273 results about "Upsampling" patented technology

The invention belongs to the technical field of image data processing, and particularly relates to a super-resolution remote sensing image reconstruction method, system and device based on frequency domain enhancement, and the method comprises the steps: S1, extracting the shallow features of a low-resolution remote sensing image; s2, inputting the shallow features into a plurality of cascaded frequencies for interactive processing, performing double-branch processing on the input features, performing inverse transformation after radial weighting on different frequency components in a frequency domain to obtain first features, and obtaining second features through depth separable convolution, an activation function, a selection scanning module and layer normalization; fusing the two branch features according to the weight, performing jump connection with the original features, performing enhancement through a feedforward network, performing repeated execution for a set number of times, and performing convolution and residual connection to obtain deep features; and S3, fusing the deep features and the shallow features, and outputting a super-resolution image through convolution and pixel rearrangement up-sampling. According to the method, texture details and edge contours of the remote sensing image can be more accurately reconstructed while the structural consistency is kept.

The invention discloses a frequency modulation and wavelet sub-band guided double-domain collaborative Transformer X-ray image denoising method, which comprises the following steps of: acquiring a noise-containing digital ray original image and a corresponding clear reference image, and constructing a data set after preprocessing the noise-containing digital ray original image and the corresponding clear reference image; constructing a network model of a double-domain collaborative coding-decoding architecture; performing 3 * 3 deep convolution on an input image to extract shallow layer features; in the encoding stage, ETB and AFMB are alternately stacked to represent local and global information, a WB-LKED module is embedded to strengthen fine-grained features, and WDB executes down-sampling and transmits high-frequency features to a decoding end; in the decoding stage, the WUB recovers the resolution through double-path up-sampling, integrates the same-scale features of an encoder, enhances details by using high-frequency features, splices the features, then carries out ETB and AFMB refining, obtains output features through 3 * 3 deep convolution, and combines a global residual error connection optimization result; and training the model by using the data set, inputting a to-be-denoised image, and outputting a final result. According to the method, the problems of weak complex noise interference resistance, poor detail retention effect and limited CNR improvement can be solved.

The invention discloses an improved YOLOv11-based steel rail defect detection algorithm and system, and aims to solve the problems of low manual inspection efficiency, high limitation of a traditional visual technology and difficulty in embedded deployment of a deep learning model. According to the scheme, YOLOv11n is used as a framework to carry out three core improvements: firstly, a C3k2-LSKA module is used to replace an original module to enhance the feature extraction capability; secondly, DySample dynamic up-sampling is adopted to replace traditional interpolation up-sampling, and small defect feature transfer is optimized; and finally, introducing windmill-shaped convolution PConv to realize sparse feature extraction. After training based on a constructed steel rail defect data set, a detection system based on PyQt5 is developed, and the detection system is successfully deployed in K230 embedded equipment integrating a high-definition camera and a touch display screen to form an acquisition-inference-display closed loop. Experiments show that the detection precision of the improved model is remarkably improved compared with that of the original YOLOv11, meanwhile, the lightweight characteristic is kept, high precision and high efficiency are effectively considered, and a reliable solution is provided for intelligent operation and maintenance of the railway track.

A deep-learning-based method for eliminating a broadband effect and a synthesized-beam effect in low-frequency Square Kilometre Array (SKA) is provided. The method includes: establishing a frequency-domain extraction module, a feature enhancement module, and a frequency-domain gating module; constructing, based on the frequency-domain extraction module, the feature enhancement module, and the frequency-domain gating module, a neural network model based on a frequency-domain self-attention mechanism; performing primary feature extraction on an input image based on the feature enhancement module to obtain a low-level feature; inputting the low-level feature into the frequency-domain gating module and performing downsampling to achieve feature encoding, so as to obtain an encoded feature; inputting the encoded feature into the frequency-domain extraction module and performing frequency-domain attention computation and upsampling to achieve feature decoding, so as to obtain a decoded feature; and inputting the decoded feature into the feature enhancement module to obtain a target restored image.

