31 results about "Haar wavelet" patented technology

The invention relates to an epilepsy prediction method based on adaptive sparse attention and a hierarchical graph convolution network, and the method comprises the steps: carrying out the time domain convolution, spectrum transformation and Haar wavelet down-sampling of an electroencephalogram signal, respectively generating time domain, spectral domain and fidelity down-sampling features, and fusing the features into a low-level feature set; on the basis of a sparse attention mechanism, constructing and applying a multi-level sparse mask to adaptively screen and weight-aggregate key discriminative features in the feature set to obtain screened features; on the basis of the feature, by constructing a local channel graph and a global frequency band graph and respectively executing graph convolution, capturing local spatial correlation of each channel in a single frequency band and global cross-frequency-band spatial dependence among different frequency bands, and fusing the local spatial correlation and the global cross-frequency-band spatial dependence into an embedded feature; and inputting the embedded features into a classifier to obtain a state probability, and triggering an alarm based on the state probability. Therefore, the problems of key information loss, insufficient time-space spectrum dependent modeling and feature redundancy are solved, and the accuracy, stability and real-time performance of epilepsy prediction are improved.

The invention provides a high-fidelity anti-compression image watermarking method and system based on spectrum-airspace decoupling, and belongs to the field of information security. Firstly, the watermark information is mapped and remodeled; a watermark encoder based on multi-granularity spectrum-spatial domain feature decoupling is constructed, discrete cosine transform is introduced to filter out high-frequency components, Haar wavelet transform is adopted to realize lossless downsampling, a fast Fourier transform dynamic filter is combined to capture global semantic features, and local texture details are combined through multi-scale spatial domain volume accumulation; designing a physical perception and visual self-adaptive dual embedding strategy, and anchoring watermark energy to an anti-compression brightness channel; constructing an anti-attack layer containing differentiable JPEG compression simulation and mixed noise simulation, and participating in network training; and constructing a decoder and designing a loss function to carry out network optimization. According to the method, the robustness of the watermark under strong compression and complex black box attacks is improved, and extremely high visual imperceptibility is realized through physical and visual constraints.

The invention discloses an image defogging method based on a wavelet domain diffusion model, which is used for solving the problems of overlong reasoning time, large calculation amount, unstable restoration effect and the like in the prior art. The method comprises the following steps: constructing a training set and a test set; performing wavelet domain feature enhancement processing on all foggy weather images and clear images in the training set to obtain corresponding second-level components and second-level wavelet domain enhancement features; loading random Gaussian noise to a second-level low-frequency approximate component in the second-level component of each clear image, inputting the second-level low-frequency approximate component and the second-level low-frequency approximate component of the corresponding foggy day image into a noise prediction network at the same time, and outputting prediction noise; de-noising the second-level low-frequency approximate component of each foggy day image based on the predicted noise, and then carrying out inverse discrete Haar wavelet transform processing to obtain second-level recovery features of the second-level low-frequency approximate component; training the noise prediction network by using a loss function to obtain a wavelet domain diffusion model; a defogged image can be obtained based on the model.

The invention provides a feature fusion method, system and device based on a frequency domain and a storage medium, and the method comprises the following steps: 1, separating low-frequency features and high-frequency features: carrying out the separation of the low-frequency features representing a global structure and the high-frequency features representing detail textures through a Haar wavelet decomposition guidance feature map; 2, generating a low-frequency weight: processing the low-frequency features, and generating the low-frequency weight which can be used for modulating a target feature map; 3, generating a high-frequency weight: processing the high-frequency feature, and generating the high-frequency weight adaptive to the spatial size of the target feature map; and 4, feature fusion: fusing the low-frequency weight, the high-frequency weight and the target feature map, and outputting the fused enhanced features. According to the method, global and detail information separation of cross-scale features is realized through Haar wavelet decomposition, and adaptive fusion is realized in combination with double-branch weight learning, so that the problems of insufficient feature fusion and poor generation quality of the generative adversarial network in image generation are solved.

