116 results about "Haar wavelet" patented technology

The invention relates to the technical field of medical image processing, and provides a medical image segmentation method based on wavelet enhancement and multi-scale feature fusion. According to the method, a CNN-Transform double-branch coding structure is combined, a multi-scale wavelet fusion module is provided, from the perspective of a frequency domain, Haar wavelet transform is adopted to extract an image high-frequency sub-band so as to enhance edge and texture detail expression, dynamic weighting is performed on different frequency band features through grouping convolution and a sub-band attention mechanism, and the discrimination capability is improved; meanwhile, a multi-scale cavity pyramid structure is fused in a spatial domain, and after cross attention dynamic fusion is introduced, a feature alignment mechanism of a wavelet domain and the spatial domain is established; and collaborative fusion of frequency domain and space domain features is realized. The method effectively improves the segmentation precision of the fuzzy boundary and the fine-grained structure under the complex background, has good universality and adaptability, and is suitable for various medical image segmentation tasks.

The invention discloses a lightweight super-resolution system and method of an adaptive wavelet attention network, and belongs to the technical field of image processing. The system is composed of a multistage wavelet attention module, a dynamic convolution kernel generation unit, a convolutional neural network and an output unit. The method comprises the following steps of: extracting features of a low-resolution image and performing Haar wavelet decomposition; calculating a cross-scale attention weight on each high-frequency sub-band and carrying out weighted fusion to highlight details; adaptively generating a dynamic convolution kernel of a Haar wavelet basis kernel weighted combination based on the fused features, and performing directional convolution enhancement on the features; high-resolution image reconstruction is realized through a lightweight residual network and pixel rearrangement; and during training, pixel domain mean square error and wavelet coefficient compensation loss joint optimization is adopted. The parameter quantity of the system model is smaller than 450KB, a 1080p video super-resolution task can be processed on mobile equipment in real time, and the texture recovery performance is improved by about 1.2 dB compared with that of an existing lightweight model.

The invention discloses an image enhancement method and system based on detail sensitivity and noise suppression fusion, and relates to the technical field of image processing, and the method comprises the steps: firstly processing an input unmarked microscopic cell image through a three-branch network containing detail sensitivity enhancement, a multi-scale Haar wavelet down-sampling module and loss function constraint; the network extracts multi-scale gradient features and separates illumination, reflection and noise components in a frequency domain. And then noise is subtracted from the original image, illumination influence is removed, and a noiseless reflection image is obtained. And then, Gamma correction is adopted to optimize the illumination component, and the Retinex method and the reflection image are fused to enhance the image quality. And finally, constraining the result in combination with the detail sensitive loss and the noise suppression loss to obtain high-quality output. According to the method, the three-branch image decomposition network is combined with detail enhancement, frequency domain component separation and loss constraint optimization technologies, so that high-quality enhancement and noise suppression of the unmarked microscopic cell image are realized.

The invention relates to the technical field of digital image security and authentic identification, and provides an image tampering detection method and system based on mixed features and RGB features, and the method comprises the steps: obtaining a to-be-detected input image, and carrying out the preprocessing of the to-be-detected input image; respectively extracting a Haar wavelet high-frequency component, a discrete cosine transform frequency domain feature and a Bayer convolution noise feature, and carrying out matrix level fusion to obtain a mixed feature; extracting RGB (Red, Green and Blue) features for the preprocessed input image; the mixed features are connected through cross-layer residual errors, and mixed feature learning features are obtained; and integrating the mixed feature learning features and the fused RGB features by using a cross-modal feature interaction architecture to obtain a prediction probability graph. Multi-modal features are fused, high-frequency response is enhanced, and the accuracy of image tampering detection is improved by adopting a dynamic fusion mechanism. The technical problems that an existing tampering detection method is insufficient in feature characterization capacity in a complex scene, low in tampering trace detection sensitivity and the like are solved.

