80results about How to "Guaranteed Segmentation Accuracy" patented technology

The invention discloses a target identification and capture positioning method based on deep learning, and belongs to the field of machine vision. The method comprises the following steps that: firstly, utilizing a Kinect camera to collect the depth and the color image of a scene; then, using a Faster R-CNN (Regions with Convolutional Neural Network features) deep learning algorithm to identify ascene target; according to an identified category, selecting a captured target area as the input of a GrabCut image segmentation algorithm; through image segmentation, obtaining the outline of the target so as to obtain the specific position of the target as the input of a cascade neural network for carrying out optimal capture position detection; and finally, obtaining the capture position and the capture gesture of a mechanical arm. Through the method, the instantaneity, the accuracy and the intelligence of target identification and positioning can be improved.

The invention discloses a retina eyeground image segmentation method based on a depth full convolutional neural network. The retina eyeground image segmentation method includes the following steps of:selecting a training set and a test set, extracting retina eyeground images to obtain optic disk positioning area images, and performing blood vessel removal operation on the optic disk positioning area images; constructing the depth full convolutional neural network, taking the optic disk positioning area images as the input of the depth full convolutional neural network, and performing the training of an optic disk segmentation model on the training set based on trained weight parameters as initial values to fine tune model parameters, and performing fine tuning on parameters of an optic cup segmentation model based on trained optic disk segmentation model parameters; and performing optic cup and optic disk segmentation on the test set by utilizing a trained optic cup segmentation model, performing ellipse fitting on final segmentation results, calculating a vertical cup-disk ratio according to optic cup and optic disk segmentation boundaries, and taking a cup-disk ratio result as important basis for a glaucoma auxiliary diagnosis. The retina eyeground image segmentation method achieves optic disk and optic cup automatic segmentation of the retina eyeground images, has high precision and fast speed.

The invention discloses an image processing method and apparatus, and belongs to the field of image recognition. The method comprises the steps of acquiring a target image; calling an image recognition model, wherein the image recognition model comprises a backbone network, and a pooling module and a cavity convolution module connected with the backbone network and parallel to each other, and a fusion module connected with the pooling module and the cavity convolution module; performing feature extraction on the target image through the backbone network in the image recognition model, processing characteristic images output by the backbone network through the pooling module and the cavity convolution module to obtain a first result output by the pooling module and a second result output bythe cavity convolution module, fusing the first result and the second result through the fusion module, and outputting a model recognition result of the target image; and based on the model recognition result, obtaining a semantic segmentation label graph of the target image. By adopting the method and the apparatus, the demands of accuracy and segmentation precision can be met at the same time.

The invention provides a parallel method of target detection and semantic segmentation based on end-to-end depth learning, which obtains a model compose of a target detection neural network Darknet-19, a target segmentation full convolution neural network FCN by training a massive labeled target detection frame and a pixel-level target segmentation image. , and tasks of parallel target detection and target segmentation are successfully realized. That method can effectively extract image features, realizes real-time image processing functions on the basi of ensuring target detection and targetsegmentation, and has a wide application prospect.

The invention discloses a 3D point cloud semantic segmentation method under a bird's-eye view coding view angle. The method is applied to an input 3D point cloud. The method comprises: converting a voxel-based coding mode into a view angle of a bird's-eye view; extracting a feature of each voxel through a simplified Point Net network; converting the feature map into a feature map which can be directly processed by utilizing a 2D convolutional network; and processing the encoded feature map by using a full convolutional network structure composed of residual modules reconstructed through decomposition convolution and hole convolution, so that an end-to-end pixel-level semantic segmentation result is obtained, point cloud network semantic segmentation can be accelerated, and a point cloud segmentation task in a high-precision real-time large scene can be achieved under the condition that hardware is limited. The method can be directly used for tasks of robots, unmanned driving, disordered grabbing and the like, and due to the design of the method on a coding mode and a network structure, the system overhead is lower while high-precision point cloud semantic segmentation is achieved,and the method is more suitable for hardware-limited scenes of robots, unmanned driving and the like.