The invention discloses a multi-scale parallax calculation module, method and device based on Haar wavelet transform and a medium, and the method comprises the steps: firstly constructing a feature pyramid through the Haar wavelet transform, extracting high-frequency and low-frequency details, and then carrying out the parallax calculation through employing a Coarse-to-Fine strategy. According to the method, the problem that random initialization is slow in convergence in a weak texture area or falls into local optimum can be effectively solved, and meanwhile, the wavelet detail component can provide richer texture and edge information for cost calculation. The method has the advantages of being high in precision, high in robustness and high in speed in dealing with parallax calculation tasks of weak-texture, non-texture or repeated-texture scenes.

The invention belongs to the technical field of image processing, and particularly relates to a traffic scene image defogging method based on bidirectional Haar wavelet transform and AOD-Net, comprising the following steps: constructing a data set; performing frequency domain decomposition on the image by using Haar wavelet transform, separating low-frequency and high-frequency features, and realizing display decoupling of structure and fog information and texture and detail information; a multi-head self-attention mechanism is introduced, a low-frequency self-attention enhancement module and a high-frequency self-attention enhancement module are constructed respectively, and global structure modeling and detail feature self-adaptive enhancement are achieved; according to the method, on the premise that the color naturalness and the structure consistency of the restored image are kept, the common problems of over-enhancement, artifact generation, edge breakage and the like in a traditional defogging network are effectively reduced, and the overall visual quality and stability are remarkably improved.

The invention relates to a river flow prediction method, which comprises the following steps of: 1, dividing a daily runoff data set, and performing multi-stage decomposition on a training set by using Haar wavelets to obtain de-noising parameters; 2, multiplexing the verification set and the test set to obtain data; step 3, carrying out first-order differential transformation, and zooming to an interval of [0, 1] by using MinMaxScaler to obtain a normalized value of a differential sequence; 4, mapping the data to a high-dimensional space through a linear embedding layer, and then carrying out position coding to obtain data; step 5, inputting the data obtained in the step 4 into a DTransformer encoder, so that a catastrophe point obtains a higher attention weight so as to strengthen difference characteristics of adjacent time steps and obtain encoder output; 6, inputting the data obtained in the step 5 into an LSTM decoder, and performing feature recombination to obtain encoder output; 7, mapping the data obtained in the step 6 to 7-dimensional output through a full connection layer; and 8, performing inverse normalization and inverse difference transformation on the data obtained in the step 7 to obtain final output.

The invention discloses a dispensing robot motion control system based on template matching, and belongs to the technical field of industrial automation and robot control. According to the invention, the problems of inaccurate positioning, wrong and leaked glue and track optimization existing in the existing micro object glue dispensing are solved. The method specifically comprises the following steps: preprocessing a collected image, carrying out denoising and feature enhancement on the collected image by adopting an improved bilateral filtering algorithm in combination with Haar wavelet transform, and then realizing accurate positioning by fusing color matching, gray mode matching and an improved dung beetle optimization algorithm; and finally, through D-H parametric method modeling and forward and inverse kinematics analysis of the robot, the motion trail of the robot is optimized. According to the invention, the dispensing positioning precision and matching efficiency can be effectively improved, the phenomena of glue mistake and glue leakage are reduced, and the method can be applied to high-precision dispensing operation of micro objects.

This invention discloses an extended attention visual processing method for unmanned power grid inspection, comprising: acquiring and preprocessing UAV inspection images; inputting the preprocessed images into an encoder-decoder segmentation network to extract shallow detail features and deep semantic features; in the shallow encoding stage, using a hybrid frequency domain downsampling module to extract high-frequency edge information using Haar wavelet transform and fusing it with spatial domain structural features; in the deep encoding stage, introducing an extended alternating perception mechanism to model local detail associations and global topological dependencies; in the decoding stage, using a feature reconstruction module to complete upsampling, feature fusion, and context refinement to progressively restore the spatial structure of power lines; training the network using a depth-alignment-connectivity loss function and outputting a power line segmentation mask; this method can reduce power line segmentation breaks and missed detections in complex backgrounds, and improve the continuous segmentation capability and structural integrity of weak targets.