The invention provides an infrared and visible light image end-to-end registration method based on phase consistency enhancement, and belongs to the technical field of electric digital data processing. The invention discloses an infrared and visible light image end-to-end registration method based on phase consistency enhancement, which comprises the following steps of: firstly, preprocessing infrared and visible light source images in an image database, and dividing the infrared and visible light source images into a training set and a test set; and constructing a feature extractor based on a phase consistency attention mask, realizing cross-modal consistency feature extraction, and focusing an important space region by using phase information. A self-defined ResSobeNet network is adopted for parameter estimation, and Haar wavelet transform is used for replacing pooling operation, so that information loss in a traditional down-sampling method is avoided. A double-branch generator structure is designed, affine parameters and an optical flow field are generated respectively, meanwhile, constraint on rigid registration and non-rigid registration is achieved, and finally a registration result is obtained through resampling of a space conversion module. According to the method, the problems that cross-modal consistent features are difficult to extract in the infrared and visible light image registration process and explicit registration steps are complicated and difficult to generalize are solved, and the method has good universality and practicability.

The invention discloses an ancient textile image restoration system based on artificial intelligence. The system comprises a multi-source image acquisition module, a damaged area detection module, a pattern generation module, a color restoration module, a texture synthesis module and a multi-scale fusion module. The system introduces a wavelet guidance-frequency domain attention mechanism and a rotation invariant Haar wavelet basis function to realize accurate identification and classification of a damaged area; a saliency-guided wavelet decomposition control and self-adaptive threshold denoising method is combined, so that the perception capability of slant textures and edge details is improved; the texture synthesis module constructs a hierarchical modeling strategy fusing Gram style loss, Wasserstein style loss and a total variation regular term, and realizes generation of high-quality textures with unified styles and smooth edges; the system can be widely applied to cultural relic digital repair and display scenes.

The invention relates to the technical field of image processing, in particular to a remote sensing image segmentation method based on multilayer wavelet transform and a dynamic memory network, which comprises the following steps: decomposing an image into a low-frequency sub-band and a plurality of high-frequency sub-bands through multilayer Haar wavelet transform, and performing feature extraction on each frequency band in combination with convolution operation to obtain a final feature map; constructing a dynamic memory unit network, performing dynamic feature extraction on the feature map by adopting a query-key-value structure, generating a dynamic memory enhanced feature with global context information, performing global-local feature modeling on the feature map, and fusing the dynamic memory enhanced feature and the global-local feature to obtain a fused feature map; and segmenting the fused feature map to obtain a final segmentation result. According to the method, low-frequency smooth information and high-frequency detail information of the image can be accurately extracted, efficient fusion of global and local features can be realized, and the accuracy and robustness of remote sensing image segmentation are improved.

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 relates to a visible light and infrared image deep fusion auto-encoder network model based on Haar wavelet transform, and belongs to the technical field of multi-modal image fusion. The model comprises an image feature extraction module, an image feature fusion module and an image reconstruction module. The image feature extraction module comprises shallow shared feature extraction, coarse-grained feature extraction and fine-grained feature optimization; the image feature fusion module fuses high-frequency and low-frequency features by using Haar wavelet inverse operation; and the image reconstruction module carries out image reconstruction by using a Decoder module. The self-encoder network model has good performance, particularly, the high-frequency and low-frequency features of the infrared picture and the visible light picture reserved in the fused image are rich, fusion is fast, and help is provided for improving the accuracy of downstream visual tasks such as target detection, target tracking and target segmentation.

The invention relates to a CT image segmentation method based on a wavelet double-U-shaped space-frequency fusion transformer network, and belongs to the field of CT image processing. According to the method, a low-frequency component and a high-frequency component of a feature map are obtained through Haar wavelet transformation, linear self-attention interaction is carried out in low frequency and high frequency respectively, a network learns a global structure from the low frequency and captures detail features in the high frequency, a U-shaped encoder-decoder structure on the outer layer of the network supplements space information for a Transform network, and therefore the network structure is optimized. The local feature extraction capability of the Transform network is enhanced, so that the network can mine local details in the image more deeply, and finally, the spatial domain feature and the frequency domain feature are taken as two branches to realize feature fusion according to space and channel dimension alternate weighting. The method can improve the precision of the segmentation result and the detail retention capability.