The invention discloses a retinal fundus image segmentation method based on a deep full convolutional neural network, and the method comprises the steps: selecting a training set and a test set, carrying out the extraction of a retinal fundus image, obtaining an optic disc positioning area image, and carrying out the blood vessel removal operation; Constructing a deep full convolutional neural network, taking the image of the optic disc positioning area as the input of the deep full convolutional neural network, performing optic disc segmentation model training on the training set based on thetrained weight parameter as an initial value to finely adjust model parameters, and performing parameter fine adjustment on the optic disc segmentation model on the basis; Using the trained visual cup segmentation model to perform visual cup and visual disc segmentation on the test set, performing ellipse fitting on a final segmentation result, and calculating a vertical cup-visual disc ratio according to a segmentation boundary of the visual cup and the visual disc. Automatic segmentation of the optic disc and the optic cup of the retinal fundus image is realized, the precision is high, andthe speed is high.

The invention discloses an image semantic segmentation method, electronic equipment and a readable storage medium. Based on an FCN model based on depth feature fusion, the image semantic segmentationmethod replaces the traditional convolution operation by cavity convolution, constructs original images with different resolutions to form image pyramids, hierarchically inputs the FCN model, fuses the output characteristics of the upper layer with the output characteristics of the lower layer, and fuses output features to a bottom layer from top to bottom layer by layer for transposed convolution, and the output features of the bottom layer perform transposition convolution so as to enable the output resolution to be consistent with a bottom-layer input image, thus improving the sensitivity to target positioning, and the segmentation precision is ensured through optimization processing of a full-connection conditional random field subsequently, thereby obtaining a better segmentation effect.

The invention relates to a power plant high-temperature pipeline defect detection and segmentation method based on OTSU and a region growing method, and the method comprises the following steps: obtaining an original infrared image, carrying out the graying, and extracting a pipeline region through the improved two-dimensional OTSU pre-segmentation; finding out a gray scale range corresponding tothe normal temperature of the pipeline through the gray scale distribution histogram of the pipeline area; judging whether the maximum value of the neighborhood gray average value exceeds a normal gray range or not to finish automatic detection and positioning of the multi-defect seed points; and taking the self-adaptive threshold adjusted according to the gray average value and the standard deviation of the grown region as a growth criterion, and taking a gradient amplitude threshold calculated based on a Prewitt operator as an additional condition to realize growth and extraction of the defect region. Compared with the prior art, the system has the advantages of high reliability and accuracy and good real-time performance, and can meet application requirements.

The invention provides a semantic segmentation method, device and equipment and a computer readable storage medium. The method comprises the steps: obtaining a display operation for a three-dimensional model; in response to the display operation, displaying the three-dimensional model on a human-computer interaction interface, wherein the human-computer interaction interface comprises semantic segmentation options; obtaining a selection operation for the semantic segmentation options; and in response to the selection operation, after a two-dimensional segmentation result of a two-dimensional image is obtained, displaying a semantic segmentation result of the three-dimensional model on the human-computer interaction interface, wherein the semantic segmentation result is determined accordingto the two-dimensional segmentation result of the two-dimensional image and the model attribute of the three-dimensional model, and the two-dimensional image and the three-dimensional model belong tothe same scene. Through the semantic segmentation method based on artificial intelligence provided by the invention, the segmentation efficiency of the three-dimensional model can be improved, and the user experience is improved.

