44results about How to "Improve robustness" patented technology

The invention provides a real-time cloud service bottleneck detection method based on a kernel density estimation and a fuzzy inference system. The real-time cloud service bottleneck detection method is characterized in that S1, real-time cloud environment can be started, and cloud service can be provided; S2, a real-time cloud service state parameter acquisition assembly can be started; S3, cloud service state parameters in a certain time period in the cluster normal operation environment can be acquired, the modeling can be carried out by adopting the kernel density estimation model, and the fuzzy inference system can be started at the same time; S4, the recently acquired state parameters can be input in the fuzzy inference system, and the bottleneck indexes can be calculated; S5, the threshold value can be set, and the historical law of every cloud service bottleneck index can be observed, and when the cloud service bottleneck indexes exceed the threshold value in a certain time period, the early warning mechanism can be started. The real-time cloud service bottleneck detection method is advantageous in that the operation state of every service in the cloud cluster can be acquired in a real-time manner, and can be stored in the database; the bottleneck detection of every real-time cloud service can be realized; when the cloud service bottleneck indexes of the real-time cloud service exceed the threshold value for a long time, the early warning mechanism can be started.

The invention discloses an image retrieval method. The image retrieval method includes the steps that firstly, the number of interior points of any two images in an image data base is calculated, and the correlation degree value between the two images is calculated according to a formula; then, the correlation degree value between a query target and any image in the image data base is calculated to obtain an image directly related to the query target and an image indirectly related to the query target; a weighted adjacent matrix A is constructed for attenuation, a compensation item for compensating for an over-small indirect correlation degree value caused by excessive attenuation is added to calculate the indirect correlation degree value between the query target vq and the image vi, and the indirect correlation degree value is adjusted so that it can be achieved that the larger the correlation degree value of an image is, the larger the indirect correlation degree value of the image and the query target vq is; the direct correlation degrees and indirect correlation degrees between the query target and all images in the image data base respectively are sequenced to obtain an image, containing the query target, in the image data base.

The invention discloses a road target depth estimation method based on scene depth mapping. The scene target depth estimation is achieved by obtaining two factors of the scene depth mapping relation and target positioning. On the basis of scene edge detection and straight line detection, aiming at solving the problem that positioning of disappearance points and disappearance line is uncertain, an expectation-maximization algorithm is used to accurately estimate a center disappearance point and a center visual field disappearance line. A camera model is utilized to complete construction of the scene depth mapping relation. Aiming at solving the problem that scene target positioning is not accurate, based on scene division, an Adaboost classifier is adopted to train and identify class marks of divided areas. A ground contact line of a vertical target is analyzed and a depth mapping relation table is queried, so that scene target depth estimation is obtained.

The invention relates to an automatic parking method based on fusion of vision and ultrasonic perception. The method comprises the following steps that visual parking space detection is carried out, ultrasonic radar parking space detection is carried out, two parking space detection results are fused, and parking space tracking and parking space display are carried out; a target parking space is determined, and a parking speed and a parking track are planned in combination with a current scene map; according to the visual detection and the ultrasonic radar detection, obstacles at a driving area around a vehicle and an edge of the driving area are detected, and the obstacles are subjected to motion estimation; the probability of collision of the vehicle itself and the obstacles is calculated, and if the collision probability is greater than a preset threshold value, an obstacle avoidance function is triggered; and parking track tracking is carried out through a low-speed controller, gear control, vehicle speed control, steering wheel steering control and obstacle avoidance control along the parking track are achieved, and automatic parking is completed. According to the method, thedetection effect of the parking space and the obstacles around the vehicle is improved, the robustness is improved, a motion obstacle track estimation module is added, and the obstacle avoidance reliability is improved.

The invention discloses a napping behavior detection method based on fast Fourier transform, which comprises the following steps: carrying out face detection for collected images, collecting coordinates of a central point of a rectangular area with a detected human face, and carrying out spectrum analysis for data through fast Fourier transform when the collected coordinates of the central point reaches required values, so as to find out a spectrum difference between napping and normal movement of a person. During the spectrum analysis, the head behaviors of a person are divided into three states, static, slightly moving and regulating moving. The head behaviors in a time period are analyzed after the spectrum analysis is carried out for a plurality of times, in order to find out a dominating state in the time period. When the regular movement in the vertical direction dominates, the person can be judged to be napping at present. The invention adopts the pattern recognition technologyand the statistical judgment method to detect the napping behavior, and has the advantages of no contact, high robustness and good instantaneity.