The application discloses a return scattering ionization map F area automatic segmentation method based on Haar wavelet downsampling, to solve the problems of fuzzy F area boundary, low contrast and high frequency information loss in the prior art. The method comprises the following steps: step 1, data acquisition, labeling and pretreatment; step 2, constructing a SegNext segmentation network based on Haar wavelet downsampling; step 3, decoding fusion and segmentation prediction output; step 4, training optimization based on edge weighted focus joint loss. The method retains the high frequency characteristics of the ionization map by using the wavelet downsampling module, and combines the focus loss and the edge weighting mechanism, so that the segmentation result is more accurate and stable in the boundary area.

The present application relates to the technical field of safety monitoring and intelligent management of rail transit systems, and specifically discloses a subway emergency management and intelligent command system and method, which uses temperature sensors and vibration sensors to respectively collect temperature data of electrical connection points and track vibration data; based on Haar wavelet transform, temperature abnormal fluctuation characteristic values are calculated to identify abnormal states of the electrical connection points, while fast Fourier transform is combined with mean clustering analysis to extract track vibration amplitude and frequency characteristics, a comprehensive characteristic vector is constructed as input, a random forest model is used for risk prediction, according to the risk score output by the model, the future running state of the track is divided into three levels of high risk, general risk and low risk, a detailed inspection and maintenance plan is automatically triggered for the high-risk area, and the train scheduling strategy is optimized to ensure operation safety, key information is displayed through data visualization technology, and the response speed and accuracy of decision makers are improved.

This invention provides a lightweight intrusion detection method, system, and storage medium based on frequency domain gating and dynamic head attention distillation. The method includes: Step 1: Constructing a teacher model, which includes a frequency domain wavelet analysis module and a dynamic head interaction attention module. The frequency domain wavelet analysis module performs multi-scale decomposition and information enhancement of network traffic features in the frequency domain through a two-level Haar wavelet transform and adaptive gating mechanism. The dynamic head interaction attention module treats the feature dimension as a token sequence and models the global dependencies between features by introducing position-aware embedding and multi-head self-attention mechanisms; Step 2: Designing a dynamic entropy adaptive topology distillation strategy, enabling the lightweight student model to inherit the discriminative features and sample relationship structure of the teacher model while compressing the feature space; Step 3: Transferring the knowledge of the teacher model to the lightweight student network. The beneficial effect of this invention is that it achieves more efficient and more complete knowledge transfer.

The invention discloses a power transmission line tiny target detection method based on frequency sensing feature fusion. The method specifically comprises the following steps: extracting a multi-scale feature map through a ResNet50 backbone network; constructing a feature pyramid, and decomposing features into low-frequency structure components and high-frequency detail components by using Haar wavelet transform; processing low frequency components using depth separable convolution and grouping MLP for capturing remote dependencies and helping similar feature aggregation; for a high-frequency component, reinforcing boundary details of a tiny target by designing a self-adaptive high-frequency filter; a channel grouping association strategy is introduced, and self-organizing optimization of features is carried out; flattening the feature map into one-dimensional features, constructing a multi-branch decoder architecture to realize cooperative training, and relieving the deficiency of positive sample query through dense prediction branches; and sending the final output features into a classifier to obtain a category and a position frame. According to the method, through a high and low frequency division strategy, the feature response of the small target is enhanced, and inter-channel feature self-organization optimization is adopted, so that the detector can more fully learn the discrimination feature of the small target, and the detection precision under the complex background of the power transmission line is improved.

The invention discloses an online LiDAR moving target segmentation method based on Haar wavelet enhancement, and belongs to the technical field of target segmentation. The problem that an existing moving target online segmentation method is low in precision is solved. According to the method, a residual multi-layer perceptron structure is adopted, the residual multi-layer perceptron structure can reduce information loss caused by depth reduction by introducing a residual mechanism, the mapping quality from the point cloud to the BEV feature map is improved, and spatial information expression is enhanced. By designing the wavelet fusion block, multi-scale feature modeling can be enhanced based on Haar wavelet transform, information expression of a BEV feature map in the horizontal direction, the vertical direction and the diagonal direction is enhanced, and the capturing capacity of a moving target is improved. The optimized coding-decoding architecture can reduce the calculation overhead while improving the segmentation precision, and meets the real-time requirements of online tasks. Through point cloud feature extraction, feature fusion modeling and segmentation network design, segmentation precision and real-time performance are improved. The method can be applied to moving target segmentation.

