32results about How to "Improve Semantic Segmentation" patented technology

The embodiments of the present invention belongs to the computer vision technological field and provide a semantic segmentation network training method, an image semantic segmentation method and corresponding devices. The semantic segmentation network training method includes the following steps that: a to-be-trained image is acquired; the to-be-trained image is inputted into a pre-established semantic segmentation network, the front network layer of the semantic segmentation network is adopted to extract the features of the to-be-trained image, and a feature map containing the block, global and edge features of the to-be-trained image can be obtained; the feature image containing the block, global and edge features of the to-be-trained image is inputted to the rear network layer of the semantic segmentation network so as to be subjected to image pixel classification, so that a semantic segmentation image containing segmentation pixel types can be obtained; and the parameters of the semantic segmentation network are update according to the semantic segmentation image. Compared with the prior art, the method of the invention separately extracts and restores the edge features of theto-be-trained image, thereby improving the training recognition effect of the edge of a segmentation region.

The invention provides a graffiti-based weak supervision semantic segmentation method and system, and the method comprises the steps: obtaining a plurality of training images which correspond to graffiti marks and edge images; selecting a training picture as a current picture, and inputting the current picture into the semantic segmentation network to obtain high-level semantic features of the current picture; inputting the high-level semantic features into a prediction correction network to obtain a segmentation result graph of the current picture, and obtaining cross entropy loss of a graffiti mark area in the current picture according to the graffiti mark of the current picture; inputting the high-level semantic features into a boundary regression network to obtain a boundary map of a target in the current picture, and obtaining mean variance loss of a boundary region in the boundary map according to an edge map of the current picture; constructing a total loss function, judging whether the total loss function is converged or not, and if yes, taking the current prediction correction network as a semantic segmentation model; and inputting the picture to be semantically segmentedinto the semantic segmentation model to obtain a segmentation result picture of the picture to be semantically segmented.

The invention relates to a three-dimensional liver image semantic segmentation method based on a context attention strategy. The method comprises the following steps: selecting a medical image data set to be subjected to liver segmentation, and dividing the medical image data set into a training set and a test set; preprocessing the three-dimensional liver image in the training set; in the encoding stage, obtaining a feature map of the liver by using a residual structure, a convolutional network and cavity convolution; in a decoding stage, obtaining a segmented image of the liver by using a context attention strategy module, transpose convolution and a depth supervision mechanism; and carrying out post-processing on the liver image obtained after semantic segmentation. The method has the characteristic of improving the semantic segmentation effect of the three-dimensional liver image, realizes a better automatic segmentation effect, and can assist doctors in diagnosis.

The invention discloses a remote sensing image semantic segmentation method fusing improved UNet and SegNet. Batch processing standardization is added between a convolution layer and an activation layer of the UNet neural network; an ELU activation function is adopted to replace a ReLU activation function; respectively training each semantic segmentation category by adopting a binary classification training mode; combining the models trained by the binary classification; in the coding process of the SegNet neural network, the coding speed is improved; performing a maximum pooling operation; introducing a result of a front setting layer in the SegNet neural network to carry out convolution operation; step short-circuit connection is carried out on a convolution operation result; the remotesensing image semantic segmentation method comprises the steps of obtaining an improved SegNet neural network by reducing the number of partial network layers of SegNet, fusing the improved UNet neural network and the improved SegNet neural network to obtain a remote sensing image semantic segmentation model, and performing semantic segmentation to improve the effect of performing semantic segmentation on a remote sensing image.

The invention discloses an image semantic segmentation method based on a convolutional neural network. The method comprises the following steps that ResNet101 is selected as a skeleton network to carry out feature extraction, a JFP model is provided to combine the last three layers output by the ResNet101, feature extraction of the ResNet101 is perfected, and the problem of image information lossis solved; then the output of the JFP is accessed to an ASPP model to further extract the spatial scale information of the image, and the part is used as a coding structure to better extract the imageinformation; and finally, the output image of the neural network is restored to the original size by using a simple decoding structure to finish semantic segmentation of the image. Meanwhile, an attention model is designed, a loss function of the model is combined with a loss function of the semantic segmentation network, the network is assisted in training, and the effect of training the model is improved. The method remarkably improves the image semantic segmentation effect in a complex scene, can be suitable for various scenes, and has the semantic segmentation processing capability for images of more than 20 types of objects.