The present invention discloses a wavelet scatternet-based SAR image segmentation method, which resolves a technical problem that conventional texture-based SAR image segmentation is inefficient and time-consuming. Implementation steps are that: performing pre-processing of denoising and uniformization on an SAR image; setting a scattering transform path, selecting a wavelet function and a window function, and generating a scattering propagation operator and a scattering operator; performing scattering transform on the SAR image, to obtain a scattering coefficient of an SAR image pixel point; performing K-Means cluster on a scattering texture feature after dimension reduction, to obtain a preliminary segmentation result; searching for affine space in which pixels of different categories are located, to form an affine classifier; and performing sliding window correction on the preliminary segmentation result by using the affine classifier, to implement SAR image segmentation. According to the present invention, it is unnecessary to perform a partitioning operation on the SAR image, and the scattering texture feature that can reduce a difference between same texture and increase a difference between different texture is extracted, and precise and fast segmentation can be performed on the SAR image. The method is used for fast segmentation on a single SAR image.

The invention discloses a lung anatomy position positioning algorithm based on a deep learning technology, which can accurately and quickly divide lung CT, and can simply, quickly and accurately realize automatic segmentation of lung lobes based on lung CT images, thereby realizing the anatomy position positioning of lung lesions. Compared with a traditional segmentation method, the method has theoutstanding advantages that (1) the process is simple, and the end-to-end segmentation mode does not need to pay attention to other processes; (2) the multi-stage and multi-output network architecture controls the network in different stages, so that the segmentation effect is better, and the segmentation precision can be ensured to the maximum extent through a semantic-based segmentation mode; and (3) the generalization ability is strong, and the data in the training process is enhanced, so that the model can learn different and diverse data, namely, the generalization ability of the segmentation model is ensured, meanwhile, the risk of over-fitting is also avoided to a certain extent, and the geometric deformation and illumination influence of CT (computed tomography) are insensitive when lung lobe division is performed on different CT.

The invention discloses a parallel underwater image segmentation method and device. The method includes the following steps: S1, dividing a group of images into different subimage groups according to similarities of image attributes; S2, allocating the different subimage groups to different computational resources respectively; S3, classifying each pixel to the corresponding clustering category via the computational resources in parallel according to the membership degree of gray, as a member of the clustering center, of each pixel of the corresponding subimage group, wherein the clustering center is a gray value. By the method and device, underground image segmentation efficiency can be improved.

The invention relates to a method and system for extracting information of an urban water body. The method comprises: S1, carrying out data preprocessing on an urban remote sensing image captured by a satellite to obtain a preprocessed image; S2, carrying out image segmentation on the preprocessed image according to a preset segmentation parameter to obtain a mean value image having feature groups including a spectral feature, a topological feature, a shape feature and an length-width ratio feature; S3, carrying out a spectral feature analysis on the mean value image to obtain a typical ground object spectral curve graph including spectral curves of all typical ground objects containing a building, a water body and a shadow, at all wave bands; and S4, according to the feature group, extracting water body information in the typical ground object spectral curve graph. According to the invention, the method and system have the following beneficial effects: the water body is effectively extracted based on the spectral features of the ground objects, the shape features, topological relations, and length-width ratio information of the ground objects; and the extraction accuracy is ensured.

The invention discloses a lung cancer-oriented intelligent target region and organ-at-risk segmentation method. Dual-path segmentation is performed by adopting a common segmentation network framework of a lung cancer-oriented target region and an organ at risk; the common segmentation network framework comprises a teacher model and a student model, and the two models both adopt nnU-Net as trunks; and for the organ at risk and the target region, pre-defined auxiliary condition information is respectively given to be embedded in an up-sampling process and is connected to a corresponding decoder in a decoding process, and predicted masks are fused to obtain a final segmentation result. According to the method, the target region and the organ at risk can be accurately and efficiently segmented.

The invention discloses an image partitioning method and device. The method comprises the steps that a first frame three-dimensional point cloud image collected under a horizontal viewing angle is acquired; and a first ground equation and feature transformation are acquired; according to the first ground equation and point cloud data in the first frame three-dimensional point cloud image, a secondground equation of the first frame three-dimensional point cloud image is determined; according to the feature transformation and the second ground equation, a first ceiling equation of the first frame three-dimensional point cloud image is determined; according to the second ground equation and the first ceiling equation, first to-be-partitioned data is determined, wherein the first to-be-partitioned data is point cloud data, except point cloud data in the second ground equation and point cloud in the first ceiling equation, in the first frame three-dimensional point cloud image; and image partitioning is performed according to the first to-be-partitioned data.

