155results about How to "Achieve robustness" patented technology

The invention discloses a retinal vessel segmentation method based on combination of deep learning and a traditional method and relates to the fields of computer vision and mode recognition. According to the method, two grayscale images are both used as training samples of a network, corresponding data amplification, including elastic deformation, smooth filtering, etc., is done against the problem of less retinal image data, and wide applicability of the method is improved. According to the method, an FCN-HNED retinal vessel segmentation deep network is constructed, an autonomous learning process is realized to a great extent through the network, convolutional features of a whole image can be shared, feature redundancy can be reduced, the category of multiple pixels can be recovered from the abstract features, a CLAHE graph and a gauss matched filtering graph of the retinal vessel image are input into the network, an obtained vessel segmentation graph is subjected to weighted average, and therefore a better and more intact retinal vessel segmentation probability graph is obtained. Through the processing mode, the robustness and accuracy of vessel segmentation are improved to a great extent.

The invention provides a pedestrian re-identification method based on multi-scale feature cutting and fusion, particularly provides pedestrian re-identification network training based on multi-scale depth feature cutting and fusion and a pedestrian re-identification method based on the network, and performs pedestrian re-identification through multi-scale global descriptor extraction and local descriptor extraction. The extraction of the global descriptor is to carry out average pooling and feature fusion on feature maps of different layers of the deep network, and the extraction of the localdescriptor is to horizontally divide the feature map of the deepest layer of the deep network into a plurality of blocks and respectively extract the local descriptors corresponding to the feature maps. In the training process, a minimum smooth cross entropy cost function and a difficult sample sampling triple cost function are used as the target training network parameters. By adopting the technical scheme of the invention, the problem of feature mismatching caused by factors such as pedestrian posture change and camera color cast in pedestrian re-identification can be solved, and the influence caused by background can be eliminated, so that the robustness and precision of pedestrian re-identification are improved.

The invention provides an all working condition standard emission control method for a thermal power unit denitration system. The method comprises the steps that a main ring PID controller, a sub-ringPID controller and a feedforward controller are arranged in a denitration control system; the main ring PID controller calculates an ammonia injection control signal according to an outlet NOx concentration set value and an outlet NOx concentration feedback signal; the feedforward controller calculates an ammonia injection feedforward control signal according to the purge correction NOx concentration, the NOx estimated concentration, the inlet NOx concentration and the flue gas flow; and the sub-ring PID controller controls the opening of an ammonia injection valve according to the ammonia injection control signal, the ammonia injection feedforward control signal and an ammonia flow feedback signal. According to the invention, a gain scheduling model and an hour mean model are added intoa main ring to realize automatic control of the variable working condition of the denitration system; an inlet concentration purge correction model is introduced into a sub-ring to realize dynamic ammonia injection correction; the ammonia injection is accurately controlled; and a technical support is provided for whole process robustness and all working condition standard emission.

A method for transmission is provided to include demultiplexing information to be transmitted into a plurality of stream blocks, encoding each of the stream blocks according to a corresponding coding scheme to generate a plurality of encoded streams, interleaving the plurality of encoded streams in a bit-level to generate a plurality of bit-level interleaved streams, modulating each of the bit-level interleaved streams according to a corresponding modulation scheme to generate a plurality of modulated symbol streams, interleaving the plurality of modulated symbol streams in a symbol-level to generate a plurality of symbol-level interleaved streams, precoding the plurality of symbol-level interleaved streams according to a precoding scheme to generate a plurality of precoded streams, and transmitting the plurality of precoded streams via a plurality of antennas.

The invention relates to the field of computer vision and particularly relates to an improved MSER image matching algorithm. A speeded up robust feature (SURF) and a maximally stable extremal region feature (MSER) are combined to carry out image feature extraction and matching so as to generate a feature vector, and then the Euclidean distance is used to carry out the coarse matching of an image so as to preliminarily correct the space geometric distortion of the image. Then the scale invariance of an H-L feature is applied, and a feature point comprising a large amount of image structure information can be detected. According to the algorithm, the complementarity of feature extraction of two parties in the multiple transformation conditions of the image can be fully utilized, and the robustness of matching between images in a complex environment in a time condition acceptable range is achieved.

