53results about How to "Reduce segmentation" patented technology

A vehicle, system and method of navigating a vehicle. The vehicle and system include a digital camera for capturing a target image of a target domain of the vehicle, and a processor. The processor is configured to: determine a target segmentation loss for training the neural network to perform semantic segmentation of a target image in a target domain, determine a value of a pseudo-label of the target image by reducing the target segmentation loss while providing aa supervision of the training over the target domain, perform semantic segmentation on the target image using the trained neural network to segment the target image and classify an object in the target image, and navigate the vehicle based on the classified object in the target image.

First, initial frame is captured from video frequency input. Next, based on configuration file generated by user in advance in monitored area, local rectangle region, where there is an object in video frequency needed to pick up, is determined. Then, in the determined rectangle region, pretreatment, watershed segmentation, color analysis and regional clustering process are carried out. Finally, video object picked up is saved. The invention is as a pioneer in video monitoring area based on content in order to solve issues: large calculated amount of picked video object, not accuracy. The invention is applicable to picking up pointed monitored object in a scene. Experiment indicates the algorithm in high real time, basic consistence between clustering processed pixel block and monitored object. Thus, the invention provides basis for further tracking video object.

The invention discloses Grabcut-based sandstone particle improved segmentation algorithm research, and provides a sandstone particle size segmentation algorithm based on an image. Firstly, a method for adaptively and automatically generating a target candidate box is provided, a foreground and background mixed model can be quickly segmented from a background image, the time for manually drawing the box is greatly shortened, and the requirement for real-time performance in an engineering field is met; secondly, the mass center of the super-pixel image of the SLIC is clustered by adopting a K-Means clustering method based on the SLIC, so that the number of samples participating in clustering can be greatly reduced; the Gibbs energy function can determine the weights of the two segmented edges; in order to reduce the number of iterations of foreground and background segmentation of an unknown label of Grabcut in a candidate box, an information entropy function with a certain weight is added to an energy function of Gibbs, segmentation information of a segmented image is supervised, and the real-time performance of the system is further optimized. The invention can be applied to real-time on-line analysis of the grain size of the sandstone in a factory.

The invention relates to a conveyor belt ore rock particle image segmentation method. The method comprises the following steps: 101, preprocessing a to-be-segmented conveyor belt ore rock particle image, and inputting the to-be-segmented conveyor belt ore rock particle image into a pre-trained first convolutional neural network model to obtain a first contour map; 102, inputting the first contourmap into a pre-trained second convolutional neural network model to obtain a second contour map; 103, performing binarization processing on the second contour map by using a preset threshold, and performing morphological closed operation processing on the binarized map to obtain a third contour map; 104, obtaining the contour, the minimum bounding rectangle of the contour and the area of the contour in the third contour map, screening by using the minimum bounding rectangle of the contour and the area of the contour, and drawing the screened contour as a segmentation map. According to the method, deep learning is utilized to realize conveyor belt ore rock image contour detection and ore rock size distribution automatic statistics, the requirement for image definition is reduced, image segmentation is accurate, and application is convenient.

The invention discloses a visual perception enlightening high-resolution remote-sensing image segmentation method, and the method enables an object boundary in an image to be divided into an intensity boundary corresponding to a spectrum homogeneous area, and a texture boundary corresponding to a spectrum change area, and respectively extracts two types of visual information which serve as the main basis of segmentation. The method comprises the steps: filtering noise and texture information in the image through a nonlinear filtering method, and obtaining the intensity gradient through a gradient operator; analyzing the texture features of the image through employing a Gabor filter, enabling the filtering output of a plurality of channels to be merged and processed through the gradient operator, and obtaining the texture gradient of the image; carrying out the fusion of the intensity gradient and the texture gradient, carrying out the conversion of the gradient image after fusion through watershed conversion, and achieving the segmentation of the image. Compared with the prior art, the method improves the boundary accuracy of image segmentation, reduces the over-segmentation and under-segmentation phenomena, and can be effectively used in the field of high-resolution remote-sensing image information processing.

Embodiments of the invention provide an image segmentation method and apparatus. The method comprises the steps of mapping all pixel points of an image into a CIELAB color space; uniformly segmenting the image into first regions with the same number as first super-pixels, and performing partition on pixel points in each first region and regions near the first region to obtain the first super-pixels corresponding to the first regions; mapping all the first super-pixels into an RGB color space, averaging pixel points in the first super-pixels to obtain corresponding second super-pixels, and constructing an undirected graph by taking all the second super-pixels as vertexes and taking a similar distance between every two adjacent second super-pixels as an edge; and forming a segmented region by the two adjacent second super-pixels with the similar distance meeting a preset condition in the undirected graph, and judging whether the adjacent segmented regions are combined or not according to a weight value of the edge in the adjacent segmented regions. According to the embodiments of the image segmentation method and apparatus, the pixel points in the image can be changed smoothly, the over-segmentation phenomena are reduced, and the image segmentation effect is improved.

