145 results about "Residual Blocks" patented technology

The invention discloses a method for predicting the elastic property of a fiber reinforced composite material, and aims to solve the problems of low precision, low efficiency and difficulty in processing multi-scale correlation of a traditional method. The method comprises the following steps: firstly, generating a representative volume unit by using a random algorithm, and constructing a high-quality training data set by combining Sobol sequence sampling and an SMOGN data enhancement technology; secondly, designing and training a deep neural network embedded with residual blocks and physical constraints, and improving the generalization ability of the model through Bayesian optimization adaptive parameter adjustment; and finally, establishing a plurality of macrostructure models, and realizing end-to-end prediction of the multi-scale elastic performance. The method realizes breakthrough in prediction precision and calculation efficiency: the time consumed by single analysis is shortened from several hours to a minute-second level, and the average prediction error is lower than 5%. The technology can be widely applied to the fields of aerospace, new energy automobiles, wind power and the like, and intelligent support is provided for design and manufacturing of high-performance composite materials.

The invention provides a self-distillation model compression method and device, an electronic product and a medium. The method comprises the steps that a teacher model needing model compression is constructed, the basic structure of the teacher model is divided into a plurality of residual blocks, and student models are constructed according to the residual blocks; through knowledge distillation of each model, enhancing result knowledge; enhancing process knowledge through attention map mapping of feature maps of the teacher model and each student model; according to the enhancement result of the effective knowledge and the enhancement result of the process knowledge, calculating a loss function of the teacher model and each student model; according to the loss function obtained through calculation, a student model with the classification effect closest to the teacher model is extracted to serve as a compression model of the teacher model. According to the self-distillation model compression method and device, the electronic product and the medium provided by the invention, a method technical support is provided for forming a simple, convenient, high-automation and user-friendly model compression and acceleration tool.

The invention relates to the technical field of image super-resolution, and provides an image super-resolution reconstruction method, and the method comprises the steps: obtaining a to-be-reconstructed low-resolution image, inputting the to-be-reconstructed low-resolution image into a trained super-resolution image reconstruction model, and obtaining a super-resolution reconstruction image corresponding to the to-be-reconstructed low-resolution image. A shallow layer feature extraction module in the super-resolution image reconstruction model is composed of a depth separable convolution sub-module and a standard convolution sub-module, and a deep layer feature extraction module is composed of a plurality of dense parallel residual blocks constructed based on an adaptive attention mechanism and a standard convolution layer which are connected in sequence. The reconstruction module is composed of a dynamic convolution layer, a moving window attention layer, a pixel reconstruction layer and a standard convolution layer, and solves the problems that an existing shallow feature extraction module cannot fully extract shallow features, the local modeling capability of a residual Swin Transform block in a deep feature extraction module is insufficient, and the reconstruction module cannot obtain features of a target area. Therefore, a high-quality super-resolution reconstructed image is obtained.

The invention relates to an EDSR (Enhanced Depth Super-Resolution) image super-resolution reconstruction method based on particle swarm optimization, and the method comprises the following steps of: (1) carrying out super-resolution reconstruction on an EDSR (Enhanced Depth Super-Resolution) image; the method comprises the following steps: firstly, inputting a low-resolution image data set as a training sample, defining an optimization space containing the number and stage of residual blocks, convolutional layer parameters, an attention module, an up-sampling mode and the like, initializing particle swarm optimization (PSO) parameters, and dynamically constructing a candidate network by particle position coding; a candidate network is dynamically constructed through particle position coding, and a residual block layer, a convolution layer, an attention module and an up-sampling module are sequentially configured. The candidate network is subjected to limited training, and individual and global optimal positions are updated through fitness function evaluation fusing PSNR and model complexity. And finally, a global optimal structure is selected for complete training for low-resolution image reconstruction, the detail reduction capability and the reasoning efficiency are remarkably improved, and a high-quality image is generated.