The invention belongs to the field of audio processing, and particularly discloses an audio depression detection method and system based on an improved convolutional neural network autoencoder, and the method comprises the steps: carrying out the potential feature extraction and compression of input original audio data, extracting the data into a low-dimensional and dense potential feature vector, and carrying out the recognition of the potential feature vector; and the information loss in the reconstruction process is reduced. According to the model part, a residual block structure is introduced into an encoder part, input low-layer information is reserved through jump connection, the learning ability of the network is enhanced, and the gradient disappearance problem is avoided. A transpose convolution operation is introduced into a decoder part, the spatial resolution is increased through up-sampling, and a high-quality signal reconstruction task is taken as a constraint, so that the extracted potential features are ensured to contain all key information for accurate classification. Finally, the potential features are directly used for depression classification, and experimental results show that the accuracy and robustness of depression detection can be remarkably improved, and the method has high practical application value.

The invention discloses a sound source localization and distance measurement method, device and equipment based on a microphone array and a storage medium, and relates to the technical field of acoustics, and the method comprises the steps: collecting a frequency sweeping signal played by a sound source through the microphone array, obtaining the audio data of each channel, obtaining a cross-correlation sequence through generalized cross-correlation and phase transformation weighting processing, and obtaining a distance measurement result; constructing a current to-be-processed set, processing other sequences again by taking the first sequence as a reference to obtain a new sequence group, discarding the reference sequence, taking the new group as a to-be-processed set, repeating the iteration step until only one to-be-processed sequence is left in the set, and performing up-sampling on the sequences to obtain a to-be-processed set; extracting a first main peak position to obtain a time delay estimation value of a decimal sampling level so as to determine an incident angle of a sound source relative to an array normal, performing time delay alignment on each cross-correlation sequence, constructing a single-channel beam forming signal based on an alignment result, and determining a second main peak position so as to determine a sound source distance, the efficiency of sound source localization and distance measurement is improved.

Certain aspects of the present disclosure provide techniques for upsampling input data including inputting input data at a first resolution into a machine learning (ML) model comprising a plurality of selectivity kernels, each of the plurality of selectivity kernels configured to perform a different type of selectivity to upsample the input data; and obtaining output data, corresponding to the input data, at a second resolution, from the ML model, the second resolution being higher than the first resolution, wherein the output data is based on a composite of outputs from the plurality of selectivity kernels.

The invention discloses a ring stationary signal extraction method based on a multi-scale gating attention mechanism and BiLSTM, and belongs to the technical field of mechanical fault diagnosis and signal processing. The objective of the invention is to solve the problem that weak fault features are difficult to accurately extract in a strong background noise and non-Gaussian abnormal interference environment in the prior art. Comprising the steps that a deep neural network model is constructed, local waveforms of different receptive fields are captured through parallel convolution branches, channel masks are generated through an attention mechanism, and hierarchical adaptive suppression of background noise is achieved; biLSTM is embedded in a bottleneck layer, a bidirectional memory unit of the BiLSTM is used for carrying out complete sequence time sequence modeling on compression features, and random abnormal pulses which do not conform to periodic logic are accurately eliminated through a long-term evolution rule of the sequence features; and finally, a decoder is adopted to carry out nonlinear shaping and detail repairing on the up-sampling features after transposed convolution. Gaussian white noise and non-Gaussian outliers can be efficiently filtered out, and pure ring stationary fault impact signals are reconstructed in a high-fidelity mode.

The invention provides a bridge vibration displacement detection method and system based on image enhancement and space-time tracking, and relates to the technical field of bridge detection and computer vision, and the method comprises the steps: obtaining a video image frame sequence of a bridge structure, carrying out the cooperative processing of image frames through employing an ACAE adaptive image enhancement algorithm, and obtaining an enhanced reconstructed image; inputting the reconstructed image into an AHSD-Net detection model, introducing a depth separable convolution module into a backbone network of the AHSD-Net detection model, performing spatial aggregation of preliminary features in a neck network by combining a dynamic upsampling mechanism, and finally outputting to obtain a target position detection result; based on a cascaded ST-Refine Track strategy, density denoising and position correction of a space dimension, cross-frame association of a time sequence dimension and adaptive smoothing processing of a state dimension are sequentially executed on a target position detection result, and a continuous and stable bridge vibration displacement curve is obtained. According to the invention, the displacement measurement precision and reliability are improved.

