102results about How to "Improve target recognition ability" patented technology

The invention provides a synthetic aperture radar target identification method based on diagonal subclass judgment analysis, which mainly solves the problem that the prior synthetic aperture radar has poor target identification performance. The method comprises the following processes: the self-adapting threshold segmentation, the morphological filtering, the geometric clustering operation and the pretreatment of image enhancement are carried out for an original image; the optimal subclass division to each target after pretreatment is carried out by adopting a two-dimension rapid global K-means clustering algorithm; the diagonal subclass judgment analysis or the diagonal subclass judgment analysis and two-dimension subclass judgment analysis are used for finding out an optimal projection matrix; training and testing images after pretreatment are projected towards the projection matrix to obtain characteristic matrixes of the training and testing images; the Euclidean distance between a testing target and the characteristic matrix of each training target is calculated, and the category attribute of the testing target is determined by adopting a nearest neighbor rule. Simulation experiments show that the invention has the advantages of good effect of inhibiting background clutter, high quality of the target image and low characteristic dimensionality and can be used in a remote sensing system.

The invention discloses a multiple-subarea matching method constraint by a space relation. A plurality of subareas are selected in a matching template as sub-templates, and the sub-templates are separately and simultaneously matched with to-be-matched images. Space position relations among the sub-templates are utilized to realize the accurate matching of the images. The method comprises the following steps: the plurality of subareas are selected in the matching template respectively as the sub-templates; the space relations among the sub-templates are determined; the plurality of sub-templates keep the space position relations unchanged to form combined templates; matching is carried out by utilizing the mobile searching performed by the combined templates on the to-be-matched images, and a plurality of similarity values of the combined templates are obtained; the plurality of similarity values are compared, and a position where a maximum similarity value exists is regarded as an optimum matching position to complete the matching. The method of the invention utilizes the mutual space relations among the plurality of sub-templates to reach the requirements for accuracy and precision of matching. The object identifying performance of the method is substantially improved in real-time performance compared with the conventional large-template gray scale or contour matching algorithms.

The invention discloses an automatic millimeter wave image target identification method and a device. The method comprises steps that (1), sub image blocks of a to-be-identified target millimeter wave image are acquired; (2), based on a trained convolutional neural network, target containing probability values of the sub image blocks are acquired; (3), based on the probability values, a probability cumulative graph of the to-be-identified target millimeter wave image is acquired; and (4), based on the probability cumulative graph, target marking is carried out so as to accomplish target identification of the to-be-identified target millimeter wave image. The invention further discloses an automatic millimeter wave image target identification device. The device and the method are advantaged in that the device and the method are suitable for automatic millimeter wave image target identification, the excellent target identification effect is realized, a problem that employing traditional manual design characteristics for the millimeter wave image can not realize the excellent detection effect in the prior art is solved, precise target positioning is realized, false alarms are reduced, safety check efficiency is improved, and manpower cost is reduced.

The invention belongs to the technical field of radars, and particularly relates to a class aggregation true and false target one-dimensional range profile feature extraction method. According to the method, same-class target features are more compact, different-class target features are hugely separated, defects of a conventional subspace discrimination method are overcome, even if serious overlap happens to different target data distribution, good recognition effects can still be acquired, and the radar true and false target classification performance is effectively improved.

The invention discloses a radar target tracking recognizer based on DSP, which comprises a transmitter, a receiver, a digital signal acquisition unit and a digital signal processor, and is used for receiving a digital difference frequency signal transmitted by the digital signal acquisition unit, and carrying out fast Fourier transformation FFT to acquire a power spectrum of the difference frequency signal and acquire the radial distance of a radar target according to the position of the power spectrum of the difference frequency signal. Compared with the prior art, the radar target tracking recognizer based on the DSP can realize quick and real-time detection and tracking for a moving target, enhances the target recognition performance of radar, does not need to adopt a special chip, is favorable for lowering the production cost, and has important productive practice significance.