The invention discloses a fabric defect detection method based on multi-feature matrix low-rank decomposition. The method comprises the steps of image blocking, multi-channel feature matrix extraction, united low-rank decomposition and saliency map generation and partitioning, wherein a fabric image is divided into image blocks with the same size, a second-order gradient direction map of each image block is calculated, a retina P-type ganglion cell coding mode is adopted to extract image features, and a feature matrix is generated; an effective low-rank decomposition model is constructed according to the feature matrix, optimal solving is performed through a direction alternating multiplier method, and a low-rank matrix and a sparse matrix are generated; and a threshold segmentation algorithm is adopted to partition a saliency map generated by the sparse matrix, and defect positions are found. According to the method, the complexity of fabric texture features and the diversity of defect types are comprehensively considered, second-order features capable of effectively representing the fabric texture features are extracted, the untied low-rank decomposition model is adopted to effectively realize quick separation of defects and a background, and the method has high detection precision.

The invention discloses a servo motor control method integrating sliding mode control and fractional order neural network control. The servo motor control method comprises the following steps of A, establishing a numerical model of a servo motor, and describing the numerical model of the servo motor, so as to obtain the output, speed error and speed error derivative of the servo motor; B, according to the speed error and the speed error derivative of the servo motor, designing a fractional order sliding mode controller, so as to obtain a sliding mode control rule for inhibiting the buffet of the system; C, adopting a neutral network approximating algorithm to approximate the obtained sliding mode control rule, so as to obtain the approximated sliding mode control rule; D, adopting a self-adaptive control method to adjust the approximated sliding mode control rule on line, so as to obtain a final control rule of the servo motor; according to the final control rule, controlling the servo motor. The servo motor control method has the advantages that the sliding mode control theory and the fractional order neural network adaptive control theory are integrated, the robustness is high, and the tracking property is good; the servo motor control method can be widely applied to the field of industrial control.

The invention discloses a fault hologram system based on standardization processing technology and a method thereof. The fault hologram system based on the standardization processing technology is used to widely, comprehensively, and synthetically display fault transient processes in a station from a perspective of fault increment reflection, facilitates timely and accurate fault condition understanding, scientifically analyzes fault reasons, rapidly judges occurring positions, natures and severity degrees of faults, and optimizes the power grid fault treatment process. The fault hologram system is developed by using data treatment techniques of data automation fusion and the like, is based on a multipoint region fault information synthesis analyzing thought after the faults, unifies and standardizes treatment methods through wave shapes, sampling point time scales and sampling frequency, achieves whole process dynamic display of data in different data sources on the same timer axis, uses booting logic based on combination of sampling characteristic value and switching value gallery attributes, automatically locks a fault sampling gallery, synthesizes the faults onto a fault analysis platform for normalization treatment, and improves fault treatment efficiency.

The invention provides a mechanical arm grabbing method and a system for achieving the method. The method comprises the following steps that S1, point cloud information of the surface (the surface capable of being seen in the view field of a camera) of an object to be grabbed is collected; S2, data of the point cloud information is further processed, and feasible grabbing points meeting constraintconditions are extracted through a grabbing programming algorithm; S3, the grabbing points are adopted as inverse kinematics input of a mechanical arm, and a motion control instruction is sent to themechanical arm and a two-finger parallel gripper; and S4, the mechanical arm executes the motion control instruction, and moves to the specified position, and according to the motion timing relationship in the motion control instruction, the two-finger parallel gripper is opened and clamped to complete the grabbing task. The mechanical arm grabbing method and the system for achieving the method have the beneficial effects that under the condition that the shape of the object is uncertain, a single visual sensor can be used for achieving the mechanical arm grabbing function with robustness.