The application discloses a mixed expert network-based optical flow estimation method and system, belonging to the field of computer vision and artificial intelligence. First, the input two frames of images are preprocessed, low-resolution features are extracted through a mixed expert feature extractor (MoEE) containing a sparse activation mechanism to reduce redundant calculation; then, dot product operation is performed on the low-resolution feature maps to construct a 4D correlation volume to capture pixel motion matching relationship; subsequently, a mixed expert updater (MoEU) is used to iteratively update the hidden state and regress the residual flow increment through a dynamic expert selection mechanism to optimize the optical flow accuracy; finally, a multi-scale upsampling module is used to reconstruct the high-resolution optical flow field to output the high-precision optical flow result. The algorithm realizes dynamic resource allocation through the MoE architecture, significantly reduces the calculation cost while ensuring the accuracy of optical flow estimation, and can adapt to resource-constrained scenarios such as automatic driving and unmanned aerial vehicles, and can balance efficient inference and flexible deployment.

The invention relates to the technical field of image processing, and discloses a cross-modal fusion-based capture detection method and a readable storage medium, and the method comprises the steps: setting parallel RGB branches and depth branches in a coding stage, and carrying out the feature extraction of an RGB image and a depth image; in the RGB branches, a state space fusion module is added behind each residual block for feature fusion, and a fusion feature map with modal specificity and global relevance is generated; and the output of each down-sampling module in the RGB branch is connected to the corresponding up-sampling unit in the decoding stage in a jumping manner, and the position precision of grabbing detection and the accuracy of attitude estimation are further improved by multiplexing detail information of shallow layer features.

In one implementation, we propose a bitwise octree coding approach based on deep neural networks and operations on 3D sparse tensors. To encode / decode a certain level of detail (LoD) in an octree, geometric features are first inherited from the previous LoD by upsampling. Then based on the already encoded / decoded voxels, the point cloud geometry is firstly refined by pruning, followed by combining with the known context information. In the end, feature aggregation and probability estimation can be applied to obtain the occupancy probabilities for actual arithmetic encoding / decoding. A corresponding probabilistic training strategy is also proposed for our bitwise octree coding approach.

The invention relates to the technical field of visual quality detection, in particular to a traffic engineering quality detection method and system based on machine vision, and the method comprises the following steps: calling a vehicle-mounted camera to collect a video stream and intercept a key frame, carrying out the motion deblurring of the key frame to generate a basic image, extracting high-frequency detail features through a first convolution kernel, and carrying out the recognition of the high-frequency detail features; extracting low-frequency structural features by using down-sampling and a second convolution kernel, splicing the low-frequency features after up-sampling the low-frequency features with the high-frequency features, calculating weights based on channel response intensity and generating multi-scale features, mapping the features into a defect probability matrix and calculating a dynamic threshold, screening regions with numerical values greater than the dynamic threshold as candidate connected domains, and extracting the candidate connected domains from the candidate connected domains; and extracting geometric parameters of the connected domain, comparing the geometric parameters with a standard disease form library, and determining disease categories. According to the method, the problem that fixed threshold segmentation cannot adapt to a complex environment is solved by eliminating motion blur, fusing high and low frequency features and combining dynamic threshold judgment and morphological parameter comparison, and the traffic facility disease detection precision and efficiency are greatly improved.

The invention discloses a real-time industrial defect detection method and system based on multi-scale attention and progressive feature global shuffling, and belongs to the technical field of industrial defect detection.The method comprises the steps that data preprocessing is conducted, and industrial defect image data are collected and divided; constructing a core detection model, enhancing a multi-scale structure and an attention mechanism through features, and enhancing the feature characterization capability of the backbone network; a progressive multi-scale fusion network is designed to be combined with cross-layer interaction and a dynamic up-sampling strategy to improve feature multiplexing efficiency, an adaptive space fusion mechanism is introduced to optimize a detection head, and feature redundancy is reduced through progressive feature global shuffling; training a model by adopting an improved loss function of gradient descent optimization and geometric shape perception based on the training set; and inputting a to-be-detected image into the model and outputting a detection result. According to the method, the high reasoning speed is ensured while the positioning precision and the classification accuracy are improved, and a reliable solution with high precision and high real-time performance is provided for industrial quality control.

