109results about How to "Accurate border" patented technology

The invention provides an image semantic division method based on a depth full convolution network and a condition random field. The image semantic division method comprises the following steps: establishing a depth full convolution semantic division network model; carrying out structured prediction based on a pixel label of a full connection condition random field, and carrying out model training, parameter learning and image semantic division. According to the image semantic division method provided by the invention, expansion convolution and a spatial pyramid pooling module are introduced into the depth full convolution network, and a label predication pattern output by the depth full convolution network is further revised by utilizing the condition random field; the expansion convolution is used for enlarging a receptive field and ensures that the resolution ratio of a feature pattern is not changed; the spatial pyramid pooling module is used for extracting contextual features of different scale regions from a convolution local feature pattern, and a mutual relation between different objects and connection between the objects and features of regions with different scales are provided for the label predication; the full connection condition random field is used for further optimizing the pixel label according to feature similarity of pixel strength and positions, so that a semantic division pattern with a high resolution ratio, an accurate boundary and good space continuity is generated.

The invention discloses an image semantic segmentation method based on deep learning. The method comprises four parts of data set processing, deep semantic segmentation network construction, deep semantic segmentation network training and parameter learning, and semantic segmentation on a test image. The RGB image and the gray level image of the input image are used as the input of the network model, the edge information of the gray level image is fully utilized, and the richness degree of input characteristics is effectively increased; a convolutional neural network and a bidirectional threshold recursion unit are combined, and more context dependency relationships and global feature information are captured on the basis of learning image local features; coordinate information is added tothe feature map through the first coordinate channel module and the second coordinate channel module, the coordinate features of the model are enriched, the generalization ability of the model is improved, and a semantic segmentation result with high resolution and accurate boundary is generated.

A water body information automatic extraction method for a multi-spectral image of a remote sensing satellite comprises the steps of (1) implementing normalized difference water index (NDWI) extraction on apparent reflectivity data of the multi-spectral image of the domestic remote sensing satellite, and forming an NDWI image; (2) implementing maximum between-cluster variance algorithm based global threshold segmentation initial water body information extraction on the NDWI image; (3) implementing local buffer area expansion and threshold segmentation on the initial water body information, and obtaining a stable water body unit result set; and (4) according to elevation data, obtaining slope and shadow information, removing a shadow from the water body unit result set, and finally finishing the water body information extraction. According to the method, by adopting a computer automatic interpretation method and utilizing a thought of from global segmentation to local iterative segmentation, the NDWI is subjected to water body information extraction in a process manner; and the method is high in extraction precision and easy in engineering.

The invention discloses a content-based video detection method and a content-based video detection device. A video program to be detected is subjected to audio decoding and audio detection by adopting a partial image decoding mode to determine a candidate region where a target segment appears first and then the only candidate region is subjected to image decoding and image detection to determine the accurate boundary of the target segment. By adopting the mode, not only the processing speed is improved greatly, but also the accurate boundary of the target segment can be given.

The invention discloses a full-automatic three-dimensional scene construction method based on a single image. The full-automatic three-dimensional scene construction method based on the single image comprises the following steps: on the basis of a machine learning method, utilizing a training image set to train to obtain a classifier which can roughly classify and label an input image; dividing the input image into three subareas, i.e. a vertical area, ground and sky, by utilizing the classifier to obtain a rough classifying and labeling result of image areas; on the basis of a "trustable" area in the rough classifying and labeling result, correcting the rough classifying and labeling result by utilizing a GrabCut image segmentation algorithm, and obtaining a precise boundary between image geometry areas; and on the basis of obtaining the precise classifying and labeling results of the image geometry areas and the precise boundary between the image geometry areas, carrying out modeling with a realistic three-dimensional scene by utilizing a computer graphics method. The method disclosed by the invention realizes the three-dimensional scene construction based on the single image.