The invention discloses an SAR image semantic segmentation method based on a fast ridgelet deconvolution structure learning model, which mainly solves the problems of inaccurate segmentation image and long model training time in the prior art. The method is performed through the following steps: 1. according to the sketch model of an SAR image, extracting a sketch graph; 2. according to the sketch graph, obtaining a region graph; using the region graph to divide the SAR image into a hybrid pixel subspace, a homogeneous pixel subspace and a structural pixel subspace; 3. Building a ridgelet filter set and a fast ridgelet deconvolution structure learning model and using the learning model to segment the hybrid pixel subspace; 5. using the visual semantics and sketch characteristics to sequentially segment the structural pixel subspace and the homogeneous pixel subspace; and 6. Combining the segmentation results of the three pixel subspaces to obtain a final segmentation result. The method of the invention improves the segmentation effect of an SAR image and can be used for detecting and recognizing subsequent SAR images.

The invention discloses an image segmentation method. The method comprises that an image to be segmented is loaded; a characteristic vector of each pixel in the image is collected; selection of a region of interest (ROI) and a non region of interest (NROI) is received; according to characteristic vectors of pixels in the ROI, the pixels in the ROI are clustered into a first amount of interest classifications, and classification information of each interest classification is determined; according to characteristic vectors of pixels in the NROI, the pixels in the NROI are clustered into a second amount of non interest classifications, and classification information of each non interest classification is determined; the probability that each pixel in the image is a pixel of interest is calculated; the color similarity of every two adjacent pixels in the image is calculated; and a label value of each pixel in the image is determined according to the probability that each pixel is the pixel of interest and the color similarity of every two adjacent pixels. The invention also discloses an image segmentation application in which the method is enforced and a computing device including the application.

The invention discloses a deep learning based optimization method for coronary arteriography image segmentation. The method includes using a Tensor object for storing a coronary arteriography image and obtaining a segmentation result through accelerated calculation in a neural network through a GPU; and optimizing the segmentation result of the coronary arteriography image through a network structure formed by combination of a cascaded module and a pixel restoration module added in the neural network. According to the invention, the speed can be improved by 0.083s in a single image iteration process and more than 1 minute can be saved for data sets of thousands of levels in quantity in reality life. Besides, the neural network generally used for image styles is trained at least for 100 thousand times and more than 100 minutes can be saved in training of the whole network. At the same time, partial structure of the network is changed, so that the method ensures image segmentation accuracy and also has an advantage of reducing time length substantially and improves the segmentation accuracy.

The invention discloses a partition method for kidney artery CT contrastographic picture vessels based on a three-dimensional Zernike matrix. The partition method comprises the following steps: firstly, acquiring a forecast voxel point set and extracting a local geometric structure of the voxel point; then, constructing a descriptor with space rotation invariance by mapping the local geometric structure into a unit ball and solving the three-dimensional Zernike matrix; using the local geometric structure characteristic descriptors composed of different order and repeatability descriptors for expressing the characteristics of the local geometric structure in a quantizing form; and finally, adopting a study classifying model based on a support vector machine for classifying the acquired local geometric structure characteristic descriptors, thereby confirming the voxel points in the vessel lumen and acquiring a final vessel partition result through region growth. According to the method provided by the invention, the semiautomatic and accurate partition for kidney artery CT contrastographic picture vessels is realized, and the working efficiency of doctors and the accuracy of clinic auxiliary diagnosis are promoted.

