95results about How to "Improve target recognition rate" patented technology

The invention discloses an image identification automobile live-action navigation device and method. The device comprises a video analysis module, an image identification module, a GPS (global positioning system) module, a route planning module, a map management module and a live-action navigation management module, wherein the video analysis module is connected with the image identification module, and the live-action navigation management module is respectively connected with the image identification module, the GPS module, the map management module and the route planning module. Automobile navigation equipment provides map display and route marking, images are identified according to images acquired by an automobile data recorder, particularly, traffic signs of variable lanes are identified, a user is reminded of selecting turning lanes or direct-through lanes when approaching an intersection and reminded of driving or parking according to actual traffic signal conditions, and live-action navigation is realized. The new function of automobile navigation is added, and user experience is enhanced.

The invention discloses a radar high-resolution range profile target identification method based on a two-dimensional convolutional network. The radar high-resolution range profile target identification method comprises the steps of: determining Q different radars, wherein a target exists within detection ranges of the Q different radars, then acquiring Q-type high-resolution range imaging data from high-resolution radar echoes of the Q different radars, dividing the Q-type high-resolution range imaging data into a training sample set and a test sample set, and recording the Q-type high-resolution range imaging data as original data x; calculating to obtain data x'' '' after short-time Fourier transform according to the original data x; setting a two-dimensional convolutional neural network model which comprises five layers, and constructing the two-dimensional convolutional neural network model by using the training sample set and the data x'' '' after short-time Fourier transform, soas to obtain a trained convolutional neural network; and performing target identification on the trained convolutional neural network by using the test sample set, so as to obtain a radar high-resolution range profile target identification result based on the two-dimensional convolutional network.

The invention discloses a quick target identifying method and system based on compressed sensing. The method comprises steps of conducting spectrum dimensionality reduction via a spectrum selection method, sampling a target image and conducting spare transformation via a compressed sensing method, conducting compressed sensing reconstruction based on an orthogonal matching pursuit method, and identifying via an SVM classifier. Sampling and spare processing are conducted via the compressed sensing algorithm can reduce data post-calculation complexity, and can break through limitations of sampling frequency; spectrum selection and target identification algorithm are combined; with the high-spectral feature identification technology, unit target detection and multiple imaging detection can be combined; with the utilization of spectrum technology and imaging technology, the detection unit has a target identifying capacity, so target identifying rate can be improved; and target identifying time can be reduced and strong anti-jamming capability can be achieved.

The invention discloses a low-altitude short-range cluster cooperative defense system and defense method. The system comprises an early-warning monitoring module, an aerial counter-measure module, a ground counter-measure module and a control center. Data acquired by the early-warning monitoring module is transmitted to the control center to form visualization air situation, when an illegal low-slow small flyer invades, the early-warning monitoring module identifies, follows and monitors the illegal flyer and indicates an azimuth and height of the illegal flyer, and the control center drives the aerial counter-measure module or the ground counter-measure module to handle the illegal flyer. The control center transmits a control instruction to the aerial counter-measure module or the ground counter-measure module and can also process information returned by an aerial unmanned counter-measure module or a ground unmanned counter-measure module in real time. The aerial unmanned counter-measure module repels or hits a target by virtue of a carried task load. The ground unmanned counter-measure module repels and hits the target or effectively suppresses the low-altitude flyer by virtue of a carried task load.

The invention discloses a complex environment adaptive active radar system based on a big data analysis technology, which comprises a radar system front-end module, an information integration controlmodule, a depth learner module, an auxiliary parameter extraction module and a data center module, wherein the radar system front-end module receives a waveform instruction sent by the information integration control module, a radar transmitting signal is generated, a radar echo signal is then received, and a radar image is formed; the depth learner module receives the radar image, the radar imageis subjected to target recognition, and target recognition result information is obtained; the auxiliary parameter extraction module obtains target parameters according to the radar image; and the data center module is used for integrating online network information and priori information for storing a typical target environment. While requirements on the prior knowledge of the target and the target environment are reduced, the detection and tracking performance of the radar system towards a weak target in a complicated environment can be effectively improved.