The invention relates to the technical field of intelligent environment regulation and control, and particularly discloses an intelligent fresh air handling unit self-adaptive environment regulation and control method and system based on the Internet of Things. Temperature and humidity data are collected in real time through multi-source sensors deployed indoors and outdoors, and a temperature stability characteristic value is calculated through zero-sequence processing and fast Fourier transform; a humidity change rate characteristic value is extracted by combining first-order difference processing and Haar wavelet transform; the feature values are constructed into a comprehensive environment state feature vector, and the comprehensive environment state feature vector is input into a trained random forest model for environment regulation and control effect evaluation; according to the method, the sensing technology, signal analysis and machine learning prediction are fused, the sensing ability, the regulation and control precision and the self-adaptability of the system are improved, and the method is suitable for complex and changeable environment scenes and has good application prospects.

The invention relates to an underwater single target tracking method based on a wavelet token and a space-time Transform. The method comprises the following steps: firstly, constructing a reference frame sequence, a search frame and a previous frame historical token into a space-time input sequence, and extracting cross-frame features through a Transform encoder; then, Haar wavelet decomposition is carried out on the historical token, and a low-frequency component representing a target structure and a high-frequency component capturing motion details are separated out; then, adaptively fusing the global features and the historical components of the current search frame by using a gating mechanism, and generating a wavelet token; and finally, inputting the wavelet token and the global feature into a prediction head, and outputting a target classification confidence map and a bounding box regression map to determine the position and the scale of the target. According to the technical scheme of the invention, the interference of underwater low-illumination noise can be effectively suppressed through the wavelet token, and the space-time continuity of target motion modeling is maintained in combination with a gating strategy, so that the tracking robustness of an underwater complex scene is effectively improved.

The invention relates to a hyperspectral classification method and system based on wavelet spatial spectrum Mamba, and belongs to the technical field of hyperspectral image processing and deep learning. According to the method, a hyperspectral image cube is used as input, an image is divided into overlapped 3D small blocks, spatial and spectral characteristic decomposition is carried out on each small block at the same time, multi-resolution analysis is carried out on reconstructed characteristics by using classical Haar wavelets, long-range dependence is captured through a Mama network of a state space model, and finally the image is input into a classifier for classification. The method effectively balances the relationship between the network model performance and the calculation complexity, has the multi-scale feature extraction capability and the global modeling capability at the same time, can reduce the calculation resource consumption while remarkably improving the classification precision, and has excellent robustness and generalization capability.

The invention relates to the technical field of precision manufacturing and intelligent control, and particularly discloses an intelligent spark machine machining parameter optimization method based on effect feedback, which comprises the following steps of: calculating an abnormal characteristic value of an electrode feeding speed by using Haar wavelet transform, and analyzing time sequence data of conductivity of a working solution by using fast Fourier transform; according to the method, abnormal characteristic values are obtained, the characteristic values are constructed into a comprehensive characteristic vector, the comprehensive characteristic vector is input into a random forest model for comprehensive analysis, an electric spark machining effect score is generated, the system can dynamically adjust the electrode feeding speed and the working fluid conductivity based on the effect score, and it is ensured that machining parameters are always in the optimal configuration; according to the method, the machining efficiency is improved, the rejection rate is reduced, the surface quality and the size precision of the product are greatly improved, and the requirement of precision manufacturing is met.

The invention relates to the technical field of intelligent environment recognition, and particularly discloses a water and soil conservation measure intelligent query method and system based on multi-dimensional parameterization. Temperature and humidity data are collected in real time through a multi-source sensor deployed in water and soil, and a temperature stability characteristic value is calculated by adopting zero-sequence processing and fast Fourier transform; a humidity change rate characteristic value is extracted by combining first-order difference processing and Haar wavelet transform; the feature values are constructed into a comprehensive environment state feature vector, and the comprehensive environment state feature vector is input into a trained random forest model for environment recognition result evaluation; according to the method, the sensing technology, signal analysis and machine learning prediction are fused, the sensing ability, the regulation and control precision and the self-adaptability of the system are improved, the method is suitable for complex and changeable environment scenes, and the method has the advantages of being high in adaptability, high in adaptability and the like. Good application prospects are realized.