The invention discloses an image segmentation method and device, equipment and a storage medium, and relates to the technical field of artificial intelligence. The method comprises the following steps: acquiring a first image sample and a second image sample, wherein the quality of annotation data of the first image sample is higher than that of annotation data of the second image sample; performing segmentation processing on the first image sample and the second image sample through a student network and a teacher network to obtain a student segmentation result of the first image sample, a student segmentation result of the second image sample, a teacher segmentation result of the first image sample and a teacher segmentation result of the second image sample; determining training loss according to the annotation data of the first image sample and the segmentation result; and training the student network based on the training loss. The image information is extracted from the samples with the low-quality labels through the teacher network, and the student network is trained, so that the problem of model overfitting caused by lack of the samples with the high-quality labels is avoided.

The invention provides a tongue image segmentation method based on color decomposition and the threshold technology. The method comprises steps that S1, a tone component of a tongue image in the HSI color space is transformed, and tone components of a tongue and an upper lip after transformation and the neighbor tissue are made to have tone difference; S2, image threshold segmentation of the transformed tone component is carried out to obtain the binary segmentation result, after morphological operation of the binary segmentation result is carried out, the initial tongue region containing thetrue tongue and the upper lip is extracted; and S3, the initial tongue region is corrected, two image threshold segmentation methods are selected, one of segmentation results of the two image threshold segmentation methods is utilized to separate the true tongue region from the upper lip region, and the final segmentation result of the tongue image is obtained. The method is advantaged in that themethod is simple and effective, and image segmentation performance is substantially improved.

The invention discloses a neural network model for segmenting an image, an image segmentation method and device thereof, and a readable storage medium, the neural network model comprises an intelligent selection module, and the intelligent selection module further comprises a feature extraction unit and an intelligent selection unit. As the feature extraction unit adopts multi-scale cavity convolution, information of different scales of an input feature map is obtained, and a large amount of rich feature information is provided for subsequent feature screening; and the intelligent selection unit carries out intelligent screening on the input feature map channel by training a weight value according to the size of the weight value, so that the intelligent selection module can reduce the parameter quantity and the calculated quantity while ensuring the segmentation precision. Therefore, the neural network model provided by the invention can quickly extract effective features of the imageby adopting the intelligent selection module, is small in calculation amount and few in model parameters, and is suitable for a mobile terminal.

The invention relates to a lung parenchyma segmentation method for extracting a CT image based on a clustering key frame, and the method comprises the following steps: carrying out the preprocessing of lung CT image data, and carrying out the lung window windowing gray scale transformation of an image CT value; performing clustering analysis on the grey level histogram similarity of the patient lung CT image processed in the step 1, and extracting key frame data in the lung CT sequence; carrying out pulmonary parenchyma segmentation on the patient key frame CT image by adopting a multi-phase level set CV model; performing lung parenchyma mapping segmentation extraction in the lung CT complete sequence according to a lung parenchyma segmentation result in the key frame to obtain a lung parenchyma initial contour; and performing morphological corrosion and expansion operation on the lung parenchyma initial contour in the lung CT image to refine the contour, and finally obtaining a lung parenchyma segmentation result in the patient lung CT data. The invention also provides lung parenchyma segmentation equipment for realizing the segmentation method.

The embodiment of the invention provides a segmentation method and device for a target object in a three-dimensional image and electronic equipment. The method comprises the following steps: accordingto a plurality of branch feature three-dimensional networks of a multi-channel three-dimensional network model, respectively carrying out feature extraction on three-dimensional images of a pluralityof modal groups to be segmented to obtain branch feature maps of a plurality of branches; according to the fusion feature three-dimensional network, performing feature extraction and fusion on the branch feature maps of the plurality of branches to obtain a fusion feature map; and according to the size amplification three-dimensional network, performing fusion and size amplification on the fusionfeature map and the branch feature maps of the plurality of branches to obtain a three-dimensional image of the segmented target object. In the embodiment of the invention, different morphological characteristics of the same target object in the three-dimensional images of different modal groups are extracted and fused, and the type and edge recognition precision of the target object is greatly improved, so that the segmentation precision of the target object is improved.