The application discloses a crack image segmentation method based on double encoders in a complex environment, which improves the overall framework of the double-path encoder-decoder structure in the prior art CN202310525413.0, designs a new Transformer transformer block structure to effectively extract the global semantic information of the image, uses the multi-window high-low frequency mechanism of the Haar wavelet transform to extract the high-frequency features and low-frequency features of the image, and enhances the local information perception ability and the interaction between the image blocks, so as to solve the problems that the existing crack detection algorithm is difficult to adapt to a complex environment and has a large amount of calculation. In addition, a new feature fusion module is designed in the application to better fuse the intermediate features of the two encoders. The method of the application can accurately realize crack detection in various complex environments and reduce the amount of calculation, and improves the environmental versatility of crack detection.

The application discloses an image defogging method based on a wavelet field diffusion model, and is used for solving the problems of long reasoning time, large calculation amount and unstable recovery effect in the prior art. The method comprises the following steps: constructing a training set and a test set; performing wavelet field feature enhancement processing on all foggy images and clear images in the training set respectively, so as to obtain corresponding secondary components and secondary wavelet field enhanced features; loading random Gaussian noise on the secondary low-frequency approximate component in the secondary component of each clear image, and then inputting the secondary low-frequency approximate component of the corresponding foggy image and the secondary low-frequency approximate component of each clear image into a noise prediction network at the same time and outputting predicted noise; denoising the secondary low-frequency approximate component of each foggy image based on the predicted noise, and then performing inverse discrete Haar wavelet transform processing to obtain secondary recovery features of the foggy image; training a wavelet field diffusion model by using a loss function on the noise prediction network; and obtaining a defogged image based on the model.

The application provides an air target adaptive segmented track recognition method and system based on a Softmax multi-classification network, which comprises the following steps: traversing a whole complex track sequence in time through a sliding window, fitting the track in the window by using a designed fitting equation to obtain equation parameters; training a Softmax classification network model based on the equation parameters, and recognizing the track pattern in the window based on the network model; and correcting the recognition result based on Haar wavelet reconstruction to eliminate discrete wrong recognition results. The application solves the problem of poor complex track pattern recognition effect and can effectively recognize the complex track pattern in an adaptive segmented manner.

The invention discloses a heart valve segmentation method based on multi-semantic feature adaptive enhancement, and the method comprises the steps: firstly transmitting a heart valve CT image into the first three shallow semantic enhancement modules of an encoder; sequentially passing through a shallow semantic feature generator based on learnable Sobel and Haar wavelet transform, loop gating attention and mixed expert attention in the modules, and then transmitting shallow semantic coding features into the rear two deep feature enhancement modules; a convolutional network layer and a dynamic association perception layer based on a dynamic association perception attention and cavity multi-core gating feedforward layer are respectively arranged in the module; the first three shallow semantic coding features and the last two deep semantic coding features are transmitted into a corresponding edge auxiliary decoding module and a mask auxiliary decoding module in an auxiliary decoder, and loss constraint is carried out by utilizing an edge mask and a valve mask respectively; and finally, transmitting the encoding features of the encoder and the auxiliary features of the auxiliary decoder into corresponding mixed semantic decoding modules in the decoder for decoding, and generating a valve segmentation result through the last layer of decoding module. According to the invention, a multi-semantic feature adaptive enhancement network architecture is designed, and the architecture can effectively capture shallow multi-semantic information and dynamic association information of deep features of the valve image, so that coding features with higher robustness are formed, and the segmentation performance is effectively improved.

The invention provides a multi-modal vehicle target identification method based on a complex environment, and the method comprises the following steps: obtaining a vehicle data set which is synchronously collected in the complex environment and comprises point cloud data and image data, and converting the point cloud data into three-dimensional voxel representation through voxelization processing; inputting the image data into an image feature extractor to obtain image features; carrying out voxel feature coding on the three-dimensional voxel representation to obtain voxel features; inputting the image features and the voxel features into a dynamic attention module which dynamically allocates weights according to the importance of the two modal features and splices the weighted features to generate fusion features; the fusion features are subjected to depth separable convolution and a Haar wavelet self-attention mechanism in sequence, and an identification result of the to-be-detected vehicle picture is obtained; according to the invention, the accuracy and real-time performance of target detection under severe weather conditions are improved.