The invention discloses a linkage control system for a small radar and a photoelectric turntable. The linkage control system comprises a small radar unit, a photoelectric turntable unit and a computer linkage control unit; the small radar unit and the photoelectric turntable unit are arranged at the front end; the computer linkage control unit is arranged at the rear end; the small radar unit comprises the small radar and a small radar data analysis module; the photoelectric turntable unit comprises the photoelectric turntable and a photoelectric turntable video acquisition module; a system control module, an image fusion processing module and multiple communication modules are arranged in the computer linkage control unit; the system control module is connected to an alarm module; and the image fusion processing module is connected to an electronic map. Compared with the prior art, synchronous display of detected targets on the electronic map can be carried out through linkage of the small radar, the photoelectric turntable and the electronic map; therefore, the target identification rate, the target display control and the early warning means of a security and protection system can be greatly improved.

The invention discloses a middle propeller cavitation noise numerical prediction characteristic extraction method based on inhomogeneous inflow. The method includes the following steps: first conducting mesh generation on a propeller computational domain and checking mesh quality and defining boundary conditions; then setting a computational model in computational fluid dynamics (CFD) software, conducting stable iterative computation to obtain water dripping performance parameters and inflow opening speed to verify the model accuracy; conducting non-stable iterative computation by utilizing stable computation as the initial value of the non-stable computation in the CFD software and displaying the cavitation period state of a propeller piece and cavitation area change of a recording piece through post processing; finally computing the propeller cavitation radiation noise through the cavitation area of the propeller piece according to a single-vacuole radiation noise theory to conduct characteristic extraction.

The invention relates to the technical field of digital signal processing, and discloses an ultralow-sidelobe pulse compression method. The method specifically comprises the following steps: S1, a transmitter designs a weighted window function w(t) according to a transmitted linear frequency modulation signal s (t), and converts the transmitted signal into s(t)*w(t); S2, a corresponding pulse compression matched filter h(t) and a corresponding sidelobe suppression filter w(t) are designed at a receiver according to the transmitted signal, wherein the sidelobe suppression filter w(t) is designed according to the weighted window function w(t); S3, the pulse compression result y(t)=[s(t)*w(t)]*[h(t)*w(t)] of an echo signal is calculated according to the matched filter h(t) and the sidelobe suppression filter w(t) at the receiver; S4, the weighted window function at the transmitter is cancelled, an equivalent filter hw(t) is designed at the receiver, and an equivalent filter hw(n) is calculated according to a frequency domain implementation method; and S5, the transmitter transmits the linear frequency modulation signal s(t), the receiver designs a pulse compression equivalent filter hw(n), and ultralow sidelobe is obtained. By adding window functions at the transmitter and the receiver respectively, sidelobe is reduced effectively.

The invention provides an SAR (Synthetic Aperture Radar) image target characteristic extraction and identification method based on KFDA (Kernel Fisher Discriminant Analysis) and an SVM (Support Vector Machine). The method comprises the following steps: performing amplitude data normalization processing on a training target sample of a known type and a testing target sample of an unknown type; performing characteristic extraction on the normalized training target sample of the known type and the testing target sample of the unknown type respectively by using a KFDA criterion; training an SVM classifier by using training target sample characteristics of known types extracted according to the KFDA criterion to generate an optimal classification face; identifying the characteristics of the testing target sample of the unknown type extracted according to the KFDA criterion through the optimal classification face. By adopting the method, the requirement on a preprocessing process is lowered, the target-aspect sensitivity of an SAR image is avoided, the dimensions of sample characteristics are compressed, and high target identification rate is obtained. The method has high popularity.

The invention provides a target monitoring method for an airport field monitoring system on the basis of video recognition. A video target is matched with a monitored target by a target recognition and location technology and a technology of converting projections of video coordinates and projections of coordinates of the field monitoring system, and the target recognition rate of the field monitoring system is improved.