The invention provides a cell nucleus segmentation method and device, and the method comprises the steps: inputting a to-be-segmented image into a trained segmentation model under the condition that the resolution of the to-be-segmented image is not greater than a preset resolution, and obtaining a segmentation result of a cell nucleus in the to-be-segmented image, wherein the preset resolution isdetermined by the fixed input resolution of the trained segmentation model, and the trained segmentation model is obtained by respectively training preset segmentation models by adopting different types of training samples; training a preset segmentation model by taking a preset natural scene image as a training sample to obtain a first segmentation model; training the first segmentation model bytaking the multi-class tissue pathological images corresponding to the various tissues as training samples to obtain a second segmentation model; and finally, training the second segmentation model by taking the multi-class histopathological image of the required tissue as a training sample to obtain a trained segmentation model. The applicability of the segmentation method is improved.

The invention discloses an eye image segmentation method based on sclera region supervision, and mainly solves the problem of low segmentation precision of a traditional method. According to the scheme, the method includes the following steps: extracting high-dimensional features of a sclera area through a residual network; performing attention adjustment on the high-dimensional features of the original eye image by using the high-dimensional features; encoding the high-dimensional features of the adjusted original eye image to obtain encoded semantic features; improving the coding semantic features through cross-connection excitation, and inputting the coding semantic features into a decoder for decoding to obtain decoding semantic features; performing channel adjustment on the decoded semantic features, and outputting a preliminary segmentation result; and calculating the total loss of the initial segmentation result and the segmentation label, comparing the total loss with a set threshold value, judging whether all filters, encoders and decoders need to be optimized, and outputting a final segmentation result of the pupil, the iris and the sclera. The method improves the segmentation precision, and can be used for human eye positioning, blink detection, sight line estimation improvement and pupil change monitoring.

The invention discloses an intelligent segmentation method for a video image target, and the method employs a YUV color space and a specific distance measurement to overcome the impact from illumination, maintains a plurality of clustering centers to process a dynamic background, and employs a maximum continuous unmatched time length parameter to exclude foreground pixels out of a background model. According to the method, structured background motion can be obtained for a long time in a limited storage space, and a compact model can be established for a dynamic background; the influence of illumination on background modeling and foreground detection can be overcome; periodically switching between a modeling stage and a detection stage is carried out to meet the application requirement oflong-time uninterrupted operation of video monitoring. The method has better segmentation accuracy and higher processing speed, and is more suitable for video object segmentation of scenes such as passenger flow statistics, traffic flow video monitoring, industrial automatic monitoring and safety protection.

The invention relates to a virtual manicure try-on method based on deep learning, and the method comprises the steps: obtaining a single image or video image containing a nail, employing a deep learning full convolution neural network method to segment and track a nail region in real time, combining with the color of the nail selected by a user and the re-coloring of a pattern template, and achieving the virtual manicure try-on function. Compared with the prior art, the method has the advantages that the deep learning and virtual reality technologies are adopted, the shot single picture or continuous video images containing the nails are processed, the nails are segmented and tracked in real time, and the function of virtual manicure try-on is achieved in combination with the selected templates such as the colors / patterns of the nails. According to the method, through the form of a mobile phone APP, customers are not limited by places and equipment, manicure try-on and selection are completed, and the experience feeling of the customers is enhanced.

The invention relates to an iris segmentation neural network model training method and an iris segmentation method and device. The iris segmentation neural network model training method comprises thesteps: acquiring an iris recognition sample set, wherein the iris recognition sample set comprises a plurality of iris recognition samples, and the iris recognition samples comprise iris recognition tags; training an iris recognition network model based on a network automatic search technology and the iris recognition sample set; determining an iris segmentation neural network model, wherein the iris segmentation neural network model takes an iris recognition network model as a basic feature network; acquiring an iris segmentation sample set, wherein the iris segmentation sample set comprisesa plurality of iris segmentation samples, and the iris segmentation samples comprise iris segmentation labels; and training an iris segmentation neural network model based on the iris segmentation sample set. Through the method and the device, the consumption of computing resources can be reduced on the premise of ensuring the segmentation precision.