The invention discloses an infrared detecting device and method for target recognition in a sea surface sun bright band. According to the infrared detecting device, the lenses of two middle infrared cameras form an angle; an infrared polarizer is arranged in front of the lens of a middle infrared camera; a wire gating polarization device is arranged at the light path intersection in front of the lenses of two middle infrared cameras; a diaphragm is arranged on the light path in front of and the wire gating polarization device; a germanium sealing window is arranged in front of the diaphragm; the germanium sealing window and a temperature control hood form a sealed temperature control cavity; and the middle infrared cameras, the infrared polarizer, the wire gating polarization device and the diaphragm are arranged in the temperature control cavity. When detection is carried out, the system of the infrared detecting device is calibrated; the image of the sea surface sun bright band is collected; and target recognition is finally realized after image processing. According to the invention, the image gray scale difference between the target and the surrounding sea area is improved; equal gray-scale massive sea area appearing in the bright band sea area is reduced; and the target recognition ability in the sea surface sun bright band is improved.

The invention relates to an image local feature extracting method on the basis of a Hilbert curve and the LBP (length between perpendiculars). In the method, the Hilbert curve is adopted to arrange LBP features of an image according to a certain sequence and the sequence can ensure an obtained feature vector to comprise a space neighboring relation among feature points so as to improve the identification performance of a mode identification system. The local feature extracting method particularly comprises the following steps of: a first step of acquiring an input object and carrying out pretreatment of filtering, removing noise and the like; a second step of extracting the LBP features; and a third step of selecting the Hilbert curve and extracting a high-order feature. The method disclosed by the invention is easy to implement, only relates to simple difference, binaryzation and feature ordering and has low complexity, i.e. low calculation complexity of both difference and binaryzation.

The invention discloses a multi-lens camera device and a monitoring system, and belongs to the technical field of the imaging. The multi-lens camera device comprises multiple lens groups, each lens group comprises multiple lenses arranged in a line, the multiple lenses included by each lens group have the same focal length, and the same focal length is the focal length of the lens group; multiplelens groups are orderly arranged on a vertical direction, and the focal lengths of at least one lens groups are different; and the pictures shot by multiple lenses included by each lens group can be spliced into a complete picture. Through the multi-lens camera device disclosed by the embodiment of the invention, since the different lens groups with different focal lengths cover different regionsin the monitored range, the targets distributed on different distances can present sharp images, thereby performing effective target identification in large depth of the monitored range, namely, the identification capacity on a monitored target is effectively improved when the large-range monitoring is satisfied.

The invention discloses a neighborhood characteristic space discriminant analysis based radar target identification method. Each class of radar target data is divided into training samples and testing sample; the within-class neighboring characteristic spaces and inter-class neighboring characteristic spaces of the training samples are established, and vertical vectors from sample points to the within-class neighboring characteristic spaces and the inter-class neighboring characteristic spaces and weighted values of the vertical vectors are calculated; within-class scattering matrixes and inter-class scattering matrixes of all the training samples are established, transformational matrixes from a high-dimensional radar target data space to low-dimensional characteristic sub-spaces are solved, all the training samples and testing samples are transformed from the high-dimensional radar target data space to the characteristic points in the low-dimensional characteristic sub-spaces according to the obtained transformational matrixes to complete characteristic extraction; a nearest neighbor method is adopted to classify the characteristic points of the testing samples to complete radar target identification. The method can effectively improve the learning capability of the sub-spaces, improve the radar target identification performance under the conditions of limited training samples and is low in computation burden.