The invention discloses a building facade three-dimensional reconstruction method based on knapsack type three-dimensional laser point cloud data, and relates to the technical field of geographic information. The method comprises the following steps: S1, acquiring building point cloud data; S2, automatically extracting building facade point cloud data; S3, automatically segmenting the single buildings; S4, acquiring a geometric position boundary of the building facade; and S5, building facade three-dimensional reconstruction. A point cloud filtering algorithm based on voxel projection densityis adopted to effectively filter non-building targets such as the ground and vegetation while a relatively complete building target is reserved, and then automatic segmentation of a single building isrealized by utilizing an image global search and profile analysis method. An RANSAC algorithm is used to carry out facade automatic segmentation and redundant facade elimination on the single building point cloud to obtain a building facade geometric position boundary; and the two-dimensional boundary line is used to constrain the original point cloud data and an RANSAC algorithm is combined tocarry out facade three-dimensional boundary straight line fitting so as to obtain a building facade geometric frame model.

The invention relates to a Kinect deep image remediation method based on colorful image segmentation. The Kinect deep image remediation method comprises the following steps: by taking a corresponding colorful image segmentation result as reference information for remedying a deep image, reconstructing an invalid area of the deep image, and furthermore by taking a colorful image as guidance information, filtering the reconstructed deep image, thereby achieving an effect of removing noise. Therefore, on the basis that reference information is provided through the simultaneously taken colorful images, the colorful images are segmented to obtain cavity reconstruction areas of the deep image, and a Kinect deep image with relatively good quality can be obtained.

The invention discloses a blue light filter film for visible light communication, which is structurally characterized in that a wide-pass band interference cutoff filter film, a high reflection film and an antireflection film are arranged on a substrate sequentially. The blue light filter film can greatly reduce noises formed due to a reason that visible illuminating light and background radiation enter a detector. Moreover, since the edge of the blue light filter film for blue light signal penetration and visible illuminating light cutoff is steep and accurate, when the visible illuminating light and near-infrared interference are filtered to an utmost extent, the loss of the blue light signals is minimum, the efficiency and the signal-to-noise ratio of the visible light communication are remarkably improved, and the blue light filter film is very beneficial to the development and the application of fields such as the visible light communication.

The invention discloses a small sand body boundary identification method and a small sand body space quantitative description method. The small sand body boundary identification method comprises the following steps: summarizing the development rules and scale of a sand body according to the geological situation of a region; summarizing the seismic response characteristics of a reservoir and carrying out seismic profile identification and interpretation through seismic forward modeling; processing earthquake in a frequency division way, extracting seismic attributes of different frequency bands, and adopting known well data analysis to preferably select seismic attributes better corresponding to the reservoir and the sand body; and finely describing the sand body by comprehensively adopting the geological rules, sand body seismic response characteristics and frequency-division seismic attribute analysis. The small sand body space quantitative description method comprises the steps of drawing the boundary of a sand body and determining a favorable area of a reservoir by the above method, and quantitatively describing the spatial shape of the sand body through well logging constrained inversion. The boundary of a described small sand body is accurate and quick to describe, and the accuracy of reservoir predication can be improved.

The invention discloses an improved watershed SAR image segmentation method based on complex wavelet texture region merging, which combines a complex wavelet feature extraction method and the watershed algorithm improved by the Gaussian low-pass filter to be used for restraining over segmentation existing in watershed-segmented SAR images. The method comprises the following implementation steps of: (1) carrying out the improved watershed transformation to an original image so as to obtain initial segmentation; (2) extracting complex wavelet energy feature of each pixel in the initial segmented image; (3) unifying the pixel feature in each irregular region block; and (4) sorting and merging different region blocks by utilizing a 'K-mean value' clustering method to reduce the region blocks and obtain final results. The image segmentation effect of the method is verified to be basically consistent with the standard by a group of simulation experiments.