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 a medical image segmentation method and system based on feature extraction optimization. The medical image segmentation method comprises the following steps: performing target region segmentation on a segmented image through a constructed image segmentation model; an image segmentation model in the invention adopts a U-Net structure and comprises an encoder and a decoder; a Haar wavelet transform and KAN convolution module is introduced into the encoder to carry out feature extraction on the input feature map; meanwhile, a channel cross attention module and a space weighting module are connected in series on jump connection of the U-Net structure, the channel cross attention module captures complex dependence of cross-stage features through channel attention, and the space weighting module generates dynamic weights by using KAN and space attention, optimizes a space-channel relation of the features, and obtains the cross-stage features. The two are combined to relieve the problems of multi-scale information loss and semantic inconsistency, and the global semantic integration capability is remarkably improved.

The invention relates to the technical field of automatic identification, and particularly discloses a self-adaptive deformation correction and identification system for a flexible curved surface bar code, which can accurately capture a complex deformation condition by monitoring the three-dimensional shape change and strain distribution of a bar code area on a flexible surface in real time, and can calculate a curvature characteristic value by using a K-means clustering algorithm, so as to realize the self-adaptive deformation correction and identification of the flexible curved surface bar code. Strain data are analyzed through Haar wavelet transform to determine deformation characteristic values, so that the deformation degree of the bar code and the influence of the deformation degree on the recognition process are comprehensively evaluated, a random forest model is constructed based on the characteristic values for deformation degree prediction and deformation score output, high-precision recognition is ensured, and when uncorrectable serious deformation is detected, the recognition accuracy is improved. The system automatically triggers an alarm mechanism, including sound alarm, visual signal and network notification, to inform management personnel in time, so that the stability and long-term reliability of the system are guaranteed.

The invention relates to the field of computer vision and target detection, in particular to the field of rotating small target detection based on a convolutional neural network, and particularly relates to a method for improving a YOLOv8 model by introducing a Haar wavelet transform down-sampling module HWD, a wavelet transform feature enhancement module WTFEM and a bounding box regression loss function MPDIOU based on the minimum point distance. Therefore, the detection precision and robustness of the small target are improved. The method comprises the following steps: firstly, replacing a down-sampling module in YOLOv8 with a Haar wavelet transform down-sampling module HWD; the HWD uses Haar wavelet transform to reduce the spatial resolution of the feature map, and at the same time, more information is reserved as much as possible. And secondly, a Wavelet Transform Feature Enhancement Module (WTFEM) is innovatively introduced into a check part of the network, so that the limitation of a traditional feature fusion mode is broken through, and the semantic understanding and detail retention capability of the model on a multi-scale target is remarkably improved. And finally, replacing the original loss function with a bounding box regression loss function MPDIOU based on the minimum point distance. According to the loss function, the Euclidean distance of the nearest vertex between a prediction frame and a real frame is calculated, and an area overlapping rate and a central point distance optimization target are combined, so that the problem of gradient disappearance of a traditional IoU in a boundary frame non-overlapping scene is solved.

The invention relates to the technical field of data management of intelligent manufacturing, and particularly discloses a 12-shaft PCB drilling machine production data acquisition and quality management system, which is characterized in that high-speed sensors are mounted on shafts of a drilling machine to acquire speed and pressure data in a drilling process in real time; a speed change rate characteristic value and a pressure stability characteristic value are extracted by adopting a method of combining differential speed analysis with Haar wavelet transform, zero-sequence pressure processing and principal component analysis, the two types of characteristic values are fused into a drilling quality characteristic vector, the drilling quality characteristic vector is input into a quality analysis model based on gradient boosting tree training, and the drilling quality is obtained. Intelligent evaluation and prediction of the drilling quality are realized, when quality abnormity is detected, the system automatically generates early warning information, and operation parameters of the drilling machine are dynamically adjusted in combination with a PID control algorithm, so that a closed-loop quality control mechanism is formed.

