361results about How to "Achieve segmentation" 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 belongs to the technical field of image processing, in order to realize automatic extraction and segmentation of retinal blood vessels, improve the anti-interference ability to factors such as blood vessel shadow and tissue deformation, and make the average accuracy rate of blood vessel segmentation result higher. The invention relates to a retinal blood vessel image segmentation method based on a multi-scale feature convolutional neural network. Firstly, retinal images are pre-processed appropriately, including adaptive histogram equalization and gamma brightness adjustment. Atthe same time, aiming at the problem of less retinal image data, data amplification is carried out, the experiment image is clipped and divided into blocks, Secondly, through construction of a multi-scale retinal vascular segmentation network, the spatial pyramidal cavity pooling is introduced into the convolutional neural network of the encoder-decoder structure, and the parameters of the model are optimized independently through many iterations to realize the automatic segmentation process of the pixel-level retinal blood vessels and obtain the retinal blood vessel segmentation map. The invention is mainly applied to the design and manufacture of medical devices.

The invention relates to a method of quickly segmenting a moving target in a non-restrictive scene based on a full convolution network, which belongs to the technical field of video object segmentation. The method comprises steps: firstly, framing is carried out on the video, and a result after framing is used for making a Ground Truth set S for a sample image; a full convolution neural network trained through a PASCAL VOC standard library is adopted to predict a target in each frame of the video, a deep feature estimator for an image foreground target is acquired, target maximum intra-class likelihood mapping information in all frames is obtained hereby, and initial prediction on the foreground and the background in the video frames is realized; and then, through a Markov random field, deep feature estimators for the foreground and the background are refined, and thus, segmentation on the video foreground moving target in the non-restrictive scene video can be realized. The information of the moving target can be effectively acquired, high-efficiency and accurate segmentation on the moving target can be realized, and the analysis precision of the video foreground-background information is improved.

The invention provides a method for segmenting a white blood cell image, which is characterized by comprising the following steps: firstly carrying out binaryzation on the green component picture of a color white blood cell image to obtain an initial interested region based on a significance attention mechanism of human vision; carrying out cancellation and mergence on a region by labeling the initial interested region to obtain an adaptive significance window of each cell; and finally realizing segmentation of nucleuses and cytoplasm in each adaptive significance window by a boundary-extending method. The segmentation method is used to realize the fast segmentation of the white blood cells, particularly the accurate segmentation for high-capacity pictures which are the non-standard dyed and contain overlapped cells.

The invention belongs to the technical field of laser radars, and discloses a depth map-based three-dimensional laser radar point cloud target segmentation method, which comprises the following stepsof: converting three-dimensional point cloud data acquired by a three-dimensional laser radar into a two-dimensional depth map; calculating an angle value formed by two adjacent points in each columnin the depth map, and traversing to obtain an angle matrix corresponding to the depth map; traversing the depth map through a breadth-first search algorithm, if the angle difference between two pointson adjacent positions of the depth map is smaller than a specified threshold value, marking the depth map as the same type, thereby finding out the part, belonging to the ground, in the depth map, and removing ground point cloud data according to the mapping relation between the point cloud and the depth map; and carrying out target segmentation on the non-ground point cloud based on an improvedDBSCAN algorithm, and judging whether the point cloud is a core point or not according to an adaptive parameter eps while considering a spatial Euclidean distance and an angle distance. According to the method, the segmentation efficiency on the depth map is improved, the real-time requirement is met, and the problems of under-segmentation and over-segmentation are effectively solved.

The invention discloses a three-dimensional carotid artery ultrasonic image blood vessel wall segmentation method based on deep learning. The method comprises the following steps: (1) obtaining a three-dimensional ultrasonic image; (2) obtaining a two-dimensional ultrasonic image of a carotid artery cross section, and performing manual marking; (3) dynamically and finely adjusting the convolutional neural network model by utilizing the manually marked image block; (4) fitting vascular adventitia-tunica media boundary initial contour; (5) using the dynamically adjusted convolutional neural network model to carry out vascular adventitia-tunica media boundary contour segmentation; (6) obtaining a vascular cavity ROI region; (7)using U-Net network to divide the vascular cavity, and extractingthe vascular cavity-tunica media boundary contour through morphological processing. According to the method, the contours of vascular adventitia-tunica media boundary MAB and vascular cavity-tunica media boundary LIB can be accurately segmented out; the workload of doctors is greatly reduced, and the vascular wall volume (VWV), the vascular wall thickness (VWT) and the vascular wall thickness change (VWT-Change) can be calculated based on the segmentation result of the method.