The invention discloses an SAR image noise suppression method based on joint sparse representation and residual fusion and mainly solves the problems of insufficient speckle noise suppression and poor detail keeping of the conventional SAR image noise suppression method. The method disclosed by the invention comprises the following steps: (1), performing block-matching on images to obtain a similar set; (2), performing local variance estimation on the images; (3), performing joint sparse representation on the similar set through local variance and the WSOMP method, so as to obtain a sparse coefficient, and calculating a residual set; (4), performing residual fusion on the residual set, and performing noise reduction through the wavelet soft threshold algorithm to obtain the fused residual; (5), updating a dictionary through the fused residual and the sparse coefficient; (6), performing image reconstruction on the similar block set through the undated dictionary to obtain a de-noised block set; (7), returning the de-noised block set to the original positions of the images, so as to obtain de-noised images. The SAR image noise suppression method obviously improves the SAR image speckle noise reduction effect and can be used for SAR image target recognition and image enhancement.

The invention discloses an unmanned aerial vehicle multi-scale target detection and recognition method, which is high in recognition rate, small in calculation amount and high in robustness. The method is realized through the following technical scheme: forming a real-time target detection and identification network model by adopting three parts, namely a trunk structure for extracting different scale features, a network neck (Neck) and a network detection head for predicting target information, performing feature extraction on an input image by adopting a CSPDarknet53 trunk network in a target detection and identification network-YOLOv4, and expanding the original three-scale feature output into four-scale feature output; using an improved two-branch PANet to reduce the number of convolution layers through which features pass; predicting a conditional probability value for each category by each detection output, directly obtaining a prediction result from the picture, and obtaining target information; and transmitting the four feature maps with different sizes to a detection head for joint training, and performing category judgment and position regression on the unmanned aerial vehicle target to obtain a detection and recognition result.

The invention discloses a boat, ground and vehicle combined monitoring method. Searching load equipment on a bag body of a moored boat is used for conducting region searching, tracking and recognizing load equipment on an unmanned aerial vehicle is used for conducting close tracking and recognizing on a target searched by the load searching equipment, and recognition results are sent back to a ground monitoring station. The invention further discloses a boat, ground and vehicle combined monitoring system which comprises the moored boat, the unmanned aerial vehicle, ground equipment and the load equipment. The ground equipment comprises a multifunctional truck and the ground monitoring station. The load equipment comprises the searching load equipment and the tracking and recognizing load equipment. The moored boat carries the searching load equipment for achieving long-term hover and region searching; the unmanned aerial vehicle carries the tracking and recognizing load equipment for achieving rapid departure and close investigation; and the multifunctional truck enables the moored boat, the unmanned aerial vehicle and the load equipment to be integrated, rapid and mobile deployment is achieved, and the ground monitoring station is used for achieving information fusion and commanding and coordinating work, so that a combined monitoring function of the boat, ground and vehicle combined monitoring system is achieved.

The invention discloses a range profile data extrapolation method based on extraction of a sparse scattering center, and aims at solving the problem that the target identification result is low when the amount of samples is relatively few in the prior art. The method comprises that (1) angle enframing is carried out on obtained multi-HRRP data; (2) ISAR imaging is carried out on each frame of multi-HRRP data, and a position set of the scattering center is obtained according to ISAR images; (3) a dictionary is generated according to the position set of the scattering center; (4) the dictionary is used to express each frame of data; (5) the sparsity of each frame of data is solved to obtain a target parameter combination of the scattering center; (6) a multi-HRRP sample which is extrapolated after each frame is obtained according to the target parameter combination of the scattering center; and (7) IFFT is carried out on each frame of data to obtain a time-domain real HRRP sample which is extrapolated after each frame, and the time-domain real HRRP sample is used for subsequent target identification. The method of the invention can substantially improve the target identification rate and be used for radar HRRP identification.