The invention discloses a remote sensing image super-resolution reconstruction method based on a Transform hopping network model, and relates to the field of image super-resolution reconstruction, and the method comprises the steps: extracting shallow layer features through a shallow layer convolution module in a hopping network model, extracting deep layer features through an encoder, and carrying out the super-resolution reconstruction through a deep layer convolution module; the multi-scale joint connection module enhances correlation among deep features of different scales through an attention mechanism, convolution and normalization to obtain multi-scale coding features, so that the calculation amount is reduced, and data loss is prevented; and performing context reweighting on the multi-scale decoding features by an adaptive weighting strategy in the decoder to obtain decoding features, fusing the decoding features with the up-sampling result of the low-resolution image, obtaining final fusion features through convolution operation, generating a super-resolution image based on the final fusion features, enhancing the expression of deep features, and obtaining the super-resolution image based on the super-resolution image. And high-frequency signals and clear structural features in the reconstructed image are reserved, the texture definition is improved, and the structure recovery capability is enhanced.

The invention provides a non-local information compensation Mama image deblurring method and system, and relates to the technical field of computer vision, and the method comprises the steps: carrying out the feature extraction of a degraded image containing a fuzzy feature, and obtaining an initial shallow feature; taking the initial shallow features as input parameters, performing n times of first iteration processing, and then performing deblurring processing to obtain a first recovery feature map; the first iterative processing comprises deblurring processing and down-sampling; the deblurring processing is completed through an improved Mama module and an enhanced FFN module; performing n times of second iteration processing on the first recovery feature map, and then performing deblurring processing to obtain a second recovery feature map; the second iteration processing comprises deblurring processing and up-sampling; and performing convolution processing on the second recovery feature map to generate a residual feature map, and adding the residual feature map with the degraded image to obtain a recovery image. According to the scheme provided by the invention, high-quality restoration of the blurred image is realized on the premise of effectively controlling calculation and storage overhead.

The invention discloses a dual-domain fusion panchromatic sharpening method and system based on wavelet transform and Mama, and relates to the technical field of image processing, and the method comprises the steps: employing two-stage wavelet decomposition at a frequency domain branch to obtain a low-frequency main body and high-frequency details for injection; a spatial Mama path and a spectral Mama path are arranged in a spatial domain branch, and spatial-spectral information deep fusion is realized through interaction of Mama; and in cooperation with progressive fusion up-sampling and channel attention reconstruction, a high-resolution multispectral image is finally output. The system comprises a data alignment module, a feature extraction module, a frequency domain decomposition module, a Mama interaction module, a fusion up-sampling module, a reconstruction module and the like. On QuickBird and IKONOS data sets, the method provided by the invention is superior to a comparison method in subjective vision and objective indexes such as MS-SSIM, PSNR and SAM, and is high in reasoning efficiency and friendly in deployment.

In various implementations, method and devices are disclosed that encode or decode a set of feature tensors used in Video Coding for Machine as a sequence of images as addressed in Features Coding Machine. For instance, the decoding method comprises obtaining an indication for enabling of a temporal resampling of a set of feature tensors at sequence level; obtaining an indication for estimating an upsampled set of feature tensors at picture level; and decoding the sequence of images by selectively activating / deactivating upsampling a set of feature tensors based on the indications. According to different variant, the indication for estimating an upsampled set of feature tensors at picture level may be derived or parsed from a syntax element fpps_inactive_upsampling_flag signaled in a Feature Picture Parameter Set (FPPS).

The application provides a method for solving pseudo edges and checkerboards based on an improved MBLLEN enhanced network, comprising the following steps: S1, downsampling structure; S2, downsampling structure optimization; S3, secondary downsampling; S4, upsampling structure selection; S5, upsampling structure optimization; S6, contact context feature map information; S7, secondary upsampling; and S8, contact context feature map information again. Due to the defects of the network structure of MBLLEN itself, it is inevitable that the defects of pseudo edges and checkerboard effects occur in the general image enhancement of real scenes. Through the modification of the application, on the one hand, the hard defects of the original structure are solved, and the occurrence of pseudo edges and checkerboards is avoided, on the other hand, the expression ability of the network is enhanced, and the training set exposure interval is not fixed, and good convergence can be achieved. The feature map is reduced, and the required computing power of the model is reduced. The context information is contacted, and the information loss of the simple linear structure is avoided.