The invention discloses a road scene segmentation method based on a residual network and expanded convolution. The method comprises: a convolutional neural network being constructed in a training stage, and a hidden layer of the convolutional neural network being composed of ten Respondial blocks which are arranged in sequence; inputting each original road scene image in the training set into a convolutional neural network for training to obtain 12 semantic segmentation prediction images corresponding to each original road scene image; calculating a loss function value between a set formed by12 semantic segmentation prediction images corresponding to each original road scene image and a set formed by 12 independent thermal coding images processed by a corresponding real semantic segmentation image to obtain an optimal weight vector of the convolutional neural network classification training model. In the test stage, prediction is carried out by utilizing the optimal weight vector of the convolutional neural network classification training model, and a predicted semantic segmentation image corresponding to the road scene image to be subjected to semantic segmentation is obtained. The method has the advantages of low calculation complexity, high segmentation efficiency, high segmentation precision and good robustness.

The invention relates to a transient electromagnetic quick three-dimensional human-machine interaction inversion method. The transient electromagnetic quick three-dimensional human-machine interaction inversion method includes that gathering geological information, physical geography information and physical logging information; using the information to analyze comprehensively to build a geo-electric model of a surrounding rock; using the model to forwardly calculate the TEM response of the surrounding rock, and calculating a pure anomaly field (three-component) according to the actual TEM response; using an electric current loop to be equivalent to the vortex of a target geological body generated by the excitation of a primary field at a time, enabling the field generated by the equivalent current loop to gradually get close to the three-component field value observed at the earth surface, and inverting to obtain the position, size, inclination angle, inclination direction and the other geometrical parameters of the current loop so as to obtain the information of the target geological body; judging whether the information is reasonable according to the known information, if so, obtaining the information of the target geological body to realize three-dimensional TEM human-machine interaction inversion, otherwise, comprehensively analyzing again to construct the geo-electric model of the surrounding rock, entering the next circulation, and stopping till the information is reasonable.

The invention discloses a Retinex low-illumination color image enhancement method based on variation constraint, and relates to the technical field of color image processing, the method comprises thefollowing steps: converting an original color image from an RGB space to an HSV space, and extracting an image of an intensity channel V as a grayscale image; constructing a conductivity weight matrixby using the extracted grayscale image; constructing an objective function by using an energy minimization principle, inputting the grayscale image and the conductance weight matrix into the objective function, and solving the objective function by using an alternating direction multiplier method to obtain an illumination image; Separating a reflection component from the original color image according to the illumination image to obtain a reflection image; The gray level image is reconstructed by using the illumination image and the reflection image; According to the method, the boundary of the image is well reserved, so that the estimated illumination image is more accurate, the reflection image contains more internal details and texture features of the image, the contrast ratio of the enhanced image is improved, and the details of the high-brightness position of the image are reserved.

The invention relates to a guiding device for tooth coronal cavity preparation and a making method thereof. The guiding device comprises a base plate (1) and a guiding structure (2) connected to the base plate, the center of the base plate (1) is provided with a coronal cavity-shaped face boundary positioning hole (5), the shape of the positioning hole (5) is the projection of the shape of a dental pulp chamber (6) and the shape of a root canal orifice (7) to a tooth crown along a guiding direction, the guiding direction (8) of the guiding structure (2) is the optimum direction of entrance into a medullary canal and the guiding direction is determined by an operator or is automatically calculated by a software algorithm. The making method of the guiding device comprises acquiring tooth three-dimensional shape data, then carrying out computer-aided design, and finally making the guiding device through a numerical control cutting or three-dimensional printing device. Through the making method, the personalized guiding device for entrance into a medullary canal is prepared and is used for accurate tooth coronal cavity preparation so that the occurrence of the mistake is reduced.