The invention relates to a turbulence degradation image semantic segmentation method based on boundary perception and adversarial learning, and the method comprises the following steps: (1), combininga simulated turbulence-degraded image with a real turbulence-degraded image based on a turbulence imaging physical model for a turbulence-degraded image, and constructing a turbulence-degraded imagesemantic segmentation data set; (2) for the data set obtained in the step (1), constructing a DeepLabV3 + semantic segmentation model based on boundary perception in combination with fuzzy and distortion characteristics of a turbulence degradation image; (3) taking the DeepLabV3 + semantic segmentation model based on boundary perception in the step (2) as a generator, and constructing a boundary perception generative adversarial network GAN model based on adversarial learning in combination with a discriminator composed of five convolutional layers; and (4) for the GAN model obtained in the step (3), performing model training on the turbulence degradation image semantic segmentation data set obtained in the step (1) to obtain a trained semantic segmentation GAN model, and performing semantic segmentation on the turbulence degradation image by using the trained semantic segmentation adversarial network GAN model to obtain a prediction segmentation graph.

The invention discloses a liver image semantic segmentation method based on an edge attention strategy, and the method comprises the steps: S1, carrying out the format conversion of an original data set; s2, preprocessing the liver image data; s3, analyzing edge features of the liver image and designing an Encoder-Decoder deep learning framework model, wherein the Encoder-Decoder deep learning framework model comprises an encoding stage and a decoding stage; s4, in a coding stage, obtaining an edge attention graph by utilizing a ResNet34 residual network and an edge attention strategy module;s5, in a decoding stage, performing feature extraction and segmented image generation by utilizing a deconvolution and edge attention strategy module; and S6, performing noise reduction processing onthe image after semantic segmentation. Through training and fusion of the edge attention strategy module and the deep convolutional network, the loss function is optimized, the depth features of the liver image are extracted and segmented, and the method has the characteristic of improving the semantic segmentation effect of the liver image.

The invention discloses a lung parenchyma CT image segmentation method based on a weighted full convolutional neural network, and belongs to the field of medical image processing. The method comprisesthe following steps: selecting a public lung data set for preprocessing, and extracting a lung parenchyma boundary in a labeled image as a semantic category; designing an improved network structure based on a standard full convolutional neural network framework, and establishing an overall structure framework of the pulmonary parenchyma segmentation convolutional neural network by using a principle that a standard path structure for encoding and decoding simultaneously comprises jump connection, expansion convolution and batch normalization; adopting a weighted loss function layer; dividing the data set; carrying out offline model training out to acquire model weight parameters; inputting a test image and outputting a segmentation result by an output layer through layer-by-layer feedforward of a network. According to an existing lung parenchyma segmentation method, a segmentation missing phenomenon is prone to occurring in a focus area in lung parenchyma, and correct segmentation of the focus area in lung parenchyma segmentation can be effectively improved through enhancement processing on important pixels.

The invention relates to a semantic segmentation method in an automatic driving scene based on BiSeNet, and the method comprises the following steps: S1, collecting urban street image data, and carrying out the preprocessing; S2, labeling the preprocessed image data to obtain labeled image data; S3, performing data enhancement on the labeled image data, and taking the enhanced image data as a training set; S4, constructing a BiSeNet neural network model, and training the model based on the training set; and S5, preprocessing the video information acquired by the camera, and performing semanticsegmentation on the urban streets in the camera according to the trained BiSeNet neural network model. The safety of automatic driving and the accuracy and rapidity of road scene segmentation can beeffectively improved.

The invention discloses an infrared image power equipment segmentation method based on a generative adversarial network. The method comprises the following steps: S1, preparing network input data; s2,constructing a generative adversarial model: Cycle-Gan; s3, designing a thermal imaging image semantic segmentation model; and S4, designing a loss function of the segmentation network. In the present invention, the problem of how to eliminate the difference between different modal data of a thermal imaging image is solved, an existing pre-training segmentation model is fully utilized, a generative adversarial network is utilized to establish a relationship between different modal data, so that a thermal imaging image can fully utilize the advantages of an existing semantic segmentation modelbased on a natural image, and a thermal imaging image semantic segmentation model with more excellent performance is realized.

The invention discloses an automatic terrain sundry identification method, and relates to a terrain sundry identification system based on deep learning. The system performs a series of processing on low-precision three-dimensional terrain point cloud data acquired based on an unmanned aerial vehicle aerial photography technology. To solve the problem that low-precision terrain feature informationis limited, geometrical features and color features of terrain point clouds are reconstructed, a multi-attribute descriptor is designed, and local information of each point in the input terrain pointclouds is described; then, the terrain point clouds are aggregated into a plurality of super points by using a multi-attribute descriptor in unsupervised classification, so that computation complexityis reduced; a directed attribute graph is defined and the super points are combined into a graph structure, namely a super point graph; and finally, the nodes of the super-point graph are classifed by using the proposed LSTM algorithm based on depth graph convolution. The system can finally perform semantic segmentation on the sundries in the terrain point cloud, and the sundries in the terrain can be displayed through different colors. Through further research and analysis, basic technical support can be provided for topographic analysis and the like.