The invention relates to the technical field of smart power grids, and particularly discloses a power grid marketing side system platform data processing and analyzing method based on edge computing, which collects power consumption and power factor data of a user in real time through a smart electric meter, and performs efficient processing by using an edge computing node close to a data source. Analyzing the power consumption change rate by adopting Haar wavelet transform, analyzing the power factor fluctuation degree by adopting fast Fourier transform, constructing a power consumption change rate characteristic value and a power factor characteristic value so as to evaluate the power utilization stability of a power grid and the stability of a power system, and integrating the characteristic values into a comprehensive characteristic vector so as to evaluate the power utilization stability of the power grid and the stability of the power system. And as the input of the support vector machine model, the power grid marketing response speed is evaluated, and a power grid operator can dynamically adjust a marketing strategy according to an evaluation result.

This invention relates to the field of power line inspection technology, specifically disclosing a power line inspection method and an automated unmanned aerial vehicle (UAV) airport vehicle. By real-time monitoring of the spectral efficiency and channel state information feedback delay time of multiple UAVs, abnormal spectral efficiency characteristic values ​​and feedback delay characteristic values ​​are calculated to assess the stability of the communication link and the risk of channel prediction errors. These characteristics are then used to construct a multi-dimensional congestion warning vector, which is input into a pre-trained deep reinforcement learning optimization decision model. The model outputs optimal frequency band allocation, time slot scheduling, and power control strategies, achieving adaptive optimization of the concurrent communication performance of multiple UAVs. This invention combines Haar wavelet transform, random matrix theory, and artificial intelligence technology to improve the system's robustness, resource utilization efficiency, and quality of service assurance capabilities in complex dynamic environments, ensuring the stability of UAV communication and thus accurately completing inspection tasks.

This invention discloses a frequency-domain illumination prior-driven unsupervised low-light video enhancement method, system, and storage medium for monitoring tunnel boring machine (TBM) slag outlets. The method utilizes a noise suppression module to achieve composite noise suppression and basic restoration of the low-light image. A frequency-domain illumination prior estimation module performs two-dimensional Haar wavelet decomposition on the restored image, combining channel-spatial attention-weighted high and low frequency sub-bands and fusing them to generate frequency-domain illumination prior features. The reflection component is accurately reconstructed using a cross-attention transformer unit in the reflection component reconstruction module. A global exponential enhancement factor is generated using an illumination enhancement coefficient module to obtain the enhanced image. Unsupervised end-to-end training of the network is achieved through gamma transform branching and multi-dimensional loss constraints. This method effectively alleviates the problems of overall darkness, detail obscuration, and severe noise in TBM slag outlet monitoring images without increasing the burden of continuous on-site lighting, resulting in enhanced videos with balanced brightness, clear details, and controlled overexposure.

This invention discloses a target detection method for remote sensing images based on an improved lightweight YOLOv4, belonging to the field of target detection technology for UAV aerial remote sensing images. The method includes the following steps: converting the format of the remote sensing dataset; dividing the converted dataset proportionally; improving the backbone network by performing Haar wavelet transform on the input feature maps of the low and high layers of the backbone network, fusing high-frequency features through convolution using the Ghost module, and adding the fused features to the features extracted from the feature maps by ordinary convolution; improving the Neck by introducing coordinate attention on the two branches before the FPN and after the last residual structure of the improved backbone network; setting training parameters, obtaining and saving weight data; selecting the model with the smallest loss function on the validation set, loading the training weights into the network, and testing with a test set. This invention can improve the target detection accuracy and real-time performance of UAV aerial remote sensing images.

The invention discloses an infrared small target detection method based on wavelet transform and manifold learning, and the method comprises the steps: decomposing an input infrared image into a low-frequency component and a high-frequency component through Haar wavelet transform (HDWT), and designing a special processing module for different components: carrying out the cross-sub-band feature fusion and U-shaped convolutional network learning of the low-frequency component through a low-frequency encoder, and carrying out the feature fusion of the high-frequency component; background structure modeling and target-background distinguishing are enhanced; for a high-frequency component, through non-local feature extraction and parameterized manifold learning, noise interference is suppressed while a target detail topological structure is kept; meanwhile, an attention mechanism is designed, the coding layer space saliency and the decoding layer channel saliency are fused, and the attention on a small target area is improved. According to the method, the frequency domain characteristics are combined, rich and highly distinguished characteristics are jointly extracted, the detection precision is remarkably improved, and the false alarm rate is reduced.