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 a SAR image target identification method based on a depth increment support vector machine. The method comprises steps of (1), inputting a SAR image; (2) pre-processing the image; (3) constructing an initial training set and an increment training set; (4), initializing the depth increment support vector machine; (5), calculating an initial identification rate of a test sample; (6), updating the depth increment support vector machine; and (7), calculating an identification rate of the test sample. According to the method, advantages of increment learning and depth learning are combined, the depth increment support vector machine is employed to carry out target identification of the SAR image, an increment training sample can be processed, and the depth target information with better discrimination property can be acquired. The method is advantaged in that SAR image target identification precision is improved, and the training time is shortened.

The invention provides a full Bayes features extraction method for synthesizing an aperture radar object, which can adaptively estimate the quantity of the attribute scattering centers and parameters of the attribute scattering centers. The full Bayes feature extraction method for synthesizing aperture radar object comprises steps of receiving a attribute scattering center model of a radar echo signal, determining distribution of posterior probability that each unknown characteristic parameters is subordinated to, establishing a whole layered Bayes model, determining unite posteriori possibility distribution of all unknown parameters in the attribute scattering center model, determining suggestion distribution of each unknown parameter, performing sampling on each unknown characteristic parameter, obtaining N sampling values of each unknown characteristic parameter, generating a Markov chain of the unknown characteristic parameter according to the N sampling values of each unknown characteristic parameters, and determining an estimation value of each unknown characteristic parameter.

The invention belongs to the digital image object recognition optimization method technical field and relates to a K-means clustering-based target recognition method. The invention aims to solve problems such as low recognition rate and high possibility of being affected by environments of a traditional sea-surface target recognition algorithm. According to the K-means clustering-based target recognition method of the invention, the RGB color space of an original image is converted into an HLS color space, so that influence on target recognition caused by sea surface, sky background color and the brightness of light can be eliminated; the clustering center of a classifier is gradually optimized through an ant colony algorithm; and therefore, target recognition rate can be improved, and the robustness of the algorithm can be enhanced.

The invention provides an SAR object identification method based on range profile time-frequency diagram non-negative sparse coding. The method utilizes non-negative sparse coding, and in the whole identification process, SAR image objects do not need azimuth angle estimation, thereby reducing the identification complex degree, avoiding the dependence of identification accuracy on object azimuth angle estimation, and helping to improve the object identification rate; and meanwhile, radar object identification is carried out based on the range profile time-frequency diagram non-negative sparse coding technique, and good identification performance is also achieved in the noise environment, thereby not influencing identification effect when not-high image quality, due to factors of defocusing or signal to noise ratio and the like during the object moving, is caused, and helping to improve the robustness performance of the radar object identification.

The invention relates to an SAR target classification method for pre-training a complex full convolutional neural network, and the method comprises the steps: S1, improving a complex convolutional auto-encoder structure according to a complex full convolutional neural network structure; S2, training a plurality of convolution auto-encoders by adopting training samples; sS3, initializing a plurality of fully convolutional neural networks by using encoder parameters in the trained plurality of convolutional auto-encoders, and further finely tuning the plurality of fully convolutional neural networks by using the training samples; S4, classifying the test samples by adopting the trained complex full convolutional neural network. According to the SAR image target recognition method based on the complex convolutional neural network, for the overfitting problem caused by the small number of training samples in SAR image target recognition based on the complex convolutional neural network, the complex convolutional auto-encoder is adopted to pre-train the complex full convolutional neural network, and the high target recognition rate is obtained through a small number of training samples.