The invention discloses a weed image segmentation method under the rape field environment. Multiple weed/rape RGB image samples are randomly acquired in the rape field; a visual attention model is established, the color characteristics, the brightness characteristics and the direction componential characteristics are extracted, each characteristic graph is acquired and each characteristic channelsaliency map is generated so that a total saliency map is acquired and the area of interest is acquired; the shape characteristics and the texture characteristics of the area of interest are extractedto perform support vector machine classification training so as to acquire the rape area; and the miscellaneous image samples and all the rape area images are fused so as to acquire the final inter-strain weed area distribution information. The area of interest is acquired through fusion of the improved visual attention model with combination of the region growth algorithm, and the whole algorithm process does not require grayscale transformation or threshold segmentation so that the processing link and the computing amount can be reduced; and the segmentation efficiency is further enhanced by extracting the characteristic parameters of the area of interest and support vector machine classification model judgment so that weed image segmentation under the background of the rape field can be realized.

The invention relates to an image mosaic processor on the basis of an FPGA (Field Programmable Gata Array), which is characterized by comprising a set of DVI (Digital Video Interactive) digital decoding circuit, a set of FPGA digital video processing circuit and a set of output interface circuit, wherein an external digital signal is connected with the DVI digital decoding circuit; a digital video signal with the minimum transformation differential signal form is subjected to decoding processing; an obtained video pixel information flow is directly fed into the FPGA digital video processing circuit; after being processed by an image processing algorithm, the signal is subjected to encoding processing to obtain the digital video signal of the minimum differential signal; and the digital video signal of the minimum differential signal is output to a mosaic wall display through the output interface circuit. The invention also provides an image mosaic method on the basis of the system. The image mosaic processor and the image mosaic method have the advantages that the image mosaic processor and the image mosaic method are not limited by the bandwidth of a PCI (Peripheral Component Interconnect) bus or the aspects of acquiring and processing capabilities.

The invention discloses a method for recognizing a license plate from a picture. The method comprises the following steps of adopting license plate positioning to determine a rectangular picture including the license plate, adopting character segmentation to respectively extract characters of a license plate number in the license plate and recognizing the characters in the license plate number based on a neural network model. According to the method, the license plate positioning is realized through combination of color positioning and Sobel positioning; through carrying out color judgment on the picture of the license plate, carrying out binaryzation and extracting a character outline and a circumscribed rectangle, the character segmentation is realized; and the characters of the license plate are recognized through a neural network algorithm. The experimental verification shows that the method of the invention has higher license plate recognition rate and can adapt to license plate recognition in different special circumstances.

The invention provides a landslide terrain detection method based on a deep neural network, and the method comprises the steps: collecting a landslide remote sensing image and digital elevation modeldata, obtaining specific range coordinates of a landslide, and generating a landslide marking map; carrying out data preprocessing on the landslide remote sensing image, the digital elevation model and the landslide labeling graph; and constructing a geologic feature extraction model taking DeepLab V3 + as a framework, and extracting rich geologic features from the remote sensing terrain data set.According to the method, landslide remote sensing image features and digital elevation model features are fused, landslide domain knowledge is combined, geological feature parameters are loaded, a terrain segmentation model is realized on the basis of the DeepLab V3 + architecture, landslide terrain segmentation of pixel-level granularity is completed, and the purpose of landslide terrain detection is achieved.

The embodiment of the invention discloses an image processing method and electronic equipment; the method comprises the following steps: converting an original pathology image into a result pathologyimage that satisfies target image parameters; taking an image area containing cells in the result pathology image as a target area; segmenting cells in the target area so as to obtain a segmentation result. The method can automatically detect and segment cells in the pathology image, thus reducing manpower and material consumption, preventing diagnosis errors caused by artificial participation, and improving the cell detection and segmentation precision.

The invention discloses an automatic auditing system for illegal bus lane occupation based on deep learning, and the method comprises the following steps: employing a bayonet camera to carry out the photographing and evidence collection of a target vehicle illegally occupying a bus lane, the evidence collection information comprising n frames (n>=1) of evidence graphs and the license plate numberinformation of the target vehicle; utilizing a yolo-V2 vehicle detection model to detect all vehicles in n frames of images, adopting a caffe-ssd model to detect a license plate area of the all the detected vehicles, and identifying a license plate number by using an lstm + ctc model; comparing the license plate numbers with the edited distance of a given license plate, and applying GoogLenet Inception-V2 network based vehicle re-identification algorithm to respectively position the position of a target vehicle in the n frames of images; segmenting an image bus lane area by using a deeplab-V2segmentation algorithm, and judging whether a target vehicle occupies a bus lane by calculating the ratio of the intersection of the target vehicle detection frame and the bus lane area to the targetvehicle detection frame; and identifying the type of the target vehicle by applying a vehicle classification network based on ResNet18, and finally judging whether the target vehicle illegally occupies the bus lane according to the type information of the target vehicle and the condition of occupying the bus lane. According to the invention, the police is saved, the illegal auditing efficiency andaccuracy are improved, and the auditing fairness is ensured.