The invention provides a steel-concrete mixing type anti-shearing connecting piece which is applicable to connecting a wave-shaped steel web plate and a concrete top plate. The top of the wave-shaped steel web plate is provided with a hole; the upper part of the hole is respectively welded with one or more combined steel bars along the vertical direction of the two sides of the wave-shaped steel web plate; at the position of a combined face of the wave-shaped steel web plate and the concrete top plate, one horizontal frange plate which is in convex-concave complementation and tight adhesion with the wave-shaped steel web plate is respectively arranged at each one of the two sides of the wave-shaped steel web plate in the vertical direction; the frange plate is provided with a plurality of vertical welding nails; a connection structure above the frange plate stretches into the reinforced concrete top plate connected with the frange plate; and a transverse steel bar in the concrete top plate penetrates through the hole on the wave-shaped steel web plate, and the concrete of the wave-shaped steel web plate is poured. The steel-concrete mixing type anti-shearing connecting piece has the characteristics of an imbedded anti-shearing connecting piece and a frange type anti-shearing connecting piece; steel products are saved, the machining is convenient, the construction is convenient and the anti-shearing strength is high; the strength, the rigidity and the durability of a connecting part are better; and the integrity of the combined part is good, and the social and economic benefits are good.

The invention discloses a level set image segmentation method based on a self-adaptive parameter, for mainly solving the problem that various parameters in a curved surface evolution equation of a conventional level set image segmentation algorithm need to be preset in advance. The realization steps comprise: (1), inputting an image to be segmented; (2), setting a time step length, and giving iteration frequency; (3), carrying out Gauss filtering processing on the input image; (4), adding the self-adaptive parameter to a level set evolution equation to replace an original parameter constant; (5), starting level set iteration operation; (6), determining whether the iteration frequency reaches an upper limit or is convergent; and (7), determining whether a termination condition is achieved, if not, returning to step (5), and otherwise, outputting a segmentation result graph. The method provided by the invention has the advantages of short time, accurate segmentation result and high stability, thereby being capable of application in such technical fields as image enhancement, mode identification and object tracking.

A method of making display having a display area capable of displaying a plurality of characters in a background, the characters including a plurality of segments, includes the steps of: providing a substrate; forming a first patterned conductor layer on the substrate having electrically separate areas defining character regions, the first patterned conductor being substantially coextensive with the display area; depositing a layer of light modulating material over the first patterned conductor layer; forming a second patterned conductor layer over the layer of light modulating material and having electrically separate areas defining the segments of the characters and the background, the second patterned conductor being substantially coextensive with the display area; depositing a dielectric layer over the second patterned conductor layer, the dielectric layer defining holes over each segment and the background area of the second patterned conductor layer; and forming a third patterned conductor layer located over the dielectric layer defining a plurality of conductors connected to the areas defining the character segments in the second patterned conductor and the background area through the holes in the dielectric layer; at least one of the conductors being connected to a segment in more than one character, whereby the display may be addressed in a matrix fashion by electrically addressing the first and second plurality of contacts, and may be driven in a positive or negative mode.

The invention relates to a liver tumor segmentation method and system based on deep learning. The method comprises the following steps: preprocessing a collected data set; a network model being built according to the preprocessed data, wherein the network model comprises a plurality of lower convolution layers and a plurality of upper convolution layers, the lower convolution layers are connected to the upper convolution layers through jumping, and feature maps obtained by the lower convolution layers are connected with the upper convolution layers through an attention module on jumping connection; inputting the preprocessed data into the network model for training to obtain an optimal network model; and segmenting a to-be-processed CT image by using the optimal network model to obtain a liver tumor region. According to the invention, wrong segmentation can be reduced, and high precision can be obtained.

The invention discloses a main girder square steel splicing structure on an escalator or moving pavement. The structure comprises two pieces of main girder square steel, wherein the ends of the two pieces of main girder square steel are fixed together by permanent butt welding with backing; the backing is made of reinforced square steel; the two ends of the reinforced square steel are extended into cavities at the ends of the two pieces of main girder square steel respectively; the outer surfaces at the ends of the two pieces of main girder square steel are provided with a plurality of punched holes in which a melted soldering flux is filled; the main girder square steel and the reinforced square steel are plugged into each other for welding; the two pieces of main girder square steel are connected by butt welding through the reinforced square steel; and the main girder square steel and the reinforced square steel are plugged into each other for welding so as to form one piece of square steel. The length of the piece of square steel can meet the requirements of manufacturing a metal framework, and the piece of the square steel also meets the requirements of manufacturing the metal framework in mechanical property, so that unnecessary sectioning of the metal framework is reduced, and the manufacturing cost of the metal framework is reduced.

The invention discloses a near-eye display device, which includes a display panel and a vibrating element, wherein the display panel includes a plurality of pixels; the vibrating element is disposed on the side of the display panel, and the vibrating element vibrates to drive the display panel to vibrate, causing multiple pixels to reciprocate and displace, wherein the vibration amplitude of the vibration element is less than 100 [mu]m.