The present invention discloses a method for intelligently identifying individual communication radiation sources in cross-period scenarios, relating to the technical field of radiation source individual identification. To address the problem of the degradation of individual distinguishability of radiation sources over time due to the inherent characteristics of communication signal radiation sources and the influence of complex electromagnetic environmental factors, the present invention, based on a generative adversarial network, constructs a generator using residual blocks to establish a mapping relationship between the feature distributions of data from two periods. By adding noise to the discriminator and discarding neurons with a certain probability, the model training process is stabilized and the risk of overfitting is reduced. Through adversarial training, a robust classifier with strong generalization capabilities is constructed. Through the above technical solutions, the present invention can achieve stable and effective individual identification of communication radiation sources in cross-period scenarios.

The invention discloses a confrontation and defense method based on dynamic gating mixed batch normalization. The method comprises the following steps: step 1, data acquisition and model initialization; 2, calling an adversarial attack algorithm to generate an adversarial sample; combining the adversarial sample and the clean sample to form a mixed sample; step 3-1, establishing a wide residual network model; the initial convolution layer passes through three network frameworks with the same structure, and the output of the last network framework sequentially passes through a standard batch normalization layer, a ReLU activation function, a global pooling layer and a full connection layer; each network framework comprises four residual blocks; 3-2, learning mixed batch normalization layer statistical distribution by using clean samples and adversarial samples; and 3-3, freezing parameters of the mixed batch normalization layer, and learning parameters of other network layers in the wide residual network model. According to the invention, by introducing a dual-path dynamic normalization and adaptive gating fusion mechanism, the robustness, generalization ability and calculation efficiency of the model are significantly improved.

A method for video processing applied to a decoder includes (i) receiving an input image; (ii) processing the input image by one or more convolution layers; (iii) processing the input image by multiple residual blocks by using partition information of the input image as reference so as to obtain reference information features; (iv) generating different-scales features based on the reference information features; (v) processing the different-scales features by multiple convolutional layer sets; (vi) processing the different-scales features by reference spatial attention blocks (RSABs) so as to form a combined feature; and (vii) concatenating the combined feature with the reference information features so as to form an output image.

The invention discloses an Alzheimer disease prediction method and system based on deep learning and electrocardiosignals. Firstly, electrocardiosignals are collected and preprocessed; then, constructing a deep learning prediction model which comprises a feature extraction module and a classification prediction module; the feature extraction module comprises a plurality of residual blocks, and each residual block comprises a convolution layer, a batch normalization layer, a nonlinear activation layer and a jump connection; the classification prediction module realizes a complete mapping process from a feature space to probability prediction through a multi-layer perceptron structure and a softmax function, and obtains a prediction probability of each electrocardiosignal fragment belonging to each category; and fusing the prediction results of the plurality of electrocardiosignal segments of the same sample by adopting a soft voting mechanism to obtain a final prediction result. The system comprises a signal acquisition module, a signal preprocessing module, a deep learning prediction module and a result output module. According to the method, Alzheimer's disease prediction is realized based on the electrocardiosignals, and the method is simple and convenient in acquisition mode, non-invasive, low in cost and suitable for large-scale screening.

The invention provides a flight delay prediction method and system, computer equipment and a storage medium, and belongs to the technical field of traffic delay prediction.The method comprises the steps that firstly, a bacterial foraging algorithm (BFA) is optimized through a genetic algorithm (GA) so as to improve the global search ability and convergence speed of the algorithm; secondly, optimizing the structure and parameters of a deep neural network DNN by using the optimized BFA algorithm, and constructing a deep neural network model based on double hidden layers; according to the model, an attention mechanism and a residual block are introduced, so that the nonlinear mapping capability and the generalization capability are improved, and model overfitting is effectively prevented. In addition, the flight data and the weather condition data are combined, so that the prediction accuracy is further improved. The performance of the model is evaluated through multiple indexes, the result shows that the method can effectively solve the problem of DNN structure and parameter selection, and the training efficiency and generalization ability of the model are remarkably improved.