The invention discloses a pixel-level label automatic generation model construction and automatic generation method and device. The pixel-level label automatic generation model construction and automatic generation method comprises the steps: firstly learning segmentation knowledge from a source domain through a pixel-level label, then transferring the knowledge to a target domain to generate a coarse label of an image, carrying out the reasoning of the coarse label through a guide filter, and generating a fine label. On the basis of refining the label, the segmentation network is optimized, and the fine-grained target label with a detailed pixel-level structure / boundary is generated, and the semantic segmentation effect is improved.

The invention is suitable for the technical field of deep learning and automatic driving, and provides a semantic segmentation method and system for automatic driving, electronic equipment and a medium, and the method comprises the steps of obtaining three-dimensional point cloud data, mapping the three-dimensional point cloud data into a two-dimensional depth map comprising a plurality of pieces of channel data, and forming a sample data set according to the plurality of pieces of channel data; building a semantic segmentation network initial model comprising a first model and a second model, using the sample data set to train the semantic segmentation network initial model, obtaining a target model, where the second model comprises an encoder and a decoder; obtaining target three-dimensional point cloud data, mapping the target three-dimensional point cloud data into a target two-dimensional depth map, inputting the target two-dimensional depth map into the target model, and obtaining a target semantic segmentation result. Through adoption of the method, the problem of low semantic segmentation performance of the three-dimensional point cloud is solved.

The invention discloses a semantic segmentation method, device and equipment and a computer readable storage medium. The method comprises steps of acquiring to-be-processed characteristic informationof a to-be-processed image, wherein the to-be-processed image comprises a first mode image and a second mode image corresponding to the first mode image, and the to-be-processed characteristic information comprises first characteristic information of the first mode image and second characteristic information of the second mode image; connecting the first feature information and the second featureinformation to obtain first connection feature information; determining compensation information of the feature information to be processed according to the first connection feature information; correcting the to-be-processed feature information according to the compensation information to obtain corrected feature information; and determining a semantic classification result of each pixel in the to-be-processed image according to the correction feature information.

The invention belongs to the technical field of intelligent application of remote sensing images, designs a semantic segmentation network based on optical and PolSAR feature fusion, and is used for realizing an end-to-end optical and PolSAR fusion ground feature segmentation task. The device is composed of a double-path twin convolution feature encoder, an attention mechanism module ASCAM and a symmetric jump connection decoder. Wherein the dual-path twin convolution feature encoder adopts ImageNet and a large number of labeled PolSAR and light images to carry out pre-training on each path of encoder. And then designing an ASCAM, obtaining a nonlinear influence relationship of light at a local position and each channel of the PolSAR on a classification task through an expansion space matrix, and combining the expansion space matrix with data of a double-channel twin convolution feature encoder to realize weighted guidance of attention on a feature fusion process. And finally, realizing fusion of weighted features through convolution, designing a symmetric jump connection decoder, performing jump connection on the fused features and coding process features in optical and PolSAR, and finally realizing end-to-end ground feature segmentation.

The invention discloses a method and device for automatically learning network preprocessing parameters, electronic equipment and a storage medium, and the method comprises the steps: carrying out the training of a semantic segmentation model through randomly constructed image processing parameter items, and enabling the image processing parameter items to be used for adjusting the display effect of each picture, the image processing parameter item comprises an image processing parameter value; inputting the plurality of pictures into a preset parameter preprocessing network to obtain an image processing target parameter value corresponding to each picture; according to the image processing target parameter value corresponding to each picture, adjusting the picture to be subjected to semantic segmentation processing to obtain a plurality of target pictures; and inputting the plurality of target pictures into the semantic segmentation model obtained by pre-training, and carrying out image semantic segmentation. According to the invention, the semantic segmentation effect of the semantic segmentation model during prediction is improved.

The embodiment of the present invention relates to the field of computer vision technology, and provides a semantic segmentation network training method, image semantic segmentation method and device, the semantic segmentation network training method includes: acquiring an image to be trained; inputting the image to be trained into a pre-established semantic segmentation Network, using the front network layer of the semantic segmentation network to perform feature extraction on the training image to obtain a feature map containing the block, overall and edge features of the image to be trained; the feature map containing the block, overall and edge features of the image to be trained The image is input into the back network layer of the semantic segmentation network to classify the image pixels, and the semantic segmentation map including the segmentation pixel type is obtained; the parameters of the semantic segmentation network are updated according to the semantic segmentation map. Compared with the prior art, the embodiment of the present invention separately extracts and restores the edge features of the image to be trained, which improves the training recognition effect at the edge of the segmented area.