The invention discloses a steady beam synthesizer weight solving method comprising the steps: firstly performing an AD sampling to an array receiving signal x(t), estimating a covariance matrix R; secondly performing a direction of arrival estimation through the covariance matrix, and obtaining a desired signal and a direction of arrival of M interface signals; thirdly obtaining M+1 sub weight vectors w0,..., wM; fourthly obtaining a total weight vector of the array according to a formula of w(t)=w0-sigma wm. The steady beam synthesizer has advantages that the array error is not needed to be complexly corrected by increasing constraint conditions, thereby at the same time of having less operation, realizing the robustness of the beam synthesizer.

The invention belongs to the field of lithium batteries and discloses a state of charge estimation algorithm for a lithium battery monomer. A second-order RC equivalent circuit model is used to ensurethe basic accuracy of battery modeling, the prior knowledge of the battery is effectively used, and besides, the high efficiency of partial least square calculation is used, a data-driven model of the battery is quickly established in an online mode, the problem of difficult online updating of parameters of the traditional second-order RC equivalent circuit model can be effectively solved, and the accuracy of battery model modeling under complex conditions is improved. The final state of charge of the battery integrates the estimation results of two battery models, and the limitation of the accuracy of using a single model under various complex conditions is avoided. Essentially, the estimation accuracy of the battery state of charge under complex working conditions such as an electric vehicle can be improved.

The invention relates to an image-based steel bar end face automatic recognition counting algorithm. First of all, the end face region which needs to be processed in the image is extracted. the imageis preprocessed, which includes image scaling, Gaussian filtering; using cloud model to classify pixel color, obtaining preliminary segmentation of image according to the classification result, extracting parameters after closed operation to the segmented large connected region, including area ratio, center of gravity, aggregation degree, linear weighting to obtain reference value and selection, and obtaining end face region to be processed. Then the end face area is counted by fixed branch separation. Some images are preprocessed, including grayscale, histogram equalization and adaptive threshold binarization, and the binary image of the face region is obtained. Estimated radius of a single face is obtained by particle size measurement method, template is constructed according to formula,template matching is carried out, and the center position of the face is obtained. Finally, the robustness of the algorithm is improved by restriction conditions. The average correct rate of the invention can reach 97%.

The invention discloses a position tracking optimization control method based on a flexible actuator of an exoskeleton robot. The method comprises the steps: building a dynamic mathematic model; to bespecific, building a dynamic mathematic model of a flexible actuator of an exoskeleton robot through employing the Newton's law of motion under the condition that the impact from a friction force isneglected; defining a rolling optimization performance index; optimizing the control performance by selecting a proper optimization index; outputting position prediction based on disturbance compensation; designing a disturbance observer; to be specific, respectively designing the corresponding disturbance observers for matching and non-matching interferences according to the measured position information; and designing a composite controller; to be specific, designing a corresponding composite control scheme by utilizing the observed disturbance information and combining generalized prediction and starting from feedforward and feedback control respectively. The method is based on a position tracking algorithm combining interference active compensation and generalized model prediction control, and has the characteristics of accurate tracking of a given reference trajectory, strong anti-interference capability, high stability, easy realization and the like.

The invention provides an embedded fatigue state detection system and method, and the system comprises an image obtaining module which is used for dynamically capturing the video of a learner througha camera, converting the video into a frame image according to a preset frame interval, and carrying out the normalization processing; a data processing module used for positioning a human face from the frame image, and then positioning and intercepting eye and mouth images, identifying the opening and closing states of the eyes and the mouth through the trained convolutional neural network, calculating the frame number frequency of eye closing and yawning in the convolutional neural network by adopting Parclos and Fom rules, and comparing the frame number frequency with a preset joint judgment threshold to judge whether the target is tired or not; and an output decision module used for controlling a loudspeaker and a display screen to display according to the judgment result of the data processing module. The system and the method provided by the invention are high in robustness, good in practicability and easy for product transformation.