The invention relates to a quality control classification method and system suitable for laryngeal knot DR images, and the method comprises the steps: carrying out the preprocessing of each obtained historical laryngeal knot DR image, and obtaining a preprocessed image; dividing a training set, a verification set and a test set according to a preset distribution proportion based on the preprocessed historical laryngeal knot DR image set; the training set and the verification set are input into a classification network model for model training, the classification network model is composed of a plurality of cascaded residual blocks, and after output of each residual block, an attention and multi-scale aggregation module is integrated, so that an attention and multi-scale aggregation model is formed; the attention and multi-scale aggregation module is composed of a CBAM space channel attention module and an MSAA multi-scale feature aggregation module which are arranged in sequence; inputting the test set into the trained classification network model, and optimizing model parameters based on a model performance evaluation result; and acquiring a real-time laryngeal knot DR image, inputting the real-time laryngeal knot DR image into the trained and optimized classification network model, and processing the real-time laryngeal knot DR image to obtain a corresponding classification result.

This invention discloses a deep learning-based Hadamard transform-based anti-screen-capture watermarking method, belonging to the field of watermarking technology. This invention combines a convolutional neural network (CNN) and residual blocks to achieve an end-to-end process of watermark embedding and extraction within the Hadamard domain. By simulating screen-capture attacks and incorporating this method between the embedding and extraction layers, the network ensures robust watermark embedding. This method can efficiently extract watermark information from photos taken by surreptitiously, protecting the copyright of digital media. The use of the transform domain for watermark embedding allows the watermark to spread over a wider area of ​​the image, significantly improving the robustness of the watermarking algorithm. Comparison with related technologies demonstrates the superiority of this invention in terms of imperceptibility and robustness.

This invention discloses an adaptive edge-aware 3D medical image segmentation method, with the following specific steps: S1, constructing an adaptive edge-aware network, which includes an encoder and a decoder, with a skip connection between the encoder and decoder; S2, acquiring and processing a 3D medical image; S3, inputting the preprocessed image from step S2 into the encoder of the adaptive edge-aware network through a patch partitioning layer, then into the decoder through residual blocks and adaptive weight matching blocks. The decoder output and the original input image are skip-connected through adaptive weight matching blocks, and finally, the image segmentation result is output through residual blocks and Fourier convolution. This invention exhibits stronger robustness and boundary accuracy in multi-organ 3D segmentation tasks, providing an efficient and scalable solution for medical image segmentation.

Disclosed in the present application are an encoding method, a decoding method, a bitstream, an encoder, a decoder, and a storage medium. The method comprises: decoding a bitstream, so as to determine the value of first syntax identifier information; when the first syntax identifier information indicates that a multi-transform combination mode is applied to the current block, determining at least two first transform coefficients of the current block; determining at least two transform cores of the current block, and performing inverse transform on the at least two first transform coefficients respectively on the basis of the at least two transform cores, so as to determine at least two first residual blocks of the current block; and determining a reconstructed block of the current block on the basis of the at least two first residual blocks. In this way, the processing efficiency can be improved, thereby improving the encoding and decoding performance.

This invention discloses a dynamic obstacle avoidance method for assistive visually impaired scenarios. First, a feature extraction module based on a hierarchical residual structure constructed within residual blocks is built, which expands the receptive field of deep features. Second, a spatial feature recovery module based on bilinear interpolation and transposed convolution upsampling is designed to make segmentation edges more accurate. Third, a discrete sampling strategy is used to extract obstacle category, distance, and contour information, and path planning is performed using a heuristic search algorithm considering safe distance constraints. Finally, priority rules and logical order rules for the obstacle avoidance warning system are assumed, and a decision is generated based on the planned path, effectively guiding blind users to avoid obstacles through auditory and tactile information. This invention solves the problem that single-stage instance segmentation algorithms struggle to simultaneously and accurately segment small target obstacles and background road surfaces, overcoming the difficulty of providing effective information for obstacle avoidance decisions in assistive visually impaired scenarios with instance segmentation results, enabling blind users to avoid obstacles more intelligently and autonomously.

Disclosed are a video signal processing method and device whereby a video signal is encoded or decoded. The video signal processing method may comprise the steps of: determining whether an intra sub-partition (ISP) mode is applicable to a current block; if the ISP mode is applicable to the current block, splitting the current block into a plurality of rectangular transform blocks in the horizontal or vertical direction; generating prediction blocks of the transform blocks by carrying out intra prediction on each of the transform blocks; and reconstructing the current block on the basis of residual blocks of the transform blocks, and the prediction blocks.