The present invention proposes a method and system for identifying cancer focus areas based on pathological images, including: obtaining multiple pathological images with labels, each pathological image has a content label marking the cancer focus area and a contour label for the outline of the cancer focus; Construct a picture classification model including encoder, content decoder and contour decoder. The encoder is used to downsample the image to obtain the downsampling feature map. The contour decoder decodes the profile feature map according to the downsampling feature map. The contour decoder has to The skip connection of the content decoder, through the skip connection, the content decoder decodes the content feature map according to the downsampling feature map and the profile feature map; input the marked pathological image into the picture classification model, and train the picture at the same time in the way of joint supervision The classification model performs content recognition and contour recognition tasks, and calculates the loss function, and updates the image classification model through the loss as the cancer focus area recognition model; the cancer focus area is identified through the cancer focus area recognition model.

The present invention relates to a semantic segmentation method based on BiSeNet in an automatic driving scene, comprising the following steps: step S1: collecting urban street image data, and preprocessing; step S2: labeling the preprocessed image data, and obtaining the labeling Step S3: perform data enhancement on the marked image data, and use the enhanced image data as a training set; Step S4: build a BiSeNet neural network model, and train the model based on the training set; Step S5: collect the camera The video information is preprocessed, and the city streets in the camera are semantically segmented according to the trained BiSeNet neural network model. The invention can effectively improve the safety of automatic driving and the accuracy and speed of road scene segmentation.

Segmentation method of infrared image power equipment based on generative confrontation network, including the following steps: S1: Prepare network input data; S2: Build generative confrontation model: Cycle-Gan; S3: Design thermal imaging image semantic segmentation model; S4: Design segmentation network loss function. In the present invention, how to eliminate the difference between different modal data for thermal imaging images is fully utilized by using existing pre-trained segmentation models, and using generative confrontation networks to establish connections between different modal data, so that thermal imaging images can be fully utilized Based on the advantages of the existing natural image semantic segmentation model, a thermal imaging image semantic segmentation model with better performance is realized.

The invention relates to a land classification method based on a high-resolution remote sensing image, and the application of the high-resolution remote sensing image can effectively achieve the dynamic monitoring of land utilization, and improves the scientificity of land management. By using the ERFNet in combination with the conditional random field (CRF), the semantic segmentation precision can be further improved on the basis of reducing computing resources, and the land use classification work can be better completed; and an efficient land use classification method plays an important role in promoting the development of land utilization, urban planning, environmental monitoring and military fields.

According to the cross-domain semantic segmentation method based on three-branch cross training, common knowledge and unique knowledge among different views are fully mined, the generalization ability and segmentation ability of each view are improved, and meanwhile diversity among the views is kept. The method comprises an asymmetric three-branch network and two indirect pseudo tag alignment techniques. The three-branch network comprises three segmentation networks with different network structures, the three segmentation networks share one feature extraction network, and any two networks generate high-credibility pseudo labels for the third network. The indirect pseudo-tag alignment technique comprises semantic feature alignment, feature center points of a segmentation network are extracted through pseudo-tags, a simple category only needs to pull close the center points of the same category, and a complex category also needs to pull away the center points of different categories; comprising the following steps: estimating the adaptive capacity, evaluating the difference between two views for generating pseudo labels, and aligning the output of each pixel point according to the intensity of the adaptive capacity.

The invention belongs to the technical field of semantic segmentation, and particularly relates to a semantic segmentation method for an unmanned aerial vehicle (UAV) aerial video based on UVid-Net. The method comprises the following steps: data collection: collecting a data set for unmanned aerial vehicle video semantic segmentation, carrying out the pixel-level marking of the type of the data set, and completing the construction of the data set needed by model training; data preprocessing including normalization, data division, image scaling and the like, and amplifying a data set to ensure a model training effect; model identification: constructing a semantic segmentation model based on UVid-Net, inputting training data, and completing the construction of a parameter model; model storage: when the loss function of the model is not reduced any more, storing the model; and model evaluation: performing performance evaluation on the segmentation result of the model through a plurality of evaluation indexes. According to the coding path, the time dynamic state of the video is captured by extracting features from multiple frames, the features of a coder layer can be reserved, and the semantic segmentation performance is improved.

The invention discloses a deconvolutional network and adaptive inference network based SAR image segmentation method and mainly solves the problem that human experience is required to extract image features in the prior art. The method is implemented by the steps of (1) sketching an SAR image; (2) extracting a complemented region graph of the SAR image; (3) training a deconvolutional network DNN for aggregation regions and homogeneous regions separately; (4) performing adaptive comparison inference on structural features of non-communicated regions in the aggregation and homogeneous regions to obtain segmentation results of the aggregation and homogeneous regions; (5) based on a watershed method, segmenting a structural region obtained in the step (2); and (6) combining the aggregation regions, the homogeneous regions and the structural region to obtain a segmentation result. According to the method, the segmentation result has relatively good regional consistency and the segmentation effect of the SAR image is improved. The method can be used for target detection and identification.