The invention provides a sparse point cloud multi-target tracking method fusing spatio-temporal information, and belongs to the field of 3D vision. According to the method, the point cloud feature extraction network is used as a trunk, multiple frames of point cloud data are input at the same time, time domain information fusion is carried out on extracted features, and therefore missing detectioncaused by point cloud sparseness is avoided. Benefited from the fusion of time-space information, tracking and detection tasks can be more tightly coupled, and detection frames of three frames beforeand after can be predicted at the same time to obtain a track segment of a current target for three continuous frames. Then a distance intersection ratio score of the current trajectory segment and the trajectory tracking result at the previous moment is calculated, and the currently split trajectory segment and the historical trajectory segment are matched and spliced by using a greedy algorithmto obtain a final trajectory tracking result at each moment. The method has the application potential of coping with multi-target tracking under sparse point clouds, has high robustness for target missing detection and false detection, and can still obtain a stable tracking result in sparse point cloud sequence input.

The invention discloses a target identification method and a target identification system for an SAR (synthetic aperture radar). The method comprises the following steps that the complex image data of a tested SAR image is subjected to SCM (subaperture cross-correlation magnitude) target signal-to-noise ratio improvement processing; the processed image data is subjected to threshold division, and then, a target outline image of the tested SAR image is determined; after the Fourier description sub features are extracted on the target outline images, the feature vector of the tested SAR image and the feature vector of each training SAR image are obtained, and the similarity between the tested SAR image and each training SAR image is calculated; the target category of the training SAR image with the highest similarity to the tested SAR image is identified into the target category of the tested SAR image. The method and the system provided by the invention have the advantages that according to the technical scheme, the target identification rate is improved through improving the image data quality, and the identification speed is also accelerated, so the target identification performance is improved.

The invention discloses an SAR target identification method based on a Bayes multinuclear learning support vector machine. The objective of the invention is to solve a problem of inaccurate SAR image target identification in the current target identification method. The method comprises following steps of 1) inputting an original SAR image and carrying out preprocessing to calculate nuclear matrixes of different characteristics; 2) according to the multinuclear learning method, combining the nuclear matrixes; 3) establishing a Bayes multinuclear learning support vector machine model for a support vector machine according to the combined nuclear matrixes; 4) using the expectation maximization algorithm to solve the Bayes multinuclear learning support vector machine model to obtain an optimal solution; and 5) using the optimal solution to carry out target identification on SAR image test data. According to the invention, by effectively combining the deduction capability of the Bayes method and the distinguishing capability of the multinuclear learning method, the identification performance is improved and the method can be used for classification of SAR images.

The invention discloses an unmanned aerial vehicle detection single station. The single station comprises a detection device, a direction finding device, a data processor and a detection terminal. Thedata processor is connected to the detection device, the direction finding device, and the detection terminal in a communication connection mode. The detecting device is used to scan the radio data of a monitored area to collect and obtain the original data of an unmanned aerial vehicle signal, and transmit the original data of the collected unmanned aerial vehicle signal to the data processor. The data processor is used for extracting and analyzing radio signal characteristic information in the original data in real time, extracting the parameter information of the unmanned aerial vehicle signal and transmitting the parameter information to the direction finding device. The direction finding device carries out direction finding processing and transmits a direction finding result to the detection terminal for displaying through the data processor. The unmanned aerial vehicle detection single station can realize the detection control of the passive unmanned aerial vehicle of non-voluntary filing, and the timely detection and discovering, the rapid positioning, the effective tracking and the track control of long-distance illegal unmanned aerial vehicle can be realized without affecting the electromagnetic environment of a monitoring area.

The invention discloses a sand body thickness prediction method based on spectrum inversion and a sand body thickness prediction system thereof. The prediction method comprises the steps that seismic wavelet and stratum reflection coefficient are acquired from original seismic data; an objective function is established, and spectrum inversion is performed according to the objective function, the seismic wavelet and the stratum reflection coefficient after parity decomposition so as to acquire an even component reflection coefficient sequence; new seismic data are synthesized according to the seismic wavelet, the even component reflection coefficient sequence and the objective function; and the sand body thickness is predicted according to the new seismic data. The spectrum inversion result resolution is better than that of the conventional inversion so that the thin layer geological target identification effect can be enhanced, the operation cost is low and the application effect is great.