The present invention discloses a deep learning-based ultrasound contrast tumor automatic identification and detection method. The method comprises the following steps of obtaining the characteristic video data of a target area; pre-processing the characteristic video data; partitioning the pre-processed characteristic video data to obtain a plurality of video blocks; inputting the video blocks in a 3D convolutional neural network to process the data and obtain a characteristic layer; inputting the characteristic layer in a full-connection layer to classify the characteristics, and finally determining whether the classification result is a tumor. By utilizing the 3D convolutional neural network, the method provided by the present invention can effectively extract the time-space effective blood flow characteristics in and between the ultrasound contrast video data frames, detects the tumor tissues based on the difference of the tumor tissues and the normal tissues and by combining the time-space characteristics of the video data and using a 3D CNN algorithm to segment the tumors, and has a better segmentation accuracy than a conventional machine learning algorithm.

The invention discloses a probability-partition-merging-based three-dimensional model segmentation method. The method comprises: (1), over-segmentation is carried out to form lots of areas; (2), segmented areas are aggregated; (3), a cascading classified is trained; and (4), according to the cascading classified, a new model is segmented and adjacent areas are merged to obtain a segmentation result. On the basis of lots of areas, the current initial segmented boundary includes a final ideal segmentation boundary; with a logistic regression model, a cascading secondary classifier is trained and is used for determining whether the adjacent areas need to be merged into one area. According to the invention, on the basis of the machine learning theory, a triangular mesh is segmented by means of cascading area cohesion and a proper segmentation result is selected based on demands. Moreover, effectiveness of the segmentation method is verified; the integration feature can be realized well; and for most of models with different types, the effect is good.

The invention discloses a sequential level set-based method for automatically segmenting images, which comprises the steps: (1) inputting an image I to be segmented and automatically initializing a level set function in an image domain by using a morphological method; (2) evolving the level set function by using an iterative approach in a corresponding subdomain to obtain the level set function and a detected image border, and using the inside domain of the edge as a next subdomain; (3) converting an image edge detected in the subdomain into a binary image and storing the binary image; (4) judging whether the area of the next subdomain is zero, and executing a step (5) if the area of the next subdomain is zero, or returning to the step (1) if the area of the next subdomain is not zero; and (5) performing a 'combination' operation of the stored binary image, combining the image edges of all subdomains, and completing the image segmentation. The method has the advantages of detecting more image edges and automatically completing image segmentation, and can be used for segmenting natural images and medical images.

The invention discloses an MR (magnetic resonance) image three-dimensional interactive segmenting method for random walks and graph cuts based active contour model, comprising the steps of S1, selecting a seed point to capture local three-dimensional MR image data including pituitary adenomas in order to acquire a segmented initial boundary surface; S2, establishing a hybrid active contour model based on the initial boundary surface to obtain a model energy function; S3, discretizing the model energy function; S4, using each pixel of the captured local three-dimensional MR image as an image node, and using 6 neighborhoods of each pixel to construct an image; giving an initial value to each of the pixels inside and outside the initial boundary surface; giving a corresponding weight to each of sides connecting between the nodes, between the nodes and source points and between the nodes and meeting points according to the discretized energy function; S5, performing image segmentation computing based on the constructed mage to obtain segmentation results, and extracting boundary surface form the segmentation results to obtain a segmented contour; S6, replacing the initial boundary surface of S1 with the current segmented contour, and repeating the steps S2 to S5 until the segmentation results converge to obtain a final segmented contour. By using the method of the invention, it is possible to provide three-dimensional segmentation for MR images of pituitary adenomas, and provide more accurate segmentation for the images of pituitary adenomas.

The invention relates to an eye fundus image vessel segmentation method based on a local enhancement active contour module. The method comprises: according to a feature vector of a Hessian matrix, vessel enhancement in an eye fundus image is carried out; curvature distribution statistics is carried out on the enhanced image to obtain an eyeball edge and get rid of the eyeball edge; and with a local enhancement active contour module, enhanced image segmentation is carried out by combining global energy information, thereby extracting an eye fundus vessel. According to the invention, on the basis of gray level distribution characteristics of the vessel in the medical eye fundus image, a local energy function is established and a global energy functional item is combined, so that the curve evolution process becomes stable; the speed is accelerated; and the vessel in the eye fundus image can be extracted precisely and effectively.