The invention provides a method and a device for material segmentation of material accumulation images. The method mainly comprises carrying out filtering processing of original material accumulation images to obtain smoothed images, and after carrying out morphological operation on the smoothed images, carrying out subtraction on the smoothed images and the original images to obtain gradient images; carrying out threshold processing on a gray value of each pixel in the smoothed images to obtain binary images; filling internal noise of the binary images through a scan-line filling method, and disconnecting connected accumulation targets in the binary images to obtain distance smoothed images; carrying out geodesic opening operation on the distance smoothed images to extract central target areas of various accumulation objects, outwardly growing in parallel from the central target areas of various accumulation objects in the gradient images to the edges of the accumulation objects to obtain boundary information of various accumulation objects. According to embodiments provided in the invention, materials in an accumulation scene can be accurately and effectively segmented.

The invention relates to a gravity data and gravity gradient data combined regularized inversion method. The method comprises the steps that 1, gravity data is acquired, inversion calculation is performed on the gravity data, and an inversion result is obtained; 2, the absolute value of the inversion result m obtained in the step 1 is acquired and normalized, a zero value in the result is set as aminimal value, and the processed result is used as a weighted matrix; and 3, gravity gradient data is acquired, the gravity gradient data and the weighted matrix are simultaneously applied to inversion calculation, and an inversion result of combined inversion of the gravity gradient data and the weighted matrix is obtained. Through the method, geological information contained in the gravity datais effectively utilized, the gravity data is utilized to construct the weighted matrix, the weighted matrix is applied to the gravity gradient inversion calculation, the resolution of the inversion result is increased, and the bottom border of the inversion result is clearer and more accurate.

The invention provides a carbonate rock microfacies type identification method and a sedimentary facies description method thereof; the microfacies type identification method comprises the following steps: a microfacies classifying step: carrying out microfacies classification for a single well coring segment so as to obtain a plurality of microfacies types; a microfacies identification step: obtaining features of the microfacies types, and identifying each microfacies type according to the features of the microfacies types, wherein the microfacies type features comprise at least one of the following features: imaging logging image feature and earthquake reflection composition feature. In a microfacies types identifying process, the method combines the imaging logging image feature with the earthquake reflection composition feature, thus more accurately identifying microfacies of the carbonate rocks.

The invention provides an image automatic segmentation method and device based on deep learning edge detection. The method comprises the steps of drawing a certain number of training samples accordingto a high-resolution remote sensing image; training an HED edge detection model through the training sample; performing edge detection on the remote sensing image to be segmented by using the trainedHED edge detection model to generate an edge probability graph; performing post-processing on the edge probability graph to generate a vector polygon; and simplifying the vector polygon to obtain animage segmentation result of the remote sensing image to be segmented. According to the image segmentation method provided by the invention, parameters do not need to be adjusted, the segmentation accuracy is higher, automatic operation can be realized, and the image segmentation efficiency is high.

The invention discloses a street view image segmentation method integrating a network and a dual-channel attention mechanism. The method comprises a training stage and a test stage. In the training stage, an image segmentation convolutional neural network model based on a high-resolution fusion network and a dual-channel attention mechanism is constructed, and the model is trained, wherein the model comprises an input layer, a hidden layer and an output layer, and the hidden layer comprises four expansion convolution blocks stacked in a dense sampling mode, a high-resolution fusion network anda dual-channel attention mechanism; in the test stage, the streetscape image to be segmented is predicted to obtain a predicted semantic segmentation image, i.e., image semantic segmentation based ona high-resolution fusion network and a dual-channel attention mechanism is realized. The method is high in segmentation precision and better in robustness.

The present invention provides a foreland basin alluvial fan fine description and prediction method. The method comprises a step of establishing a foreland basin alluvial fan geological model, a step of establishing the stratigraphic framework of an alluvial fan, a step of carrying out seismic facies and attribute inversion to delineate the plane distribution range of an alluvial fan, a step of carrying out three-dimensional electrical inversion to divide the phase band boundary of the alluvial fan of each period, carry out the fine identification of internal lithology and lithofacies, and determine the spatial distribution characteristics of the lithology and lithofacies, and a step of applying OpenDtect software to analyze the alluvial fan vertical evolution and establishing the deposition model of a study area alluvial fan. The method provided by the invention is mainly based on the sedimentology theory and the geological model foundation established by alluvial fan exposure investigation, logging, seismic, electrical method and other materials are fully utilized, the means of logging geological interpretation, seismic sequence analysis, seismic facies interpretation, attribute analysis and electrical method interpretation are integrated, and each phase band of the alluvial fan and the internal lithology and lithofacies spatial distribution characteristic are effectively identified.