The invention discloses a sparse learning RCS sequence feature extraction method, belonging to the neighborhood of radar target recognition technology. At first, sparse learning is carried out on thetrain sample, Using the sparse coefficient as the adjusting factor of the distance between the features of samples, A sparse learning projection matrix is established, which automatically selects a small number of neighborhood samples through sparse learning, and can better preserve the local structure information of the sample neighborhood, thereby improving the performance of target recognitionand overcoming the shortcoming of the traditional projection method that only global structure features can be extracted.

A system and method for registering 3D data sets is disclosed based on manual fiducial selection. The technique is useful in imaging obscured targets with 3-D imaging laser radars. For such an exemplary method, which defines a three-dimensional linear shift vector for each data voxel, four fiducials are required to completely define the mapping for a 3D space. An exemplary registration algorithm as disclosed provides an approach to automatically make fine adjustments to the 3D data registration. The tedious technique of shifting data sets relative to each other, in many degrees of freedom, is eliminated. Instead, a fine adjust is applied to the digital mapping function, through fiducial perturbation.

The invention discloses a radar object distance image recognition method based on kernel discrimination local tangent space arrangement, and the method comprises the following steps: obtaining a plurality of training distance images in a known target type and test distance images in a to-be-tested target type; carrying out the preprocessing of the obtained training distance images and the test distance images; obtaining a mapping matrix V from a high-dimensional distance image space to a low-dimensional feature space based on the rules of local tangent space reconstruction error minimization and inter-class scattering maximization; extracting the features of the training distance images and the test distance images: y=VTk, wherein VT is the transposition of the mapping matrix V, and k is the kernel vector of one of the training distance images or the test distance images; comparing the features of the training distance images and the test distance images through employing a nearest neighbor method, enabling each test distance image to be divided into the target class where the nearest training distance image belongs.

The invention provides a radar reflection signal processing device. The device comprises a multiplexing module, a plurality of signal processing modules, a demultiplex module and a trace point extraction module. The multiplexing module is used for receiving multiple beams of reflection signals and used for distribution of the reflection signals. The signal processing modules are used for processing one beam of reflection signals. Each signal processing module comprises a data obtaining module used for receiving one beam of reflection signals and detecting whether information of the reflection signals is lost or not, an impulse interference filter module used for impulse interference filtration, a Doppler processing module used for Doppler filtration, a clutter rejection module used for clutter rejection processing, a mixed constant false alarm detection module used for estimating the noise level, and a noise matching module used for judging whether the noise level is matched or not. The demultiplex module is used for integrating the multiple reflection signals into a data flow. The trace point extraction module is used for generating a trace point picture. In this way, the radar reflection signal processing device can process multiple beams of reflection signals at the same time, and effectively improve the processing ability and speed of radar reflection signals.

The invention discloses an automatic detection and extraction method for low and medium frequency line spectrums in ship radiation noise. The method comprises: acquiring signals; accumulating frequency spectrums at multiple moments; smoothing the accumulated frequency spectrum; carrying out smoothing item removal on the logarithm decibel spectrum; extracting line spectrum; if the line spectrum ishigher than the line spectrum threshold H, reserving the line spectrum, otherwise, removing the line spectrum; and carrying out traversal comparison on amplitudes of all frequency points of the frequency spectrum to obtain a final line spectrum extraction result. According to the method, the line spectrum characteristics of the target ship can be automatically detected and extracted by collectingand analyzing the radiation noise of the ship. The method is suitable for weak line spectrum detection under the condition of low signal-to-noise ratio and non-low signal-to-noise ratio, has obvious advantages compared with an existing line spectrum detection method, effectively improves the target feature automatic extraction capability, optimizes the underwater sound target recognition effect, and improves the intelligent level of underwater sound equipment.