The invention discloses a multi-task depth feature space attitude face recognition method, which comprises the following steps: firstly, carrying out angle measurement on an attitude face image, and extracting image depth space features by utilizing a residual network; then, adding a residual transformation mapping module to realize transformation from the side face depth feature to the front face depth feature, so that a main task of the network is formed; then, adding a module on the basis of an original residual transformation mapping module to realize reconstruction of original side face depth characteristics so as to realize feedback which is a secondary task of the network; and finally, using the cosine similarity to measure the similarity between the to-be-compared face and the depth feature representation of all people in the database, so that face authentication recognition is carried out. According to the method, the robust representation of the front face depth space feature can be obtained according to the side face depth space feature, so that the side face recognition rate is greatly improved, and the method has a very good application prospect in attitude face detection and recognition.

The invention discloses a security password manager based on multi-cloud storage and a use method thereof, the use method comprises the following steps: for accounts and password information of a useron any website, firstly, adopting a (k,n) threshold secret sharing algorithm (combining RAONT-RS algorithm with Shamir algorithm); encrypting information, splitting the encrypted information into n parts, improving the reliability of data by adopting a commitment scheme, and storing the data at n cloud providers; when the password is obtained, the commitment scheme verifying the data firstly, then combining the at least k parts to restore the ciphertext, and finally obtaining the plaintext account and password information through decryption. According to the password manager, the problem of data reliability existing in the multi-cloud storage password manager is solved, the robustness of the password manager under attack is improved on the basis that the data storage security is guaranteed, and meanwhile efficient retrieval can be achieved.

The invention discloses an ice cover radar image ice layer fine segmentation method based on an FCN-ASPP network, and relates to the field of computer vision and mode recognition. According to the invention, the radar amplitude image is used as a training sample of a network, corresponding data amplification is carried out for the problem of less ice layer image data, and the wide applicability ofthe method is expanded. Lee filtering is carried out on the ice cover image. In order to save edge information as much as possible, a threshold judgment process is added to a filtering process. FCN-is constructed, and FCN-is constructed; according to the ASPP ice layer segmentation deep network, the ASPP layer is improved, so that the extraction capability of the network on small-scale characteristics is enhanced. The preliminary classification result is further processed through CRF, and the segmentation result is further refined on the basis of achieving end-to-end pixel level segmentation.In addition, the network greatly realizes the autonomous learning process.

The invention discloses a multi-target tracking system combining a deep learning SSD algorithm with a KCF algorithm, and the system comprises the following steps: setp 1, obtaining an image, and transmitting the image into an SSD deep learning model for target recognition; step 2, the SSD algorithm carries out target identification through GPU acceleration, judges an identification result, filtersimproper targets, and then records position information of each target; step 3, judging whether the acquired image is a first frame of image in the to-be-tracked image sequence or not; if yes, executing the step 4, and if not, executing the step 5; step 4, establishing an object for each target according to the new target position information obtained by the SSD algorithm; an object and a position of target tracking are determined through SSD detection, a KCF algorithm is used for tracking, a target moving track is recorded, and in the tracking process, the SSD algorithm performs optimizationcorrection at the same time to prevent tracking offset, tracking failure, tracking target errors and increase the tracking speed until a target disappears, and the obtained target track is used for service layer analysis.

A parallel multiple robust digit watermarking-method based on multi-copyright certification includes making Walsh code modulation on copyright information (watermark) of different owner and parallel-embedding modulated watermark into carrier image; detecting demodulation of post watermark information based on relevant parallel blind detection when parallel detection is carried out.

The invention provides a cross-media feature learning retrieval method based on semi-supervision. The method comprises the following steps that: S1: establishing a multimedia database; S2: solving theprojection matrixes of different media types; and S3: carrying out cross-media retrieval; and S3: carrying out cross-media retrieval. The S2 comprises the following steps that: 2.1: defining a targetfunction; 2.2: optimizing the target function; and 2.3: projecting the original feature of the multimedia data to a public space. The S3 comprises the following steps that: 3.1: extracting the feature of the media data submitted by a user: according to the media type of the data submitted by the user, using a pre-trained model to extract the feature of the data; 3.2: projecting the feature vectorof the media data into a common space; 3.3: calculating a similarity between the projected feature vector and other vectors in the common space; and 3.4: returning the first k pieces of media data with the highest similarity. By use of the method, calculation complexity is lowered, noise robustness is realized, and retrieval accuracy is improved.