This invention relates to an image denoising method, system, and readable storage medium. The method includes: performing Gaussian filtering and downsampling on a current frame image to obtain a first downsampled image; performing Gaussian pyramid decomposition to obtain a current frame image group; upsampling to obtain an upsampled image, and subtracting it from the current frame image to obtain high-frequency information; performing motion estimation and texture estimation at different scales on the current frame image group and a reference frame image group, and correcting them with guided filtering; performing spatiotemporal filtering on the first downsampled image and the reference frame image based on motion estimation weights; upsampling the denoised downsampled image to obtain a denoised upsampled image; and weighted fusing the denoised upsampled image and high-frequency information according to texture estimation weights to obtain a final image; downsampling the final image and performing Gaussian pyramid decomposition to obtain a reference frame image group for denoising the next frame image. This invention can preserve more texture details while increasing the image signal-to-noise ratio.

The invention provides an image super-resolution method based on a discrete soft attention feature conversion network, relates to the technical field of image processing, and constructs an improved discrete soft attention feature conversion network based on an SCET network, which comprises a convolution layer, an up-sampling convolution module and a backbone network, the backbone network comprises N discrete feature fusion modules, an STA module and a convolution up-sampling module which are connected in sequence, and each discrete feature fusion module comprises a convolution layer for reducing the dimension of an input channel and a DFEModule structure. The problem of long-distance dependency relationship modeling can be solved; the problem of information loss in network propagation caused by the fact that all feature information cannot be fully utilized is solved, the problems that the design of a multi-head attention module is too simple, the perceptual range of attention operation is set too general and the effect of the model is reduced can be solved, and the disorder of reconstruction logic of local details can be solved. And the problem of high efficiency in high-resolution super-resolution is solved.

This invention discloses a non-contact, high-throughput multimodal perception method and system for fish phenotypic recognition. First, underwater optical visual images and multibeam forward-looking sonar images are acquired, ensuring that each frame of the acoustic image corresponds to and is aligned with the visual image. Then, both images are downsampled twice and input into a feature network to extract acoustic features (V, V1, V2) and visual features (S, S1, S2). Next, V1 is upsampled and concatenated with V, and S1 is upsampled and concatenated with S, then fed into a first alignment enhancement network to obtain A1 and A2, respectively. Similarly, V2 is concatenated with V1, and S2 with S1 and fed into a second network to obtain B1 and B2; V2 and S2 are fed into a third network to obtain C1 and C2. Finally, A1, A2, B1, B2, and C1, C2 are fused through attention to obtain X, Y, and Z, respectively. X, Y, and Z are input into a detection model for detection, outputting the target bounding box and key points of the fish body. The target length of the fish body is calculated using camera intrinsic parameters. This invention constructs a highly fault-tolerant and practical deep learning fusion computing method.

We provide video coding, encoders, and decoders that further improve efficiency based on trained networks. [Solution] To handle lumer-chroma channels of different sizes, the chroma component is upsampled so that the resulting upsampled chroma component has a resolution matching one of the lumer components. The lumer and upsampled chroma components are then encoded into a bitstream. To reconstruct the picture portion, the lumer component and an intermediate chroma component matching the resolution of the lumer component are decoded from the bitstream, and then the intermediate chroma component is downsampled. The subsampled chroma format is handled by an autoencoder / autodecoder framework while preserving the lumer channel.

The invention discloses a single-stage speech synthesis method and system based on VITS2, and relates to the technical field of speech synthesis, and the method comprises the steps: obtaining a to-be-synthesized speech, and extracting a corresponding audio feature; the audio features are input to a multi-scale feature fusion module for multi-scale feature extraction, invalid region filtering and fusion, and optimized features are obtained; inputting to a decoding generation module based on the optimization features, and carrying out multiple times of up-sampling to obtain an intermediate audio; inputting the intermediate audio into a feature extraction module to obtain a fundamental frequency signal and a Mel spectrum; and inputting the fundamental frequency signal and the Mel spectrum into a U-Net post-processing module, and performing up-down sampling and residual fusion to obtain a synthesized audio signal spectrum. The naturalness and fidelity of the synthesized speech are remarkably improved, and meanwhile the training and reasoning efficiency is guaranteed.