The invention relates to a machine learning model training method and device and a face recognition method and device, and relates to the technical field of artificial intelligence. The training method comprises the following steps: extracting a feature vector of each image sample data by using a machine learning model, and determining a classification result of a target in each image sample dataaccording to each feature vector; judging whether the targets in the image sample data are classified into correct classifications or wrong classifications according to the labeling results and the classification results of the targets in the image sample data; under the condition that the target in any image sample data is classified into the wrong classification, reducing the first feature similarity between the image sample data and the correct classification, and increasing the second feature similarity between the image sample data and the wrong classification; and calculating a loss function for training a machine learning model according to all the reduced first feature similarities and the increased second feature similarities.

The invention discloses a heterogeneous distributed detection information target identification optimization method based on threat assessment, which is applied to an attack and defense countermeasuresystem simulation system. The method is characterized in that a heterogeneous distributed sensor network is formed based on signal-level semi-physical systems such as multiple satellites and multipleradars, the property of an attacking target is evaluated to form a threat sequence, and a ground command control system schedules the satellites and the radars to preferentially track and identify ballistic missile targets with high threat values according to the threat values in the threat sequence, and finally an interception or striking sequence is determined according to a target identification result. According to the method, redundant and complementary information fusion is carried out on collected detection information, target and environment information is collected and processed to agreater extent, accuracy and reliability of battlefield target identification are improved, and therefore the method is of great significance in subsequent situation evaluation, threat estimation andinterception strategy formulation, and the winning probability in confrontation simulation is greatly improved.

The invention discloses a target recognition method based on compressed sensing. The method includes the steps of acquiring standard sample maps of at least two types of targets; obtaining characteristic atoms of the types of targets using a characteristic atomic extraction method; arranging each characteristic atom of each type of target diagonally into a dictionary [psi]p of each type of target, and arranging the dictionaries of the types of targets in parallel into a comprehensive dictionary [psi]; performing compressed sampling on an original image x to be recognized by the by using a measurement matrix [phi] to obtain a compressed sampled signal y; combining the comprehensive dictionary [psi], the measurement matrix [phi] and the sampled signal y, and calculating a sparse coefficient [theta] of the original image to be recognized through reconstruction; and processing the sparse coefficient [theta] to obtain a coefficient map, and performing classification and counting according to the number of rows and the size of the connected domain in the coefficient map, to achieve the recognition of the types of the targets in the original image. The target recognition method proposed in the invention can improve the speed of target recognition in the image and realize multi-target recognition.

The invention relates to a recognition method for finding dynamic target in air in real time, and deeper features can be automatically extracted and learned. According to the technical scheme, target activity rules, aerial target data feature expression, behavior rule discovery and knowledge accumulation based on hierarchical classification are realized, and an aerial target behavior feature structure system platform is constructed; aiming at the target data of which the behavior intention is mastered and target behavior rule analysis, a target behavior rule model which are composed of a data preprocessing module, a track mining module and a track element statistics module can be constructed; the data preprocessing module is used for screening target numbers and cleaning target behavior data, and the track mining module is used for calculating classic tracks and track similarity for each classified track to generate classic track data; and the track element statistics module determines the position and the movement track of a target by means of a sequence image of the target, longitudinally compares the movement of the target, and obtains a saliency detection target of the image in a frequency domain.

The invention discloses a generalized grouping detection method which includes the following steps: S1. generalized point trace generation; S2. generalized grouping detection criteria; and S3. generalized grouping detection algorithm. The generalized point trace enables grouping detection to have selectivity, which increases anti-interference capability. The generalized grouping detection criteria enable the grouping detection to be flexible, which increases object identification capability. The generalized grouping detection algorithm enables the grouping detection to have generability, which increases situation cognition capability. The generalized grouping detection algorithm is not only applicable to narrow band radar object tracking and broadband radar object tracking, and is also applicable to object tracking of visible light sensors, infrared sensors and multispectral sensors, etc. The generalized grouping detection algorithm is applicable to single object tracking, and is also applicable to multi-object tracking. The generalized grouping detection algorithm is applicable to dense multi-object tracking and is also applicable to sparse multi-object tracking. The generalized grouping detection algorithm is applicable to group object tracking of physical domain and is also applicable to group object tracking of functional domain. The generalized grouping detection method can increase capabilities of exploration systems for objects on land, on air, at sea, near space and outer space, etc, and can simplify commanding and controlling systems.