This invention provides a weight training method and system for generating geological lithological texture models based on LoCon, belonging to the field of geological lithological texture generation technology. It employs an LDM model as the large model framework for weight training and texture generation. Convolutional layers are integrated into the convolutional weights of the residual blocks in the U-Net model within the LoCon model. The low-rank matrix of the LoCon model is added to the pre-trained weight matrix in the U-Net model to construct the generated geological lithological texture model. The model is trained using a training set. The trained weights are output, loaded into the large model framework for inference, and the training parameters are adjusted based on the inference results. Training is repeated until the preset inference result requirements are met. An API is provided to create an interface between the LDM large model framework and the trained weight matrix. This fills a gap in the geological industry's material library and lowers the technical threshold for geologists to participate in digitization.

Various implementations described herein include methods and systems for video coding and decoding. In one aspect, a method includes receiving a video bitstream, the video bitstream including a set of inter-mode coding blocks and a corresponding set of transform coefficients. The method includes deriving a set of inter-mode residual blocks from a set of transform coefficients. The method includes determining whether to apply one or more inseparable transform kernels to an inter-mode residual block set according to a value of a first indicator in a video bitstream. The method includes applying a first inseparable transform kernel when the indicator has a first value, and abandoning application of the first inseparable transform kernel when the indicator has a second value. The method further includes reconstructing the set of video blocks using the set of inter-mode residual blocks and the corresponding set of prediction blocks.

This invention discloses a cross-domain fault diagnosis method for bearings, aiming to address the problem of insufficient diagnostic generalization caused by significant differences in the distribution of bearing fault features and the scarcity of target domain labels under varying operating conditions. The method first constructs a multimodal input including a standardized one-dimensional time-series signal, a short-time Fourier transform time-frequency plot, and a phase space reconstruction plot. It then extracts temporal dependencies, time-frequency textures, and phase space topological features in parallel through a three-branch network consisting of a temporal convolutional network, residual blocks, and an inverse residual layer. These features are dynamically fused using a Transformer encoder to generate global features. High-confidence pseudo-labels are generated based on a momentum prototype distance metric, and class-level domain alignment is achieved by combining a homoscedastic uncertainty weighted composite loss function. Finally, iterative training outputs the target domain fault diagnosis results. Experiments show that the method achieves cross-domain diagnosis accuracy of 98.04% and 99.79% on the bearing datasets from the University of Paderborn and Case Western Reserve University, respectively, demonstrating both high accuracy and strong generalization, making it suitable for bearing fault diagnosis in industrial variable operating conditions.

The application discloses an image super-resolution reconstruction method, system, device and medium, relates to the technical field of image super-resolution reconstruction, and comprises the following steps: extracting an image feature map; dividing an input feature map into multiple routing units, introducing a hybrid depth condition calculation in a deep feature extraction network composed of multiple residual blocks; and based on the output of the deep feature extraction network, reconstructing an image with a resolution higher than a threshold value. By introducing a hybrid depth condition calculation mechanism with displacement perception, the application designs a structured routing with displacement perception and a budget tendency for the shift window mechanism, so that the shift window sub-block is protected in terms of computing resources, and the cross-window information interaction link is more complete, so that block-like artifacts and fractures across the window boundary are less likely to occur when recovering long edges, repeated textures and large structures.

The present invention discloses an infrared target detection method based on cyclic reuse convolution, which is applied to the field of image processing technology. The method includes: using multiple reuse convolution encoders and multiple bidirectional attention aggregation decoders in a cyclic reuse convolution network to process a target image to be detected, so as to obtain target detection information; the convolution kernels of each reuse convolution encoder are the same, and the bidirectional attention aggregation decoder is used to fuse the shallow encoding features and deep features of adjacent layers; the level of the deep features is greater than that of the shallow encoding features; using the residual blocks in the cyclic reuse convolution network to detect the target detection information, so as to obtain a target detection result. In the present invention, the convolution kernels of the reuse convolution kernel encoders are the same, thereby reducing the parameters. In the decoder, bidirectional attention aggregation with low complexity is adopted to guide the progressive fusion of multi-scale features, so that both multi-scale features can be extracted, and the number of parameters can be reduced, and the computational complexity can be reduced.