The invention relates to the technical field of multimedia processing, and particularly discloses a dynamic interaction control method and system for digital multimedia equipment, which is characterized in that the interaction effect between multimedia equipment is optimized through real-time monitoring and intelligent adjustment, and the edge data transmission speed and updating frequency are acquired in real time by using intelligent monitoring modules arranged in the equipment; and analyzing the data by applying fast Fourier transform and Haar wavelet transform to evaluate the stability and synchronization precision of data transmission, constructing a comprehensive feature vector by a data transmission anomaly feature value and an update frequency change feature value calculated based on an analysis result, and inputting the comprehensive feature vector into a random forest model to predict an interaction effect score, therefore, whether the current interaction state is stable or not is judged, once the unstable state is detected, the PID controller is adopted to automatically adjust the edge data transmission speed and the updating frequency, and rapid recovery and optimization of the system are achieved.

The invention discloses a line fault detection method and device based on morphological wavelets, relates to the technical field of line detection, and solves the problem that in the prior art, when a single-phase grounding fault occurs in a small-current grounding system, the fault current is small, the characteristics are not obvious, and when line fault detection is carried out by using time domain, frequency domain and time frequency analysis methods, the fault current is not obvious. And the technical problems of low fault line selection accuracy and poor reliability exist. The method comprises the following steps: acquiring zero-sequence current signals in a period before and after a line fault occurs; pre-processing the zero sequence current signal; decomposing the preprocessed zero sequence current signal by using a morphological Haar wavelet; calculating modulus maxima of the plurality of lines according to the decomposition result; correcting the modulus maxima of the plurality of lines based on the line length; screening fault lines according to the corrected modulus maxima; the accuracy and reliability of single-phase earth fault line selection can be improved.

The invention relates to the field of deep learning infrared image segmentation, in particular to an infrared image real-time semantic segmentation method based on a wavelet pooling three-branch network, and the method comprises the following steps: S1, carrying out the modeling of interference radiation and motion rules on the basis of a public infrared target data set, and making an infrared airplane data set under interference; s2, designing a Haar wavelet pooling module, and keeping key frequency domain information while reducing the resolution of the feature map by means of the advantage of frequency decomposition of the Haar wavelet pooling module; and S3, constructing a three-branch real-time semantic segmentation network model for respectively extracting space, semantic and boundary information. And S4, introducing a Haar wavelet pooling module into the three-branch network model as a down-sampling layer of the model to reduce information loss during down-sampling. On an infrared aircraft data set under self-made interference, the method not only can meet the real-time processing requirement of the infrared image, but also can improve the segmentation precision.

The invention discloses a power data desensitization method based on multi-granularity dynamic sensitivity grading, terminal equipment and a storage medium. The power data desensitization method comprises the following steps: dividing power data into four sensitive grades of identity information, transaction records, power consumption behaviors and equipment data; full life cycle protection of data acquisition, storage and transmission is realized through a collaborative desensitization mechanism of field data national secret SM4 encryption / dynamic mask and recorded data Laplacian noise injection / behavior mode generalization; according to the system, Haar wavelet decomposition is adopted to separate user power consumption time sequence data into a low-frequency approximation coefficient and a high-frequency detail coefficient, Paillier homomorphic encryption aggregation is carried out on a low-frequency component, Gaussian noise is added to a high-frequency component, then equalization processing is carried out, and an aggregation power consumption sequence is reconstructed through wavelet inverse transformation.

The invention relates to the technical field of cold-chain logistics intelligent temperature control, and particularly discloses a fruit and vegetable multi-temperature-zone cold-chain distribution accurate cold storage and temperature control method, a transportation space is divided into a plurality of independent temperature control zones, each temperature control zone is specially used for storing specific types of fruits and vegetables, and temperature and power supply data in each zone are monitored in real time; a temperature abnormal fluctuation characteristic value and a power supply fluctuation characteristic value are respectively calculated by using fast Fourier transform and Haar wavelet transform, the stability of temperature and power supply is evaluated, and the influence degree of power supply on the temperature stability is further analyzed by using a gradient boosting tree model. And calculating a power supply regulation value based on the support vector regression model, and dynamically adjusting the power supply of the corresponding region to maintain the optimal temperature condition.