The invention provides an improved One-Class SVM (Support Vector Machine) algorithm based multi-mode composite detection target identifying method. The method is carried out through the following steps: on the pretreatment of the signals outputted from a passive millimeter wave detector, a long infrared detector and an FMCW radar in steadily scanning state, selecting the features with high distinguishing degree as classification features; using the improved One-Class SVM algorithm based algorithm to train the training data; constructing a classifier with a high target identification rate; extracting the features of a to-be-tested signal; sending the processed feature data to the classifier for target identification; and obtaining the determination result afterwards. According to the invention, the method considers the rapid development of the existing high speed real time signal processing system and the large amount of information to be processed in composite detection, and starting from the perspective of feature layer integration, uses a One-Class SVM algorithm featuring the sample and feature weights, which greatly improves the target identification rate of a multi-mode composite detector.

The invention provides a radar target recognition method based on a dpKMMDP model, and is used for solving the technical problem of low recognition rate of the existing radar target recognition method based on a classifier model. The realization steps are that 1. a radar SAR image training sample set X is extracted; 2. the kernel function matrix G of the training sample set is calculated; 3. the dpKMMDP model is constructed and joint condition posterior distribution of all the parameters of the model is calculated; 4. condition posterior distribution of all the parameters is calculated; 5. the initial value of each parameter is set; 6. cyclic sampling of all the parameters is performed, and the sampling result of the parameters required by the T0 time test phase is saved; 7. a test sample set X<^> is extracted and the hidden variable set Z<^><n> of the test sample set X<^> is calculated; and 8. the hidden variable set Z<^><n> is substituted in an LVSVM classifier to calculate and output the target category number of the test samples. Classification of complex nonlinear classifiable radar SAR image data can be realized, and the method is used for recognition of radar targets.

The present invention provides a method for distinguishing nonlinear projection of one-dimensional distance image which belongs to field of radar target recognition. Random two sorts targets in multiple targets are classified into one group, one-dimensional distance image of each group target is processed nonlinear transform and mapped to high-dimensional linear characteristic space, a nonlinear projection plane transform matrix is build in high-dimensional linear characteristic space, characteristic is obtained and classified by minimum-distance criterion, and the sorts input target belonged to is determined finally by voting mechanism. Steps: random two sorts of targets in training target are matched into one group; matrix WAlpha is determined by kernel function and radar target one-dimensional distance image training vector matrix Pi and (K)ij in the group; nonlinear projection vector of input target one-dimensional distance image xt is determined; Euclidean distance between nonlinear projection vector and library template vector is determined; sort number of input target one-dimensional distance image is determined; sort number with most stat votes is sort belonged to input target. The present method can improve target identification performance efficiently.

The invention discloses a fixed area night fog all-weather color video imaging system and a building method thereof. The system comprises a visible light video collection system, a visible light image collection module, a background image updating module, a visible light background image bank, an infrared light video collection system, an infrared light image collection module, an image preprocessing module, an infrared light background image bank, an infrared light target detection module, an image registration module, a video fusion module, a video signal emitting module, an RTSP transmission module, and a display module. The system uses an improved algorithm integrating a mixed Gaussian model and a symmetric difference method as well as an imaging concept of fusion of the infrared video and visible light background image. After processing by a data processing center, the system can output a clear color video image at night, in fog weather or other weather with poor visibility, so as to realize the color effect of the infrared light video truly. The system is easy to realize, is low in cost, is easy to install, and realizes long-distance real-time monitoring.

The invention discloses a radar target recognition method based on noise backgrounds of plural AGC models. The radar target recognition method mainly solves the problems that high recognition rate and small amount of computation can not be achieved at the same time when target recognition is carried out in a noise background in the prior art. The radar target recognition method includes the steps of using radar repeated distance image training samples to calculate variance images of the radar repeated distance image training samples, calculating signal to noise ratio of radar repeated distance image testing samples, correcting the variance images of the repeated distance image training samples according to the signal to noise ratio, and calculating decision values of posteriori probabilities of lower classes of each category according to the corrected variance images and the repeated distance image testing samples to determine target category labels. The radar target recognition method has the advantages of being high in recognition rate and small in amount of computation under the noise backgrounds, and being applicable to radar target recognition.

