face image and the feature image into a face living body detection classification model obtainedthrough machine learning training in advance, and outputting a face living body detection result; performing feature extraction processing on the to-be-detected face image to generate a plurality offeature images, which comprises: performing illumination normalization processing on the to-be-detected face image to generate an illumination normalization processing image; performing feature extraction processing on the to-be-detected face image by adopting an LBP algorithm to generate a texture feature image; converting the face image to be detected into an HSV space from an RGB color space, and generating an HSV image; and performing DCT on the to-be-detected face image to generate a frequency spectrum image. The method is short in detection time and high in human face living body detection precision, and the influence of illumination on recognition can be reduced.

The invention discloses a transformer substation equipment oil leakage detection method and system based on deep learning and belongs to the technical field of substation equipment. The system comprises a photo acquisition module, a model training module, a photo preprocessing module, an equipment oil leakage defect identification module and an information sending module, wherein the photo acquisition module acquires to-be-detected equipment images sent by terminal equipment, after being preprocessed in the photo preprocessing module, the images are inputted into an equipment oil leakage defect recognition model pre-trained by a model training module to obtain the equipment oil leakage defect information in the images, and then the equipment oil leakage defect information is sent to the terminal equipment through the information sending module to achieve defect recognition of the power equipment. The system is advantaged in that problems that an existing transformer substation intelligent inspection robot is prone to being affected by environmental factors in the inspection process, and consequently inspection accuracy and working efficiency are low are solved.

A license plate recognition system and a recognition method are disclosed, the invention relates to the field of image recognition technology, which solves the problem that the dynamic image capturedby the camera will be blurred in different degrees due to the influence of the external environment, and the license plate information in the dynamic image is difficult to be recognized. The technicalscheme essentials are as follows: obtaining the vehicle image of the recognized vehicle; extracting gradient information of gray change in the vehicle image and forming a corresponding gradient imageaccording to the gradient information; processing the histogram of the gradient image and adjusting the contrast of the gradient image; highlighting a license plate area in the gradient image according to an inherent gray change characteristic of the license plate; identifying vehicle information in the vehicle area according to an inherent geometric feature of a license plate. The vehicle information is presented by a display device, and the license plate information can be recognized quickly and accurately under the influence of adverse environment, and the recognition efficiency of the license plate information is improved.