The invention discloses a defect detection method for a high-density packaged chip, which relates to the technical field of semiconductor detection, and comprises the following steps: preprocessing the high-density packaged chip; imaging the preprocessed high-density packaged chip by using a multispectral imaging device to obtain a multispectral imaging image group; the method comprises the following steps: firstly, carrying out multi-level decomposition by adopting a discrete wavelet transform technology based on Haar wavelets, and then carrying out thresholding processing to obtain a multi-spectral imaging image group with sparse representation; carrying out compressed sensing on the sparse-represented multispectral imaging image group based on a compressed sensing technology to obtain a multispectral imaging image group after compressed sensing; the method comprises the following steps of: marking an image group of multispectral imaging after compressed sensing, transforming a visual Transform model input by RGB (Red, Green, Blue) into a visual Transform model input by a plurality of wavebands, and training the visual Transform model input by a plurality of channels; according to the method, the accuracy of defect detection of the high-density packaged chip is improved through multispectral imaging, sparse representation and compressed sensing technologies.

The invention discloses a hidden space adversarial sample generation method and system based on multi-scale feature separation, and the method comprises the steps: employing a neural network quantization training method based on straight-through estimation, and training a hierarchical vector quantization variational auto-encoder; carrying out differentiable Haar wavelet transformation on the input image by adopting a wavelet packet transformation algorithm, decomposing the input image into a low-frequency component and a high-frequency component, and realizing multi-scale feature separation; inputting the high-frequency component into a hierarchical vector quantization variational auto-encoder, and extracting and quantizing global high-frequency features and local high-frequency detail features; in the potential space, a learnable disturbance variable is introduced, a potential vector after disturbance is constructed, and the potential vector is reconstructed into an adversarial sample through a decoder; and based on a preset disturbance target, carrying out iterative optimization on the disturbance vector until a confrontation sample which satisfies an attack success condition and is optimized in visual quality is generated. According to the method, a wavelet domain variational auto-encoder and a hidden space iterative attack algorithm are fused, and an adversarial sample with high fidelity and clear interpretation is generated.

The invention discloses an open vocabulary multi-target tracking method based on confidence coefficient adjustment and wavelet convolution, belongs to the technical field of computer vision and target tracking, and can enhance the adaptive modeling capability of a target motion mode by designing a confidence coefficient weighted Kalman updating mechanism and developing a confidence coefficient noise adaptive Kalman filtering algorithm. And meanwhile, the Haar wavelet transform is introduced, so that the receptive field of convolution operation is remarkably expanded, and the capturing capability of high-frequency motion details and the maintaining capability of low-frequency contour information are effectively improved. And in combination with the joint cost matrix of the motion features and the appearance features, a more adaptive target tracking model is formed, and the tracking accuracy and stability are effectively improved. The method can effectively deal with rapid movement and temporary shielding of the target, accurately tracks various types of targets with undefined types, and is suitable for open vocabulary multi-target tracking tasks in various complex environments.

The invention relates to a bilateral sparse representation-based image compressed sensing method, which comprises the following steps of: obtaining an original image, and segmenting the original image to obtain a plurality of image blocks; performing bidirectional Haar wavelet transform on the image blocks to obtain a sparse matrix, vectorizing the sparse matrix, and obtaining a one-dimensional sparse vector; and carrying out compression and image reconstruction on the one-dimensional sparse vector to obtain a final reconstructed image. According to the method, the precision of the compressed sensing algorithm in the image reconstruction process can be greatly improved, and meanwhile, the calculation processing efficiency is remarkably improved.

The invention relates to the technical field of deep learning and load identification, in particular to a deep learning load identification method and system based on bilateral filtering denoising and multi-wavelet feature fusion and a medium, and the method comprises the steps: firstly converting an acquired training data set into an image, and carrying out the preprocessing of bilateral filtering denoising; graying the de-noised image, extracting low-frequency and high-frequency components by using Haar wavelet transform, and extracting low-frequency approximation and high-frequency information in horizontal, vertical and diagonal directions by using Daubechies wavelet transform; pixel unification and normalization are carried out on the feature map, a training set and a test set are divided after category label integers are coded, and a convolutional neural network containing two branches is constructed to extract depth features and splice and fuse the depth features; and finally, extracting fusion features through a full connection layer, and taking sparse classification cross entropy as a loss function to train a CNN model in an off-line manner to obtain a load identification model. The method can improve the accuracy and stability of load identification, and is suitable for various electric equipment load identification scenes.