The invention relates to a pedestrian re-identification method based on human skeleton key point segmentation and column convolution. The method comprises the following steps: (1) preparing a pedestrian data set; (2) sending the training set to a network A, obtaining upper body picture information and lower body picture information, wherein the upper body picture information refers to a high-dimensional matrix X containing pedestrian upper body information, and the lower body picture information refers to a high-dimensional matrix Y containing pedestrian lower body information; (3) sending thehigh-dimensional matrix X into a network B; (4) sending the high-dimensional matrix Y into a network C; and (5) verifying by using the test set, and performing result detection. According to the method, all parts of the human body can be distinguished more accurately based on region segmentation of the human skeleton key points, human body part alignment can be conducted more effectively, and therobustness of the model can be effectively improved. According to the invention, information can be transmitted between pixel rows in the image, and the target recognition rate is effectively improved.

An embodiment of the invention discloses a method and device for adjusting illumination balance. The method comprises the follow steps: carrying out region division on an object region image based on statistical characteristics of a color histogram, determining a classifier corresponding to each region according to correspondence relation between the statistical characteristics and classifiers, and obtaining a color histogram of the object region image according to correspondence relation between the classifiers and histograms; carrying out comparison to judge whether a distribution equilibrium value of the color histogram of the object region image is within the range of a distribution equilibrium value of a color histogram of a sample image; and if not, adjusting light balance parameters of the object region image to enable the distribution equilibrium value of the color histogram of the object region image to be within the range of the distribution equilibrium value of the color histogram of the sample image. The method and device for adjusting illumination balance can solve the technical problem of illumination imbalance in the prior art, thereby improving object recognition rate.

The invention provides a stable imaging method applicable to two-dimensional wide-area scanning. The core of the method is geographical stabilization and electron image movement compensation matching, forward movement of a photoelectric system is compensated in a geographical stabilization manner, the movement of the azimuth is compensated in a geographical stabilization and electron image movement combined manner, and clear imaging of the photoelectric system in the two-dimensional wide-area scanning process is finally realized. The method is mainly realized through algorithms, the structural form is simple, the imaging quality of the photoelectric system can be effectively improved, and the reliability of the photoelectric system can be improved.

Provided is an object identification method based on high-resolution one-dimensional image information fusion. The purpose of the invention is to solve how to perform information fusion on the observations of different radars in a network in order to increase resolution. The method comprises a step 1: detecting two aircrafts by using two kinds of radars with different waveforms generated by BSS so as to obtain echo data of the two different radars; a step 2 of performing pulse compression on the echo data to obtain two groups of high-resolution one-dimensional images of an object; a step 3 of performing data fusion on the two groups of high-resolution one-dimensional images by using a weighted average method to obtain a new group of one-dimensional images, wherein a fusion weight is randomly selected between 0 to 1; a step 4 of subjecting the two groups of high-resolution one-dimensional images to a set threshold to acquire the number of sampling points; a step 5 of multiplying the number of sampling points of the new high-resolution one-dimensional images by a preset distance resolution (actual distance between the two sampling points) to obtain the length of the object; and a step 6 of performing object identification by using Bayesian classification.

The invention belongs to the field of the image data analysis technology and specifically discloses a SAR image analysis method based on self-adaptive fuzzy C mean-value clustering fuzzification. The method comprises the steps: firstly, processing and outputting the fuzzification result of a fuzzy decision table based on the present FCM (Fuzzy C Mean-value); secondly, gradually increasing the values of various category numbers Cm based on the value of the incompatible degree and outputting the corresponding fuzzification result of each condition attribute when the number Cm is increased by 1; when the incompatible degree is more than the preset threshold value, gradually adjusting the category number Cm of the corresponding condition attribute from the condition attribute with minimum importance degree; and finally, outputting the fuzzification pre-treatment result of the present SAR image based on the corresponding fuzzification result of the present category number Cm of each condition attribute. The SAR image analysis method is used for target identification of SAR images, the output fuzzification pre-treatment result is capable of remarkably improving the correct identification rate of the target.