This invention discloses a lightweight lesion image segmentation method based on grouped context fusion, specifically relating to the fields of image processing and artificial intelligence. The input image is processed by a convolutional initial layer to extract preliminary features. The segmentation method relies on the MoMNet network, which adopts an encoder-decoder paradigm. The encoder uses the MoM module to extract hierarchical features in all four stages. The convolutional initial layer outputs the extracted preliminary features and passes them to the MoM module, which includes downsampling operations in four stages. The decoder fuses multi-scale information through an upsampling module and skip connections. At the same time, multiple prediction heads are jointly optimized by binary cross-entropy and Dice loss to achieve progressive learning for the image segmentation task.

Embodiments of the present application disclose a video decoding method, a video encoding method, a device, computer equipment and a storage medium. The video decoding method comprises: decoding encoded information of at least one block from an encoded video bitstream, the encoded information indicating whether a super-resolution encoding mode is applied to the at least one block, wherein the super-resolution encoding mode is applied in response to the at least one block being down-sampled from a high spatial resolution to a low spatial resolution by an encoder; and when the encoded information indicates that the super-resolution encoding mode is applied to the at least one block, generating a reconstructed block by using the super-resolution encoding mode to up-sample information of a first block in the at least one block, wherein the first block has the low spatial resolution, the reconstructed block has the high spatial resolution which is higher than the low spatial resolution, the at least one block comprises transform coefficients, and the reconstructed block comprises sample values in a spatial domain.

A method for point cloud segmentation is disclosed. The method may comprise performing a down-sampling and feature transformation step on an embedding feature extracted from point cloud data to produce a representation of the point cloud data, wherein performing the down-sampling and feature transformation step comprises performing a down-sampling sub-step for reducing a number of points by a pooling layer and performing one or more times of a feature transformation sub-step; and performing an up-sampling and feature transformation step on the representation of the point cloud data to produce the point cloud segmentation, wherein performing the up-sampling and feature transformation step comprises performing an up-sampling sub-step for restoring the number of points by an un-pooling layer and performing one or more times of the feature transformation sub-step. The feature transformation sub-step comprises at least one of capturing a local feature of the point cloud data by a local perceiver; capturing a global feature of the point cloud data by a selective state space model (SSM) block; and capturing a cross-channel dependency of the point cloud data by a channel modulator.

A wireless communication system provides a baseband signal processing circuit configured to sample a baseband signal at twice a baseband frequency to produce an upsampled signal that provides a sequence of samples of the baseband signal at twice the baseband frequency; and combine each sample within the upsampled signal with a preceding sample within the upsampled signal to produce a modified baseband signal that provides a sequence of modified symbols at twice the baseband frequency; a modulator circuit coupled to modulate frequency of a carrier signal based upon symbol values of symbols within the modified baseband signal to produce a frequency modulated carrier signal; and an antenna coupled to transmit the modulated carrier signal.

A system and method of processing Automotive Ethernet and High-Speed Ethernet signals. The method includes forming a first data path according to a first transmission protocol and a second data path according to a second transmission protocol. The method includes selecting the first data path to receive a first bitstream compliant with the first transmission protocol at a first data rate. The method includes encoding the first bitstream using a first encoding scheme to produce a first encoded bitstream. The method includes upsampling the first encoded bitstream to produce an upsampled bitstream at a different data rate. The upsampled bitstream aligns with transmission requirements of a Physical Medium Attachment (PMA) device. The method includes forwarding the upsampled bitstream to the PMA device.

The application provides a lightweight variable bit rate multi-view image compression method and system, and relates to the technical field of image compression. The method comprises: downsampling, feature extraction and feature scaling of a single-view image to obtain a latent representation corresponding to a target bit rate; quantization and lossless entropy coding of the latent representation to obtain a final compressed bit stream; subsequent lossless entropy decoding and inverse scaling to restore the latent representation; feature fusion and upsampling of the restored latent representations of different views to generate a reconstructed compressed image. The model comprises a main encoder, a feature scaling module, a quantization module, an autoregressive entropy model, an arithmetic encoder, an arithmetic decoder, a feature inverse scaling module and a decoder. The application can efficiently compress image data, reduce computational complexity and storage space occupation while preserving image details and quality, thereby providing faster speed and lower bandwidth requirement for image transmission and storage.