The application discloses a processor, a chip, a device and a code rate estimation method, and belongs to the chip technical field.The processor comprises a code rate calculation unit and an accumulation unit.The code rate calculation unit is used for calculating a first code rate corresponding to an i-th residual block after the i-th residual block is acquired; and a second code rate is calculated based on N residual blocks.The accumulation unit is used for calculating the sum of the second code rate and the first code rates corresponding to the N residual blocks, so as to obtain a code rate estimation result.The processor does not consider the dependent relationship among the N residual blocks when calculating the first code rate corresponding to each residual block.After the current i-th residual block is acquired, the first code rate corresponding to the i-th residual block is calculated based on the i-th residual block itself.In this way, the problems of calculation delay and low efficiency caused by the fact that the code rate of the current residual block is calculated after all the residual blocks having the dependent relationship are acquired in the related art are avoided, and the calculation efficiency of the code rate estimation result of the encoding unit is improved.

The embodiments of the present application provide a method, device, electronic device and medium for sorting recommended information, which relate to the field of information processing technology and can more accurately determine the display order of recommended terms. The technical solution of the embodiments of the present application includes: for each term to be recommended, each feature vector in the feature group set of the term to be recommended is input into different expert modules of the recommendation model respectively, and the feature vector output by each expert module after processing the feature vector through the multiple residual blocks included is obtained. Among them, the recommendation model is a model obtained by training the multi-task learning MMOE model, and each expert module of the MMOE model includes multiple residual blocks. Then, the score of the term to be recommended is determined by the recommendation model based on the feature vector output by each expert module, and then the recommendation order of the terms to be recommended is determined based on the score of each term to be recommended.

The invention relates to the technical field of 5G communication channel estimation, and discloses a 5GMIMO channel estimation optimization method based on deep learning. According to the method, a deep learning model architecture is constructed, and model parameters are dynamically initialized according to channel coherence time so as to adapt to 5GMIMO channel multipath propagation characteristics. Received signal streams from a plurality of antenna elements containing different orthogonal frequency division multiplexing sub-carrier frequency sequences are processed, channel impulse response estimates are generated based on frequency domain correlation and input into a model for nonlinear transformation. And performing model parameter collaborative optimization, updating the network weight by calculating the gradient variation, constructing a target function according to channel delay extension and model depth association, and adjusting the estimation output towards the direction of minimizing the mean square error. And triggering model structure adaptation based on a norm of gradient variation, acquiring model copy parameters matched with the current channel state from an adjacent cell base station, normalizing the model copy parameters, and adjusting the number of attention heads or the number of residual blocks.

The invention discloses a spherical particle size measurement method and device based on deep learning numerical prediction, and belongs to the field of particle measurement and image processing. The method comprises the following steps: acquiring interference fringe image data of spherical particles with different sizes, marking a corresponding particle size label for each image, and dividing the images into a training set, a verification set and a test set; the training set is used to train a pre-constructed neural network model, the verification set monitors the training process and stores the optimal model weight, and the test set tests the prediction precision of the model; the pre-constructed neural network model is based on an original UNet + + network and comprises a plurality of residual blocks and a space attention module, and a full-connection output head is added to realize end-to-end mapping from an image to a size value; and finally, inputting an interference fringe image of a to-be-measured particle into the trained network, and directly outputting a particle size prediction value. According to the method, high-precision, real-time and end-to-end prediction of the spherical particle size is realized, and the method is suitable for scenes such as cloud particle field on-line monitoring.

The invention discloses an interpretable dichotomy pathological image quality control method based on prototype learning, and the method comprises the steps: firstly carrying out the preprocessing of a full-width scanning pathological image, and extracting an effective region; an encoder extracts multi-level features through a convolution layer, a bottleneck block and a residual block, a prototype updating layer is embedded in a potential space, prototype vectors of focusing / out-of-focus categories are dynamically optimized, similarity vectors are generated by calculating the Euclidean distance between sample features and the prototypes, and the similarity vectors are used for calculating the sample features and the prototypes; inputting a linear classification layer output category probability; the decoder reconstructs the image through transposition convolution and jump connection, and complements details in combination with an optimization prototype; model training is combined with coding and decoding loss, classification loss and prototype loss, and finally high-precision classification is achieved. According to the method, image features are extracted through an encoder-decoder architecture, a category prototype is dynamically optimized in combination with a prototype learning mechanism, and high-precision classification and interpretability are achieved through multi-loss joint training.