The application discloses a kind of defect detection methods for high-density packaging chip, it is related to semiconductor detection technical field, including, to high-density packaging chip is pretreated;Utilize multispectral imaging equipment, the high-density packaging chip after pretreatment is imaged, obtains multispectral imaging image group;Sparse representation multispectral imaging image group is obtained by adopting the technology of multi-stage decomposition based on Haar wavelet discrete wavelet transform first, then thresholding processing;Compressed sensing is carried out to sparse representation multispectral imaging image group based on compressed sensing technology, and the multispectral imaging image group after compressed sensing is obtained;The multispectral imaging image group after compressed sensing is labeled, and the visual Transformer model of RGB input is transformed into the visual Transformer model of multi-band input, and the visual Transformer model of multi-channel input is trained.The application improves the accuracy of high-density packaging chip defect detection by multispectral imaging, sparse representation and compressed sensing technology.

The application discloses a kind of weak semi-supervised cardiac MRI segmentation methods based on pseudo-label frequency domain dynamic hybrid supervision, belong to medical image segmentation technical field, including the following steps: S1, a small amount of cardiac MRI image in training set is provided with scribble label;S2, establish main segmentation model and auxiliary segmentation model, output corresponding predicted value;S3, generate supervision pseudo-label using Haar wavelet transform and frequency domain dynamic hybrid method;S4, combined with partial cross-entropy loss constraint strategy establishes total loss function to optimize training three segmentation models, realize the segmentation of cardiac MRI image.The application generates reliable supervision signal by pseudo-label frequency domain dynamic hybrid supervision method, significantly improves the utilization efficiency of unmarked data using a small amount of scribble label and a large amount of unlabelled data;Using weak semi-supervised strategy, only a small amount of scribble label can realize high-quality segmentation, reduce the dependence on pixel-by-pixel accurate annotation, solve the problem of high medical image annotation cost.

This invention relates to the field of image segmentation technology and discloses a method for segmenting uveitis lesions based on a U-shaped encoder-decoder network. First, a multi-branch feature fusion module is constructed as a basic unit, stacked to form a multi-scale feature enhancement encoder. Second, a frequency domain cross-layer feature fusion module is designed in the skip connections, utilizing Haar wavelet decomposition to extract boundary information and combining spatial and channel attention to filter and fuse features. A hybrid Mamba interaction bottleneck module is set at the top of the encoder to align and aggregate multi-level deep features, obtain global context through visual Mamba, and perform mutual correction of global and local features using a joint attention map. In the decoding stage, an uncertainty-aware recursive gating module is introduced, using the Shannon entropy output of the prediction head as the uncertainty map to dynamically modulate the initial gating generated by the gating network, achieving adaptive control of the intensity of shallow detail introduction.

The application discloses a Hadamard matrix multi-scale sequencing method and system, wherein the method comprises the following steps: initializing parameters, and constructing an M-row and M-column Hadamard matrix; extracting each row or column vector of the Hadamard matrix and then performing normalization, rearranging the elements of each row or column vector to form a 2 n Row 2 n Column square matrix H i ; performing n-layer two-dimensional Haar wavelet decomposition on the square matrix H i , obtaining a coefficient matrix H a , taking the absolute value |H a | of the coefficient matrix H a , summing up |H a |, and recording i into a sequence number value I(i); storing the values of Sa(i) and I(i) into a vector Sa and a vector I respectively; taking i=i+1, and judging whether i is greater than M; arranging the elements in the vector Sa in ascending order to obtain a new sequence R; extracting each row or column vector of the Hadamard matrix after normalization according to the new sequence R, and traversing M times to obtain M two-dimensional 2 n ×2 n square matrices RH i . The application solves the problem that the existing coding cannot simultaneously consider fast imaging and high signal-to-noise ratio imaging.