The invention discloses an aurora oval position determining method based on deep learning and mainly solves the problem of inaccuracy existing in an existing aurora oval position determining method. The aurora oval position determining method comprises the following steps: (1) selecting complete aurora oval images from ultraviolet auroral images to form an aurora oval image set; (2) performing image segmentation on original images by utilizing shape information and maximum similarity region merging criterions; (3) performing coordinate transformation on images obtained by segmentation by utilizing geomagnetic coordinate information of the images; (4) obtaining boundary point coordinates in the aurora oval equatorial direction and the polar direction according to transformed images; (5) constructing a geomagnetic physical parameter database at the corresponding moment of an aurora oval by utilizing the shooting time of the images in the aurora oval image set; (6) inputting the aurora oval boundary point coordinates as well as geomagnetic physical parameters into a deep learning network to determine the aurora oval position. According to the aurora oval position determining method, the aurora oval position determination accuracy is improved. The aurora oval position determining method can be used for researching the effect of the geomagnetic physical parameters on aurora activities.

The invention discloses a resume layout analysis algorithm fusing visual and textual characteristics. The analysis of a resume layout comprises the following steps: step 1, obtaining text lines and coordinates thereof from a pdf reading program or an ocr engine; step 2, encoding the text of the ith row by using a neural network to obtain a text embedding vector text _ emb (i); step 3, extracting an image of a corresponding row to obtain an image embedding vector img _ emb (i); step 4, extracting word size and word length features, and normalizing the word size and word length features to obtain a feature vector; step 5, aggregating the vectors obtained in the steps 2, 3 and 4 to obtain a line embedding line _ emb (i); and step 6, performing sequence labeling on the row vector sequence [line _ emb (i)] by using a neural network. Semantic division is carried out on the resume by combining visual features and text semantic features of the resume, and independent paragraph units are identified.

The invention discloses a mechanical arm autonomous moving grabbing method based on visual-tactile fusion under a complex illumination condition. The method mainly comprises approaching control over a target position and feedback control over environment information. According to the method, under the complex illumination condition, weighted fusion is conducted on visible light and depth images of a pre-selected area, recognition and localization of a target object are completed based on a deep neural network, and a mobile mechanical arm is driven to continuously approach the target object; in addition, the pose of the mechanical arm is adjusted according to contact force information of a sensor module, the external environment and the target object; and meanwhile, visual information and tactile information of the target object are fused, and the optimal grabbing pose and the appropriate grabbing force of the target object are selected. By means of the method, the object localization precision and the grabbing accuracy are improved, collision damage and instability of the mechanical arm are effectively prevented, and harmful deformation of the grabbed object is reduced.

The invention discloses a method, a device and equipment for generating a timing sequence action nomination and a storage medium. The method comprises the following steps: acquiring a plurality of video frames in a video; calling a time sequence action nomination generation model to predict the plurality of video frames to obtain a time sequence boundary confidence map and an action integrity probability map corresponding to the video, and fusing the time sequence boundary confidence map and the action integrity probability map to obtain a fused feature map; and outputting the timing sequenceaction nomination of the video according to the fusion feature map. The two time sequence boundary confidence maps and the action integrity probability map based on the dense boundary are used for predicting the boundary of time sequence action nomination based on global nomination level information (L * L dimension), and is not similar to the method that the BMN predicts the boundary of the timing sequence action nomination only based on the local information, so that the accurate boundary can be predicted in the generation task of the timing sequence action nomination.