The invention discloses an active contour synthetic aperture radar (SAR) image segmentation method based on Fisher distribution, which mainly aims to overcome the disadvantage of existing Gamma distribution to the SAR image segmentation technology, the method comprises the following specific implementation steps of (1) making use of Fisher distribution to fit the intensity statistical characteristics of an image area, and establishing an energy functional function based on the Fisher distribution according to a regional competition model; (2) introducing a level set function, and combining a length constraint item and a level set rule item to re-express the energy functional function obtained in step 1; (3) minimizing the energy functional function obtained in step 2 by adopting Euler-Lagrange calculus of variations, estimating Fisher distributed parameters by making use of logarithmic moment estimation, and then performing numerical solution to a partial differential equation, thereby obtaining the segmentation result of an SAR image. According to the method provided by the invention, the level set method evolutionary split curve and Fisher distributed parameter estimation are combined to minimize the energy functional function, thereby realizing segmenting SAR images.

The invention discloses a V-belt base glue and rope core bonding divider, which comprises a rack, wherein a base glue coil support for installing a base glue coil, a base glue sending assembly used for clamping a base glue on the base glue coil and conveying the base glue forwards, a base glue cutting assembly, a driving roller, and a driven roller connected with a front and back rope core tensioning device are sequentially arranged on the rack from the front to the back, and the driving roller is arranged on a spindle connected with a power device; the base glue cutting assembly is provided with a cutter used for cutting off the base glue, and the cutter is connected with a telescopic device; a dividing assembly is arranged at the lower side between the driving roller and the driven roller, and comprises a lifting device and a blade bearing connected with the lifting device, and a blade shaft and a plurality of round blades arranged on the blade shaft are arranged on the blade bearing; a counting sensor for detecting a base glue head is arranged on the rack, and the power device, the base glue sending assembly, the telescopic device, the lifting device, the front and back rope core tensioning device and the counting sensor are connected with a control system through signals. The V-belt base glue and rope core bonding divider can achieve an automation object.

The invention discloses a three-dimensional grid object segmentation method based on a point cloud processing network. The three-dimensional grid object segmentation method comprises the following steps: preparing a three-dimensional grid data set and a pre-trained point cloud processing network PointNet; calculating a central point coordinate of a triangular patch in the three-dimensional grid data as point cloud data corresponding to the three-dimensional grid data; randomly selecting N points in the point cloud, and inputting the N points into a pre-trained Point Net to obtain a segmentation result; giving probability distribution to unselected points in the point cloud by using a KNN algorithm to obtain all segmented point cloud data; and mapping the point cloud segmentation result back to the three-dimensional grid data, and correcting the result by using a conditional random field to finally obtain a segmentation result of the three-dimensional grid. On the basis of deep learning, complete data can be analyzed, global features are extracted from the data, and data-driven segmentation is achieved instead of a traditional three-dimensional feature extraction operator based on artificial design.

The invention discloses a vegetation coverage detection method and device thereof in a grazing sheep feeding path of a grassland. The vegetation coverage detection method comprises the following stepsof 1 collecting and pre-processing a vegetation image; 2 establishing a vegetation characteristic database; 3 segmenting the vegetation image and calculating the coverage degree of various vegetation. The detection device comprises an image acquisition mechanism, an image preprocessing module, a feature library training module and a vegetation segmentation module. The method and the system of theinvention can correct the focal length of the camera and the coefficient when the image pixel is calculated by comparing the vegetation image with the calibration image of the artificial method, thereby improving the accuracy of the vegetation image. By adopting the principal component analysis method to reduce the dimension of the image, the redundancy of the characteristic parameters can be reduced. The method and the system of the invention conveniently divide a vegetation image into a plurality of single vegetation images by establishing a BP neural network training model, thereby calculating the coverage of each vegetation and improving the detection efficiency of the vegetation coverage in the grazing sheep feeding path.

The invention discloses a multimodal image assisted diagnosis system and a construction method thereof. The multimodal image assisted diagnosis system comprises a hospital server, an embedded development platform and a neural computing stick, wherein the neural computing stick is connected with the embedded development platform; the embedded development platform is connected with the hospital server; the hospital server, the embedded development platform and the neural computing stick are all in the same local area network environment; the hospital server stores patient information and corresponding multimodal image data; and a pre-trained deep learning model is deployed in the neural computing stick, thereby being capable of reducing the burdens on patients and doctors at the same time, and improving the working efficiency.

The invention relates to a complex makeup migration method based on facial analysis and illumination migration. Facial analysis is combined with an illumination migration algorithm and a makeup migration algorithm. The method comprises the following steps: firstly, carrying out whitening and buffing preprocessing on an input face image by using a color balance and bilateral filtering algorithm; secondly, migrating makeup of the reference face image to the input face image by utilizing a face analysis and facial makeup migration algorithm, and finally migrating illumination of the reference face image to the input face image by utilizing an illumination migration algorithm to complete migration of complex facial makeup. The method is very easy to realize through software, and can be widelyapplied and popularized to computer vision, visual design, digital entertainment, film production and the like.