This disclosure relates to a bitrate estimation apparatus and method, a video encoder, an electronic device, a storage medium, and a computer program product. The bitrate estimation apparatus includes: a parameter determination module for calculating and storing at least one bitrate estimation reference information corresponding to each residual block in an encoding unit; a bitrate estimation module for determining the input index of the last non-zero residual block in the encoding unit after storing the bitrate estimation reference information of all residual blocks in the encoding unit in the parameter determination module; and a bitrate estimation module for performing bitrate estimation on a target residual block based on at least one bitrate estimation reference information corresponding to the residual block on which bitrate estimation of the target residual block depends, wherein the input index of the target residual block is less than or equal to the input index of the last non-zero residual block. Embodiments of this disclosure can effectively reduce the amount of data that needs to be stored for bitrate estimation and reduce the computational resources required for bitrate estimation.

The invention relates to a device and a method for re-parameterizing a residual network for computational efficiency. A computer-implemented method (20) of reparameterizing a residual network (M), where the residual network (M) is a pre-trained neural network comprising residual connections skipping residual blocks, the method comprising the step of evaluating (S22) a baseline performance (P) of the residual network (M) on a first data set. When the application performance reduction (R) with respect to the baseline performance (P) is less than a given allowable reduction ([delta]), a loop (S23) is implemented, where the loop comprises the step of selecting a residual block (b belonging to M) of the residual blocks for reparameterization. A second cycle is performed through a set i < epsilon, 2 < epsilon >,..., 1: replacing all non-linear activation functions fj (x) < b with a new function fj (x) = (1-epsilon) * fj (x) + epsilon * x, and performing a retraining on M on a second data set. And re-parameterizing the residual block b into a single layer.

The application relates to a diversity image style transfer method based on a gating block, a computer device, a readable storage medium and a program product. The diversity image style transfer method is realized by using a diversity image style transfer network. The diversity image style transfer network comprises a style generation network. The style generation network comprises an encoder and a decoder which are connected in sequence. The encoder is used for inputting a content image. The decoder is used for outputting a stylized image. The decoder comprises a decoding gating block and a decoding backbone network which are connected in sequence. The decoding gating block comprises at least a first branch and a second branch which are independent of each other and share an input. The outputs of the first branch and the second branch are transmitted to the decoding backbone network. The sizes of convolution kernels of the first branch and the second branch and / or the number of residual blocks in a bottleneck layer are different. Each branch of the decoding gating block has a gating factor. The gating factor is used for adjusting the usage degree of each branch in the decoding gating block.

This invention provides a pixel-level grasping detection method, apparatus, and storage medium, relating to the field of machine vision technology. The method includes: introducing skip connections: based on a generative residual convolutional neural network (GRN), concatenating the output features of the convolutional layers and the output features of the deconvolutional layers of the GRN; introducing an attention mechanism: introducing an ECA channel attention mechanism into the residual blocks of the GRN, assigning weights to each channel of the feature map through an ECA module; improving the loss function: assigning higher weights to the region where the object is located through a weighted mask to obtain an optimized loss function; and detecting the object to be grasped based on the improved GRN. The pixel-level grasping detection method provided by this invention can more accurately capture the local geometric information and global structural relationships of an object, thereby improving the accuracy and robustness of detection.

This application provides a method, apparatus, computing device, and storage medium for kinship verification, comprising: acquiring at least two images; inputting each image into a feature extraction model to obtain output features of multiple specified residual blocks for each image and global features of individuals for each image; for any image, combining the output features of the multiple specified residual blocks to obtain a first combined feature; inputting the first combined feature into a local attention model to obtain local features of individuals; combining the local features of individuals and the global features of individuals for each image to obtain a second combined feature; inputting the second combined feature into a kinship verification model to obtain kinship verification results between individuals. By extracting significantly different local features of individuals from each image, combining them with the global features of individuals, and then inputting them into the kinship verification model, accurate kinship verification results are obtained.