The invention provides a method and device for quantizing local features of a picture into visual vocabularies, wherein the method comprises the steps of S1, deciding to-be-selected vocabularies from a first layer of a visual vocabulary tree; S2, computing the confidence degree of a route in which each to-be-selected vocabulary in the current layer exists respectively by utilizing the distance between the local feature and each to-be-selected vocabulary in the current layer and the confidence degree of a route in which a father node of each to-be-selected vocabulary in the current layer exists; and S3, selecting the to-be-selected vocabularies in the route of which the confidence degree is more than or equal to a predetermined confidence degree threshold value in the current layer and judging whether the current layer is the last layer or not, wherein if the current layer is the last layer, the vocabularies selected in the current layer are determined as the visual vocabularies of the local features; and if the current layer is not the last layer, the vocabularies selected from the current layer enter the next layer, child nodes of the selected vocabularies are determined as the to-be-selected vocabulary in the next layer and the operation returns to the step S2. With the adoption of the method and device for quantizing the local features of the picture into the visual vocabularies, based on the improvement of the robustness of quantization errors, the computation overhead during the quantization process is decreased.

The invention provides a dangerous article image detection method, a dangerous article image detection device, equipment and a medium under the articulated naturality web, and relates to the technicalfield of articulated naturality webs. The method comprises the following steps: acquiring a dangerous article image sample in an articulated naturality web server, and preprocessing the dangerous article image sample to obtain a preprocessed image sample; transmitting the preprocessed image sample to a preset neural network model for training to obtain a feature neural network model, wherein thepreset neural network model is stored in the articulated naturality web server; and sending the parameters of the feature neural network model to a client, so that the client detects a to-be-detectedobject through the received feature neural network model. According to the invention, the training of the feature neural network model and the detection of the to-be-detected image are separated, thememory and CPU pressure of the client are liberated. The robustness of the feature neural network model is improved. The labor cost is reduced, and the detection speed and accuracy are improved.

The invention provides a multi-fracture surface fleet regulation and control method and a multi-fracture surface fleet regulation and control device, relating to the technical field of power transmission. According to the method and the device, the high-robustness and high-convergence intelligent control is realized. The multi-fracture surface fleet regulation and control method comprises the following steps: changing a reference trend topology according to a repair plane; transferring fracture surface power, load power and partial generator power to a plan type data designated position; and redistributing unbalanced power between power generation and load of a power grid of each region.

The invention belongs to the technical field of optical communication systems and high-speed optical signal processing, and particularly relates to a transmitting end iterative dispersion compensation method for an intensity modulation direct detection system. According to the method, in the amplitude limiting process of the receiving end, the amplitude of even symbols in two times of sampling points in the iteration process is reserved and only the amplitude of odd points is replaced, so that the optimizable space of the signal is better utilized, and the iteration convergence speed and the final dispersion compensation effect are improved.

The invention discloses a keep-alive and self-healing method and a keep-alive and self-healing device for an advanced telecom computing architecture control surface. The method comprises the following steps that: a frame management module acquires a control surface heartbeat keep-alive intelligent platform management interface (IPMI) command regularly initiated to a switch board by the common single board and/or a control surface heartbeat keep-alive IPMI command regularly initiated to the common single board by the switch board, wherein the common single board and the switch board are positioned in a frame where the frame management module is positioned and are in the successful software electrification state; the switch board sends a control surface heartbeat keep-alive IPMI command response to the frame management module through the common single board, or the common single board sends the control surface heartbeat keep-alive IPMI command response to the frame management module through the switch board; and if the control surface heartbeat keep-alive IPMI command response is not received in continuous appointed time, the frame management module determines that a link from the switch board to the common single board or from the common single board to the switch board fails, and resets the failed single board. The method improves the accuracy of detection positioning and the robustness of the system.

The invention relates to the technical field of image processing, in particular to a mixed domain based image global contrast and local detail level adjustment method. The method includes: establishing a Gaussian pyramid of single-channel images, processing the Gaussian pyramid layer by layer from top to bottom, and performing mixed calculation in a spatial domain and a frequency domain by the aid of images on each layer and processing results of images on the upper layer so that final adjusted images are obtained. The mixed domain based image global contrast and local detail level adjustment method is capable of arbitrarily adjusting global contrast and local detail level of images to be adjusted and inhibiting and avoiding undesirable effects such as halos, jags, ringing, edge overturning and the like on multiple scales. Besides, the mixed domain based image global contrast and local detail level adjustment method is rapid in calculation process and high in robustness.

The invention belongs to the field of image processing, particularly relates to a group brain map individualization method, system and equipment based on an individual brain connection map, and aims to solve the problem that an existing brain map function division mode is difficult to reflect differences among individuals and realize cross-individual analysis. The method comprises the following steps: acquiring a T1 weighted nuclear magnetic resonance image of a target and a dispersion nuclear magnetic resonance image of a to-be-detected target, extracting surface grid triangular nodes and edges of white matter and grey matter surface structures, taking a vertex of a grey matter and white matter interface as a node of the individual brain connection map, and taking a spatial neighborhood as an edge of the individual brain connection map, obtaining a fiber bundle fingerprint through the corrected diffusion MRI image, taking the fiber bundle fingerprint as a node attribute of the individual brain connection graph, and finally obtaining a partitioned individual level brain map through a brain region classification graph convolutional network. According to the method, the individualized brain map drawing process is established, and the robustness of individual difference drawing is improved by utilizing the fiber bundle connection fingerprints and the map convolutional network.