The invention discloses a small sample remote sensing image target identification method based on metric learning, and belongs to the technical field of remote sensing image identification. The methodcomprises the following steps: searching remote sensing image data, and preparing a remote sensing image target slice data set for small sample target identification; performing feature extraction onthe multi-scale remote sensing image target slice by using a deep residual network to obtain a feature vector of a test target and a feature vector set of various training samples; finding the firstk feature vectors similar to the feature vector of the test target in the feature vector set of each type of training samples based on a k-nearest neighbor algorithm; finally, based on a metric learning method, calculating the similarity between the test target and various training samples, and completing identification of the test target. The method has the characteristics of simple network structure, simple training method, high average target recognition accuracy and the like, and is suitable for recognition application of remote sensing image targets in small sample scenes.

The system comprises a collimating mirror, a micro-lens array, a relay imaging system and an area array detector which are sequentially arranged along the optical axis direction of incident light, a polaroid array is arranged between the collimating mirror and the micro-lens array, and a static interference system is arranged between the micro-lens array and the relay imaging system. The collimating mirror, the polaroid array, the micro lens array and the static interference system are coaxial; the polarizing film array divides an incident light field into four polarization channels, so that each polarization channel forms a transmission light field with different polarization states; and the static interference system is used for enabling transmission light fields with different polarization states to form a multi-polarization interference light field so as to realize synchronous measurement of polarization interference of four polarization channels. According to the invention, the size and weight of the whole system are effectively reduced, and the microminiaturization and integration of the infrared polarization snapshot imaging spectrometer can be realized while the polarization state, imaging and optical path difference are precisely modulated.

The invention discloses a target recognition training system and method based on an EEGNIRS. The target recognition training system comprises a signal acquisition cap, an electroencephalogram signal amplifier, a near-infrared brain function imager, a computer, a training system interface, a P300 decoding unit and an NIRS analysis unit. A trainee wears a signal acquisition cap and opens a trainingsystem interface; after training is started, the display interface randomly displays a target picture and a non-target picture, a trainee watches the display interface, and the signal acquisition capacquires an electroencephalogram signal of the trainee, amplifies the electroencephalogram signal and transmits the amplified electroencephalogram signal to the P300 decoding unit for decoding, so that a target recognition rate is calculated, and the recognition capability is further evaluated; and an NIRS analysis unit in the computer is used for quantifying the brain fatigue degree of the trainee and adjusting the frequency of picture presentation in the display interface. According to the invention, multi-stage progressive training can be realized, the image presentation frequency can be adaptively adjusted according to the brain fatigue degree of a trainee, the target recognition rate can be analyzed according to P300 electroencephalogram, and the ability of the trainee can be evaluated.

The invention discloses a ship target recognition method based on feature migration, and belongs to the field of ship target recognition. According to the method, the problem of poor recognition effect of the to-be-recognized target due to the fact that the to-be-recognized target is different from the known trained target data in the aspects of appearance and imaging quality in an existing methodis solved. According to the invention, HOG features of ship images with different resolutions are extracted; and, according to the transfer learning method based on spatial alignment and probabilityadaptation, the HOG features of the source domain and the HOG features of the target domain are mapped to the same feature space, then probability adaptation and instance weight adjustment are carriedout in the same feature space, and a new source domain vectorization HOG feature and a new target domain vectorization HOG feature are regenerated, the support vector machine is trained by using thenew source domain vectorization HOG feature, and target recognition is performed on the to-be-recognized image by using the trained support vector machine. The method can be applied to identificationof the ship target in the remote sensing image.